Tokushima University / Educator and Researcher Directory --- Kita, Kenji

Personal Web Page

https://researchmap.jp/K_Kita (researchmap)

Field of Study

○	Computer Science, Information Engineering, Information Retrieval, Natural Language Processing, Spoken language processing, Media technology

Subject of Study

○	Vector space models, Probabilistic language models, Multimedia information retrieval, Cross-media information retrieval, Music identification based on audio fingerprinting, EGG-based emotional processing, Medical image processing, Medical AI, Dental AI (information retrieval, natural language processing, computational linguistics, corpus, Latent semantic indexing, multidimensional search, Hamming space search, audio fingerprint, music retrieval, image retrieval)

Book / Paper

Book:

1.	Kenji Kita, Kazuyuki Matsumoto, Minoru Yoshida, Masami Shishibori and Masaki Oono : Hands-on Data Preprocessing in AI and Machine Learning, --- Data Science with Python ---, Kagakujyoho Shuppan Co., Ltd., Tokyo, Aug. 2021. (Summary) Research on artificial intelligence (AI) itself has been going on since the dawn of computing, with the "first AI boom" in the late 1950s and 1960s, the "second AI boom" in the 1980s and 1990s, and now we are in the midst of the "third AI boom". The major difference between this third boom and the previous first and second booms is the emergence of intelligent information processing systems that approach human capabilities in many fields. This can be attributed to a variety of factors, but the most significant are the accumulation of various kinds of big data, the accelerated progress of IT technologies that make such big data available, and the emergence of new machine learning paradigms such as deep learning. One of the keys to the success of AI and machine learning systems is the pre-processing of data. The essence of machine learning is to correctly make predictions and inferences about unknown data by learning the structure, regularities, and universality that lie behind large amounts of data. However, if the quality of the data used for learning is poor, the system will not be able to learn correctly, and the resulting system will not be accurate. In order to build a highly accurate system, it is important to eliminate poor data and process the data to convert it into a form that is easy to learn. This is exactly what preprocessing is responsible for. In extreme cases, it can be said that the success or failure of preprocessing guarantees the quality of the entire machine learning system. According to one theory, in the field of AI and machine learning system development, 60 to 80% of the time spent by engineers is spent on data collection and preprocessing. For efficient system development, it is essential to master preprocessing techniques. This book specifically focuses on preprocessing techniques, which have not been sufficiently covered in conventional books on machine learning and data science, and aims to be a practical book that allows readers to understand the techniques not only by explaining them, but also by actually running programs. The sister volume of this book, "Fundamentals," introduces basic preprocessing techniques. This book, "Applications," explains more advanced preprocessing techniques and preprocessing techniques for media data such as text, images, sound, and music. (Keyword) artificial intelligence / machine learning / Python / Preprocessing / data analysis / data mining / Google Colab / Big Data / Feature Selection / Dimensionality Reduction / NumPy / pandas / scikit-learn / matplotlib / TensorFlow / Keras / Data Cleaning / Feature Extraction / Scaling / Standardization / normalization / Normalization / Data Cleansing / Data Augmentation / Binning / Discretization / Categorical Data / Category Data / One-hot Encoding / One-hot Vector / Label Encoding / Frequency Encoding / Count Encoding / Ranked Frequency Encoding / Label-count Encoding / Feature Hashing / Global Hashing Space / Per-field Hashing Space / MurmurHash3 / Entity Embedding / Distributed Representation / Embedding / Table2Vec / Imbalanced Data / Resampling / Undersampling / Downsampling / Oversampling / Upsampling / Random Oversampling / Random Undersampling / ENN / SMOTE / ADASYN / Time-series Data / Point Process / Windowed Statistics / Aozora Bunko / Word Segmentation / morphological analysis / MeCab / Bigram / Trigram / N-gram / SentencePiece / Stop Word / Stop Word List / vector space model / Bag-of-Words / tf-idf / gensim / Topic Model / LDA / Gibbs Sampling / word2vec / Skip-gram / Continuous Bag-of-Words / Negative Sampling / Hierarchical Softmax / CNN / Forward Propagation / Activation Function / Back Propagation / Sum of Squares Error / Cross Entropy / Epoch / Gradient Descent / Batch Learning / Online Learning / Mini-batch Learning / iteration / Convolution Layer / Pooling Layer / Fully Connected Layer / Feature Map / Zero Padding / Maximum Pooling / Average Pooling / Dropout Layer / SDG / Shear Mapping / Fine-tuning / VGG16 / Sampling Rate / Quantization Bit Rate / Cocktail-party Effect / Spectral Subtraction / Harmonic/Percussive Sound Separation / Short-time Fourier Transform / Inverse Short-time Fourier Transform / Wiener Filter / Audio Source Separation
2.	Minoru Yoshida and Kenji Kita : Mining Numbers in Text: A Survey, IntechOpen, Jul. 2021. (Link to Publication Site) ● Publication site (DOI): 10.5772/intechopen.98540 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.5772/intechopen.98540 (DOI: 10.5772/intechopen.98540)
3.	Kenji Kita, Ryota Nishimura and Kazuyuki Matsumoto : Introduction to Data Preprocessing in AI and Machine Learning, --- Getting Started with Python ---, Kagakujyoho Shuppan Co., Ltd., Tokyo, Jun. 2021. (Summary) Around 2010, the term "Big Data" began to rapidly permeate the world around us. At the same time, the construction of intelligent AI systems based on machine learning from big data has even become a boom these days. This book focuses on the pre-processing of data, which is an inevitable part of building such machine learning systems, and although the primary focus is on machine learning systems such as AI, the various techniques described in this book can also be used effectively in data mining and data analysis to acquire useful knowledge from big data. However, the various techniques described in this book can also be used effectively in data mining and data analysis to obtain useful knowledge from big data.The term "big data" is pleasant to the ears and may sound like a treasure trove, but in fact big data is a mixture of cobbles and stones, a mixture of treasure and garbage. In fact, big data is a mixture of treasure and trash, and the larger the scale of the data, the more trash there is, and the more difficult it becomes to extract the treasure from it. In the field of computer science, there is a saying, "Garbage In, Garbage Out" (or GIGO for short). To build a good machine learning system, it is important to eliminate garbage from data and process the data into a form that is easy to learn, which is the essence of preprocessing.Recently, many frameworks and libraries for machine learning such as TensorFlow and Keras are available free of charge, and using these frameworks and libraries may seem to lower the threshold for programming. Using publicly available datasets for machine learning, anyone can easily create an AI system. However, there is a big problem when building your own system using your own data set. As is often said, real-world data is dirty. If you use the data as it is, that is exactly what you will get.If we use the data as it is, we will end up in a "Garbage In, Garbage Out" situation. Whether or not a useful AI system can be built from real data depends on the success of pre-processing. According to one theory, in the actual construction of AI and machine learning systems, 60-80% of the time spent by engineers is spent on data collection and preprocessing.This book focuses on preprocessing techniques, which have not been sufficiently covered in conventional books on machine learning, and aims to be a practical book where readers can understand the techniques not only by explaining them but also by actually running programs. This book has a sister volume, "Practical Edition", which is planned to be published in the near future. In "Practical Edition", more advanced preprocessing techniques and preprocessing techniques for media data such as text, images, sound, and music are explained. (Keyword) artificial intelligence / machine learning / Python / Preprocessing / data analysis / data mining / Google Colab / Big Data / Feature Selection / Dimensionality Reduction / NumPy / pandas / scikit-learn / matplotlib / TensorFlow / Keras / Data Cleaning / Feature Extraction / Scaling / Standardization / normalization / Normalization / Data Cleansing / Data Augmentation / Binning / Discretization / Outlier / IQR / Boxplot / Smirnov-Grubbs' test / Missing Value / List-wise Case Deletion / Pair-wise Case deletion / Single Imputation / Multiple Imputation / Filter Method / Chi-squared Test / analysis of variance / Pearson Correlation Coefficient / PCA / NMF / CCA / LDA / Kernel PCA / t-SNE / UMAP / MDS / self-organizing map / autoencoder / SMOTE
4.	Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Analysis of Information Spreading by Social Media based on Emotion and Empathy, InTechOpen, Sep. 2019. (Summary) The number of social media users has increased exponentially in recent times,and various types of social media platforms are being introduced. While socialmedia has become a convenient communication tool, its use has caused varioussocial problems. Some users who cannot imagine the emotions their posts mayinduce in readers cause what is termed as the flaming phenomenon. In somecases, users intentionally repeat strong remarks for self-advertisement. To identifythe cause of this phenomenon, it is necessary to analyze the posted contents or thepersonalities of the users who cause the flaming. However, it is difficult to reach ageneralized conclusion because each case varies depending on the circumstancesand individual. In this chapter, we study the phenomenon of information spreadingvia communication on social media by conducting a detailed analysis of replies andnumber of retweets in Japanese, and we reveal the relation between the feedback onsuch posts and the emotions or empathy they result in. (Keyword) social media / information diffusion / flaming / buzz / Twitter (Link to Publication Site) ● Publication site (DOI): 10.5772/intechopen.88886 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.5772/intechopen.88886 (DOI: 10.5772/intechopen.88886)
5.	Kazuyuki Matsumoto, Hayato Shimizu, Minoru Yoshida and Kenji Kita : Time-Series Analysis of Video Comments on Social Media, Worlds largest Science, Technology & Medicine Open Access book publisher, Jul. 2017. (Summary) In this study, we propose a method to detect unfair rating cheat caused by multiple comment postings focusing on time-series analysis of the number of comments. We defined the videos that obtained a lot of comments by unfair cheat as unfair video and defined the videos which obtained without unfair cheat as popular video. Specifically, our proposed method focused on the difference of chronological distributions of the comments between the popular videos and the unfair videos. As the evaluation result, our proposed method could obtain higher accuracy than that of the baseline method. (Keyword) video comments / shared videos / comment analysis, / time-series analysis (Link to Publication Site) ● Publication site (DOI): 10.5772/intechopen.68636 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.5772/intechopen.68636 (DOI: 10.5772/intechopen.68636)
6.	Kazuyuki Matsumoto, Fuji Ren, Minoru Yoshida and Kenji Kita : Refinement by Filtering Translation Candidates and Similarity Based Approach to Expand Emotion Tagged Corpus, Jan. 2017. (Summary) We attempted to expand corpus without translating target linguistic resource. The result of the evaluation experiment using the machine learning algorithm showed the effectiveness of the expanded emotion corpus based on the original languages unannotated sentences and their similar sentences. (Keyword) emotion tagged corpus / Japanese-English parallel corpora / emotion estimation (Link to Publication Site) ● Publication site (DOI): 10.1007/978-3-319-52758-1_15 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85011342735 (DOI: 10.1007/978-3-319-52758-1_15, Elsevier: Scopus)
7.	邰暁英 and Kenji Kita : 信息检索技术导论(情報検索技術入門), 科学出版社, Beijing, Sep. 2006. (Summary) 本书系统地介绍了信息检索的原理与技术．具体内容包括:基于向量空间模型的检索，全文信息检索的各种方法及实现算法，图像检索技术基于Web信息检索等．本书注重概念与实用性的统一，书中配有实例，每章后配有习题，以方便读者理解，复习，巩固所学知识． (本書は，情報検索の原理と技術を紹介している．ベクトル空間モデルや全文検索アルゴリズム，画像検索技術，Web情報検索などについて説明している．)
8.	Koji Soda, Yoshio Kashida, Shigeru Sugiyama, Masahiro Nomura, Kenji Kita, Mayumi YAMAMOTO and Tomohiko Maeda : レポート作成指導ハンドブック, Center for University Extension, Tokushima, Jan. 2005. (Summary) 入学初年次生のレポート作成指導をするための具体的な事例を示した指導書(p.1-p.30)．本書の必要部分をコピーして学生に配布できるように，学生を読者にした記述スタイルを取っている．レポート作成手順を示し，学生が従来戸惑うことの多かった部分に具体的事例を示すとともに，最近問題となっている著作権などに関連する注釈や引用の具体例を明記した．またレポート作成の手助けとなるよう徳島大学付属図書館ホームページ利用法も示した．
9.	Kenji Kita, Yasuhiko Kato, Takashi Omoto and Yoneo Yano : Corpus Linguistics: Readings in a Widening Discipline, --- Automatically Extracting Collocations from Corpora for Language Learning ---, Continuum International Publishing, London, Aug. 2004. (Summary) One kind of information about usage which is valuable for many purposes is information about collocations - pairs or sequences of words that tend to co-occur frequently. This book chapter, we have taken up the problem of how to acquire collocational knowledge and discussed its importance for language learning.
10.	Kenji Kita : 応用言語学事典, Kenkyusha, Tokyo, Apr. 2003. (Summary) 応用言語学事典の1項目である「隠れマルコフモデル(Hidden Markov Model)」について説明した．
11.	Kenji Kita, 津田和彦 and Masami Shishibori : 情報検索アルゴリズム, KYORITSU SHUPPAN.CO.,LTD, 東京, Jan. 2002. (Summary) 前半部分では，情報検索の概要，情報検索の適用分野や関連分野，情報検索システムの性能評価など，情報検索全般に関わる基礎事項について説明した．また，後半部分では，情報検索の具体的な例として，ベクトル空間モデルに基づく類似文書検索手法，文字列照合に基づく全文検索手法，索引を用いた全文検索手法などについて説明した．
12.	Kenji Kita : 確率的言語モデル, University of Tokyo Press, Tokyo, Nov. 1999. (Summary) 言語データ(コーパス)に基づいた自然言語処理の中核的技術である確率的言語モデルのうち，代表的なモデルであるNグラムモデル，隠れマルコフモデル，確率文法，最大エントロピーモデルについて詳細に説明した．また，確率的言語モデルの応用例として，形態素解析および品詞付け，単語のクラスタリング，構文的愛昧性の解消，言語の識別と分類などについて説明した．
13.	Yasuhito Tanaka and Kenji Kita : Terminology and Knowledge Engineering, --- JCKE Multilingual Corpus of Major Asian Languages ---, TermNet, Vienna, Aug. 1999. (Summary) In this book chapter, we discuss the development of a multilingual corpus of major Asian languages, including Japanese, Chinese, Korean, as well as English. The multilingual corpus here not only contains the same texts in more than one language, but also has sentence-level language-to-language correspondence, as well as allowing easy computer processing. (Keyword) corpus
14.	Kenji Kita, Satoshi Nakamura and Masaaki Nagata : 音声言語処理, 森北出版, Tokyo, Nov. 1996. (Summary) コーパスに基づいた音声言語処理の中核をなす音声言語の確率モデルとして，N-gramモデル，隠れマルコフモデル，確率文脈自由文法について述べるとともに，これらのモデルの音声認識や自然言語処理への応用について述べた．
15.	榑松明, 相沢輝昭, 阿部匡伸, 飯田仁, 浦谷則好, 菊井玄一郎, Kenji Kita, 桑原尚夫, 嵯峨山茂樹, 匂坂芳典, 鹿野清宏, 杉山雅英, 杉山雅英, 鈴木雅実, 堂坂浩二, 永田昌明 and 森元逞 : 自動翻訳電話, Ohmsha, Ltd., Tokyo, Jan. 1994. (Summary) 自動翻訳電話及びそれに関連する音声言語処理技術(音声認識技術，言語翻訳技術，音声合成技術，データベース技術)についての書であり，担当セクションでは，隠れマルコフモデル(HMM)と文脈自由文法を用いた連続音声認識の手法と評価実験結果について述べた．
16.	Kenji Kita, Takeshi Kawabata and Hiroaki Saito : Generalized LR Parsing, --- GLR Parsing in Hidden Markov Model ---, Kluwer Academic Publishers, Boston, USA, Oct. 1991. (Summary) We describe the application of GLR (Generalized LR) parsing to speech recognition. In particular, we will focus on a method called HMM-LR, which is an integration of Hidden Markov Models and GLR parsing. In HMM-LR, GLR parsing is used as a language source model for word/phoneme prediction/generation. We also introduce a speech recognizer based on the HMM-LR method.
17.	Toshiyuki Hanazawa, Kenji Kita, Satoshi Nakamura, Takeshi Kawabata and Kiyohiro Shikano : Readings in Speech Recognition, --- ATR HMM-LR Continuous Speech Recognition System ---, Morgan Kaufmann Publishers, California, USA, Oct. 1990. (Summary) Speech recognition based on acoustic information alone does not attain good performance. Linguistic information can be used to improve recognition accuracy. This paper describes a continuous speech recognition system using HMMs and predictive LR parsing, and an improvement of the HMM-LR continuous speech recognizer using multiple codebooks, HMM state duration control and fuzzy vector quantization.

Academic Paper (Judged Full Paper):

1.	Zipaer Mayinuer, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Using Machine Learning to Classify Information Related to Child Rearing of Infants from Twitter, Advances in Machine Learning & Artificial Intelligence, Vol.4, No.2, 32-40, 2023.
2.	Motonobu Yoshida, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : System to Correct Toxic Expression with BERT and to Determine the Effect of the Attention Value, Communications in Computer and Information Science, Vol.1842, 239-253, 2023. (Summary) This paper describes an extended experiment on a system that converts sentences with toxic expressions into safe sentences, along with the evaluation and influence of the system. In recent years, toxicity on social media has created many problems. We evaluated the effectiveness of the proposed system for identifying toxic sentences using a prediction model based on Bidirectional Encoder Representations from Transformers (BERT) and then converting them into safe sentences using attention values and a score that indicates whether the sentences are appropriate after the predictive conversion. Six patterns of methods were tested, with Pattern 6 being the most effective for mitigating toxicity. This pattern is a technique that changes the way to take the top sentences of a beam search for each number of treatments, in addition to converting words with an attention value above a threshold and their adjacent words and phrases and words registered in the toxic dictionary. We used multiple indicators to judge the effectiveness of this method and evaluated its ability to make the text safe while preserving its meaning. (Keyword) Toxic expression / BERT / classification / Text correction / Attention value (Link to Publication Site) ● Publication site (DOI): 10.1007/978-3-031-43471-6_11 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85174230206 (DOI: 10.1007/978-3-031-43471-6_11, Elsevier: Scopus)
3.	Hidetoshi Nakao, Msakazu Imaoka, Mitsumasa Hida, Ryota Imai, Misa Nakamura, Kazuyuki Matsumoto and Kenji Kita : Determination of individual factors associated with hallux valgus using SVM-RFE, BMC Musculoskeletal Disorders, Vol.24, No.1, 534, 2023. (Summary) Introduction This cross-sectional study aimed to determine the factors related to hallux valgus (HV) and their importance using support vector machine-recursive feature elimination (SVM-RFE). Methods A total of 864 participants aged 18 years were enrolled. The Manchester scale was used to determine the presence of HV (summed scores for both feet 4). The questionnaire included items such as age, sex, height, weight, and foot measurements. These internal factors were analyzed to determine if they are related to HV using SVM-RFE. Results The results of tenfold cross-validation using SVM-RFE revealed that the numbers of feature selections were 10, 10, and 9 for age, sex, and body weight, respectively, and these factors were shown to be related to HV. HV was found to be more common in women than in men (women, 24.9%; men, 7.6%), but the sex difference was not significant in older people. Conclusion Age and sex were found to be important factors associated with HV identified via feature selection using SVM-RFE. (Keyword) feature selection / Hallux Valgus / Manchester Scale / SVM-RFE (Tokushima University Institutional Repository) ● Metadata: 118876 (Link to Publication Site) ● Publication site (DOI): 10.1186/s12891-023-06303-2 (Link to Search Site for Scientific Articles) ● PubMed @ National Institutes of Health, US National Library of Medicine (PMID): 37386376 ● Summary page in Scopus @ Elsevier: 2-s2.0-85163790710 (Tokushima University Institutional Repository: 118876, DOI: 10.1186/s12891-023-06303-2, PubMed: 37386376, Elsevier: Scopus)
4.	Kazuyuki Matsumoto, Reishi Amitani, Minoru Yoshida and Kenji Kita : Trend Prediction Based on Multi-Modal Affective Analysis from Social Networking Posts, Electronics, Vol.11, No.21, 2022. (Summary) This paper propose a method to predict the stage of buzz-trend generation by analyzing the emotional information posted on social networking services for multimodal information, such as posted text and attached images, based on the content of the posts. The proposed method can analyze the diffusion scale from various angles, using only the information at the time of posting, when predicting in advance and the information of time error, when used for posterior analysis. Specifically, tweets and reply tweets were converted into vectors using the BERT general-purpose language model that was trained in advance, and the attached images were converted into feature vectors using a trained neural network model for image recognition. In addition, to analyze the emotional information of the posted content, we used a proprietary emotional analysis model to estimate emotions from tweets, reply tweets, and image features, which were then added to the input as emotional features. The results of the evaluation experiments showed that the proposed method, which added linguistic features (BERT vectors) and image features to tweets, achieved higher performance than the method using only a single feature. Although we could not observe the effectiveness of the emotional features, the more emotions a tweet and its reply match had, the more empathy action occurred and the larger the like and RT values tended to be, which could ultimately increase the likelihood of a tweet going viral. (Keyword) multi-modal buzz prediction / information diffusion / affective analysis (Tokushima University Institutional Repository) ● Metadata: 118006 (Link to Publication Site) ● Publication site (DOI): 10.3390/electronics11213431 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.3390/electronics11213431 (Tokushima University Institutional Repository: 118006, DOI: 10.3390/electronics11213431)
5.	Tsubasa Miyashita, Kazuyuki Matsumoto, Minoru Yoshida, Ryota Nishimura and Kenji Kita : Extraction of Lifestyle Habits Described in Blogs of Diabetics, IEEJ Transactions on Electronics, Information and Systems, Vol.142, No.10, 1144-1155, 2022. (Summary) Tackling diabetes, an increasingly common lifestyle-related illness, requires information on patients lifestyle and habits. Blogs maintained by patients afflicted with the incurable illness may be useful for analyzing how lifestyle affects health-status. In this paper, we propose an intrinsic expression extractor that extracts keywords related to lifestyle and health-status from blogs of patients diagnosed with type-2 diabetes. To counter class imbalance and add to the corpus, the proposed method tags each keyword with information based on cue-words extracted from manually tagged data. The named-entity recognition (NER) for the extracted keywords uses a bidirectional gated recurrent unit neural network (BiGRU) and was evaluated for accuracy by cross-validation. We obtained F1-score of approximately 0.76. Although the accuracy of extraction can further be improved, the novel approach has applications in analyzing and improving the lifestyle of diabetes-afflicted patients. (Keyword) disease-fighting weblogs / named entity recognition / natural language processing (Link to Publication Site) ● Publication site (DOI): 10.1541/ieejeiss.142.1144 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390293633437552768 ● Search Scopus @ Elsevier (DOI): 10.1541/ieejeiss.142.1144 (DOI: 10.1541/ieejeiss.142.1144, CiNii: 1390293633437552768)
6.	Kohei Torii, Eiichi Honda and Kenji Kita : Development of an AI-based Dental Support System for Panoramic X-ray Images, Dental Radiology, Vol.62, No.1, 24-34, 2022. (Summary) Research and development of AI-based diagnostic systems in the medical and dental fields is flourishing worldwide. However, there are few practical dental support systems and dental databases. Support for medical record input is needed to reduce the burden at diagnosis. We have been developing a dental database with detailed annotation information and building an AI system for automatic teeth detection, teeth numbering, teeth contour estimation, and disease diagnosis from panoramic X-ray images since 2019. Three dentists and one expert of dental radiology created our database, which includes teeth number based on FDI method, coordinates of teeth contour, and various dental conditions, using Anotee, a software developed for creating dental databases. Our system consists of multiple deep neural networks that were trained using 1,781 panoramic X-ray images and annotations, which excluded deciduous teeth and rare dental conditions. The deep neural network to classify dental conditions is based on EfficientNetV2-S and can diagnose multiple dental conditions. To verify our systems usefulness, we evaluated two dental condition classifications for 20 dental conditions and 10 dental conditions such as caries, periodontitis, root canal filling, inlay, composite resin, crown, pontic, implant, and impacted tooth. We performed 5-fold cross validation and calculated precision, sensitivity, and specificity. Experimental results were encouraging. For the diagnosis of 20 conditions, precision was 90.4%, sensitivity was 86.1%, and specificity was 99.4%; for the diagnosis of 10 conditions, precision was 92.9%, sensitivity was 90.0%, and specificity was 99.1%. The system achieved high accuracy, suggesting that AI systems are useful in assisting medical record input support. (Keyword) Panoramic X-ray Images / medical imaging / image processing / artificial intelligence / Dental Support System (Tokushima University Institutional Repository) ● Metadata: 118606 (Link to Publication Site) ● Publication site (DOI): 10.11242/dentalradiology.62.24 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.11242/dentalradiology.62.24 (Tokushima University Institutional Repository: 118606, DOI: 10.11242/dentalradiology.62.24)
7.	Kazuyuki Matsumoto, Takumi Matsunaga, Minoru Yoshida and Kenji Kita : Emotional Similarity Word Embedding Model for Sentiment Analysis, Computacion y Sistemas, Vol.26, No.2, 875-886, 2022. (Summary) We propose a method for constructing a dictionary of emotional expressions, which is an indispensable language resource for sentiment analysis in the Japanese. Furthermore, we propose a method for constructing a language model that reproduces emotional similarity between words, which to date has yet not been considered in conventional dictionaries and language models. In the proposed method, we pretrained sentiment labels for the distributed representations of words. An intermediate feature vector was obtained from the pre-trained model. By learning an additional semantic label on this feature vector, we can construct an emotional semantic language model that embeds both emotion and semantics. To confirm the effectiveness of the proposed method, we conducted a simple experiment to retrieve similar emotional words using the constructed model. The results of this experiment showed that the proposed method can retrieve similar emotional words with higher accuracy than the conventional word-embedding model. (Keyword) Emotion recognition / emotional similarity / neural networks (Tokushima University Institutional Repository) ● Metadata: 117201 (Link to Publication Site) ● Publication site (DOI): 10.13053/CyS-26-2-4266 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.13053/CyS-26-2-4266 (Tokushima University Institutional Repository: 117201, DOI: 10.13053/CyS-26-2-4266)
8.	Kazuyuki Matsumoto, Manabu Sasayama, Minoru Yoshida, Kenji Kita and Fuji Ren : Emotion Analysis and Dialogue Breakdown Detection in Dialogue of Chat Systems Based on Deep Neural Networks, Electronics, Vol.11, No.5, 2022. (Summary) In dialogues between robots or computers and humans, dialogue breakdown analysis is an important tool for achieving better chat dialogues. Conventional dialogue breakdown detection methods focus on semantic variance. Although these methods can detect dialogue breakdowns based on semantic gaps, they cannot always detect emotional breakdowns in dialogues. In chat dialogue systems, emotions are sometimes included in the utterances of the system when responding to the speaker. In this study, we detect emotions from utterances, analyze emotional changes, and use them as the dialogue breakdown feature. The proposed method estimates emotions by utterance unit and generates features by calculating the similarity of the emotions of the utterance and the emotions that have appeared in prior utterances. We employ deep neural networks using sentence distributed representation vectors as the feature. In an evaluation of experimental results, the proposed method achieved a higher dialogue breakdown detection rate when compared to the method using a sentence distributed representation vectors. (Keyword) natural language processing / dialogue breakdown / human-computer dialogue system / sentiment analysis / emotion recognition (Tokushima University Institutional Repository) ● Metadata: 116757 (Link to Publication Site) ● Publication site (DOI): 10.3390/electronics11050695 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.3390/electronics11050695 (Tokushima University Institutional Repository: 116757, DOI: 10.3390/electronics11050695)
9.	Akira Fujisawa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Emotion Estimation Method Based on Emoticon Image Features and Distributed Representations of Sentences, Applied Sciences, Vol.12, No.3, 2022. (Summary) This paper proposes an emotion recognition method for tweets containing emoticons using their emoticon image and language features. Some of the existing methods register emoticons and their facial expression categories in a dictionary and use them, while other methods recognize emoticon facial expressions based on the various elements of the emoticons. However, highly accurate emotion recognition cannot be performed unless the recognition is based on a combination of the features of sentences and emoticons. Therefore, we propose a model that recognizes emotions by extracting the shape features of emoticons from their image data and applying the feature vector input that combines the image features with features extracted from the text of the tweets. Based on evaluation experiments, the proposed method is confirmed to achieve high accuracy and shown to be more effective than methods that use text features only. (Keyword) emoticon / emotion estimation / multimodal information (Tokushima University Institutional Repository) ● Metadata: 116758 (Link to Publication Site) ● Publication site (DOI): 10.3390/app12031256 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.3390/app12031256 (Tokushima University Institutional Repository: 116758, DOI: 10.3390/app12031256)
10.	Kazuyuki Matsumoto, Ryota Kishima, Seiji Tsuchiya, Tomoki Hirobayashi, Minoru Yoshida and Kenji Kita : Relationship Between Personality Patterns and Harmfulness: Analysis and Prediction Based on Sentence Embedding, International Journal of Information Technology and Web Engineering, Vol.17, No.1, 1-24, 2022. (Summary) This paper hypothesize that harmful utterances need to be judged in context of whole sentences, and extract features of harmful expressions using a general-purpose language model. Based on the extracted features, we propose a method to predict the presence or absence of harmful categories. In addition, the authors believe that it is possible to analyze users who incite others by combining this method with research on analyzing the personality of the speaker from statements on social networking sites. The results confirmed that the proposed method can judge the possibility of harmful comments with higher accuracy than simple dictionary-based models or models using a distributed representations of words. The relationship between personality patterns and harmful expressions was also confirmed by an analysis based on a harmful judgment model. (Keyword) Deep Neural Networks / Harmful Expression / Internet Flaming Detection / MBTI / Personality (Tokushima University Institutional Repository) ● Metadata: 116756 (Link to Publication Site) ● Publication site (DOI): 10.4018/IJITWE.298654 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1360298336571568640 ● Search Scopus @ Elsevier (DOI): 10.4018/IJITWE.298654 (Tokushima University Institutional Repository: 116756, DOI: 10.4018/IJITWE.298654, CiNii: 1360298336571568640)
11.	Kazuyuki Matsumoto, Seiji Tsuchiya, Minoru Yoshida and Kenji Kita : Construction and Expansion of Dictionary of Idiomatic Emotional Expressions and Idiomatic Emotional Expression Corpus, International Journal of Computer & Software Engineering, Vol.6, No.2, 2021. (Summary) Objective: In the study of sentiment estimation from language, methods focusing on words, phrases, sentence patterns, and sentence-final expressions have been proposed. However, it is difficult to deal with a wide variety of emotional expressions by only assigning emotions to words and phrases. In particular, it is difficult to analyze metaphorical expressions and idiomatic expressions on a word-by-word basis, and it is impossible to register all expressions in a dictionary because new expressions can be created byflexibly replacing words. However, it is difficult to determine the constraints on the words to be replaced, and not all expressions can be registered in the dictionary as sentence patterns.Methods: In this paper, we construct a dictionary of idiomatic sentiment expressions, which contains idioms expressing emotions. In this paper, we construct a pseudo-emotional corpus by collecting utterances containing emotional idioms from social media and automatically assigning emotions expressed by the idioms.Results: This corpus includes expressions other than idioms, and can be an effective resource for estimating emotions in sentences that do not contain idioms. In this study, we create an emotion estimation model for utterances based on the constructed corpus, and conduct evaluation experiments to explore the problems of the idiomatic emotion corpus. In addition, using the constructed sentimentcorpus, we investigate how to expand the dictionary of sentiment expressions in idiomatic phrases by using deep learning methods.Conclusion: Using the corpus of idiomatic sentiments constructed by the proposed method as training data, models with and without idioms were constructed by machine learning models. The results show that the F-values of all emotions with idioms exceed 0.8. On the other hand, when idioms were not included, the F-values tended to decrease overall. However, the F-values of emotions such as "shame"and "excitement" were around 0.7, indicating that the characteristics of emotional expressions other than idioms were expressed. (Keyword) emotion recognition / idiomatic expression / idiomatic emotional expression / corpus (Tokushima University Institutional Repository) ● Metadata: 116606 (Link to Publication Site) ● Publication site (DOI): 10.15344/2456-4451/2021/174 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.15344/2456-4451/2021/174 (Tokushima University Institutional Repository: 116606, DOI: 10.15344/2456-4451/2021/174)
12.	Reishi Amitani, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Buzz Tweet Classification Based on Text and Image Features of Tweets Using Multi-Task Learning, Applied Sciences, Vol.11, No.22, 2021. (Summary) This study investigates social media trends and proposes a buzz tweet classification method to explore the factors causing the buzz phenomenon on Twitter. It is difficult to identify the causes of the buzz phenomenon based solely on texts posted on Twitter. It is expected that by limiting the tweets to those with attached images and using the characteristics of the images and the relationships between the text and images, a more detailed analysis than that of with text-only tweets can be conducted. Therefore, an analysis method was devised based on a multi-task neural network that uses both the features extracted from the image and text as input and the buzz class (buzz/non-buzz) and the number of likes (favorites) and retweets (RTs) as output. The predictions made using a single feature of the text and image were compared with the predictions using a combination of multiple features. The differences between buzz and non-buzz features were analyzed based on the cosine similarity between the text and the image. The buzz class was correctly identified with a correctness rate of approximately 80% for all combinations of image and text features, with the combination of BERT and VGG16 providing the highest correctness rate (Keyword) multi-task learning / buzz classification / social media / trend analysis (Tokushima University Institutional Repository) ● Metadata: 116681 (Link to Publication Site) ● Publication site (DOI): 10.3390/app112210567 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.3390/app112210567 (Tokushima University Institutional Repository: 116681, DOI: 10.3390/app112210567)
13.	Minoru Yoshida, Kojima Takumi, Kazuyuki Matsumoto and Kenji Kita : Toward Analyzing Relations between Sleep Time and Social Networking Service Texts: Prediction of the Tweet Time Span Using the Last Tweet of the Day, International Journal of Advanced Intelligence (IJAI), Vol.12, No.1, 1-9, 2021. (Summary) Sleeping habits are one of the major issues in today's healthcare. In this paper, we consider the problem of analyzing sleeping habits of people using social networking service (SNS) texts. As the first step toward predicting user's sleeping time using SNS texts, we assume that the time span between the user's last post in one day and the first post the next day can be used as a pseudo-indicator for the user's sleeping time if the user posts the text sufficiently frequently. We call such tweet time spans ``pseudo-sleeping time'' if the first tweet of the next day include ``Good morning'' or similar words. We try to predict such pseudo-sleeping time using the text (tweet) of the preceding tweet (i.e., the last tweet of the day). Preliminary experiments show that the tweet text contains some useful information to predict the user's pseudo-sleeping time. (Keyword) sleeping time / SNS / text mining (Tokushima University Institutional Repository) ● Metadata: 116122 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1570291227698842880 (Tokushima University Institutional Repository: 116122, CiNii: 1570291227698842880)
14.	Yuko Sakuta, Junichi Tsunobe, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Estimating Psychological Impressions from Color Images Using Earth Mover's Distance, Transactions of Japan Society of Kansei Engineering, Vol.19, No.1, 97-104, 2020. (Summary) In this paper, we describe a method for estimating psychological impressions (e.g., Pretty, Elegant, Romantic, Cool, Formal, etc) from color images. Essentially, the Earth Movers Distance (EMD) is employed to map a given image onto the color image scale, a two-dimensional emotion space comprised of a WARM-COOL axis and a SOFT-HARD axis, on which many color combinations and Kansei words are located. To validate our proposed method, we collected real images from the Web and compared our EMD-based results with human evaluation results. The comparison demonstrated that EMD-based Kansei mapping is a promising method for extracting psychological impressions from color images. (Keyword) Earth Mover's Distance / EMD / Psychological impressions / image processing / Color image scale / emotion / Kansei mapping / Color images (Link to Publication Site) ● Publication site (DOI): 10.5057/jjske.TJSKE-D-19-00038 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390002184878233472 ● Search Scopus @ Elsevier (DOI): 10.5057/jjske.TJSKE-D-19-00038 (DOI: 10.5057/jjske.TJSKE-D-19-00038, CiNii: 1390002184878233472)
15.	Kazuyuki Matsumoto, Seiji Tsuchiya, Takumi Kojima, Hiroya Kondo, Minoru Yoshida and Kenji Kita : Classification of Smartphone Application Reviews Using Small Corpus Based on Bidirectional LSTM Transformer, International Journal of Machine Learning and Computing, Vol.10, No.1, 148-157, 2020. (Summary) This paper provides the classification of the review texts on a smartphone application posted on social media. We propose a high performance binary classification method (positive/negative) of review texts, which uses the bidirectional long short-term memory (biLSTM) self-attentional Transformer and is based on the distributed representations created by unsupervised learning of a manually labelled small review corpus, dictionary, and an unlabeled large review corpus. The proposed method obtained higher accuracy as compared to the existing methods, such as StarSpace or the Bidirectional Encoder Representations from Transformer (BERT). (Keyword) attention mechanism / review classification / small corpus / transformer (Tokushima University Institutional Repository) ● Metadata: 114311 (Link to Publication Site) ● Publication site (DOI): 10.18178/ijmlc.2020.10.1.912 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.18178/ijmlc.2020.10.1.912 (Tokushima University Institutional Repository: 114311, DOI: 10.18178/ijmlc.2020.10.1.912)
16.	Hidetoshi Nakao, 今岡真和, 岡健司, 森藤武, 橋本雅至, Kazuyuki Matsumoto and Kenji Kita : Factor Analysis of Individual Factors Associated with Low Arch Height Ratio of the Foot by SVM-RFE, Transactions of the Japanese Society for Medical and Biological Engineering, Vol.57, No.6, 190-197, 2019. (Keyword) SVM-RFE (Link to Publication Site) ● Publication site (DOI): 10.11239/jsmbe.57.190 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390002184878527360 ● Summary page in Scopus @ Elsevier: 2-s2.0-85081002741 (DOI: 10.11239/jsmbe.57.190, CiNii: 1390002184878527360, Elsevier: Scopus)
17.	中尾英俊, 今岡真和, 岡健司, 肥田光正, 森藤武, 橋本雅至, Kazuyuki Matsumoto and Kenji Kita : 健常者の足背高測定による足部アーチ評価とアーチ高に影響を及ぼす因子, Journal of the Society of Biomechanisms, Vol.43, No.4, 256-262, 2019. (Link to Publication Site) ● Publication site (DOI): 10.3951/sobim.43.4_256 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390001277380551168 ● Search Scopus @ Elsevier (DOI): 10.3951/sobim.43.4_256 (DOI: 10.3951/sobim.43.4_256, CiNii: 1390001277380551168)
18.	Koji Bando, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Twitter User's Hobby Estimation Based on Sequential Statements Using Deep Neural Networks, International Journal of Machine Learning and Computing, Vol.9, No.2, 108-114, 2019. (Summary) With more and more frequency, users communicate with each other on social media. Many users start on Twitter or Facebook to find friends who have the same hobby. Our study proposes a method to estimate the users interests (hobby) based on tweets on Twitter. One tweet does not, in and of itself, contain a lot of information, and some tweets are not related to the users hobby. Therefore, we propose a reliable hobby estimation method by extracting features from multiple, sequential tweets. The proposed method uses Recurrent Neural Networks (RNN) which can accommodate time-series information. We also used a Convolutional Neural Networks (CNN) which can treat contextual information. We used an averaged vector of word distributed representation as a feature. Using the proposed method based on Long Short-Term Memory Recurrent Neural Networks (LSTM-RNN), we obtained a 23.72% improvement as compared with a baseline method using a Random Forest (RF) regression as a machine learning algorithm. (Keyword) Twitter / hobby estimation / neural network / random forests / support vector machine (Tokushima University Institutional Repository) ● Metadata: 113915 (Link to Publication Site) ● Publication site (DOI): 10.18178/ijmlc.2019.9.1.773 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85064990114 (Tokushima University Institutional Repository: 113915, DOI: 10.18178/ijmlc.2019.9.1.773, Elsevier: Scopus)
19.	Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Emotion Recognition for Japanese Short Sentences Including Slangs Based on Bag of Concepts Feature Trained by Large Web Text, Current Analysis on Instrumentation and Control, Vol.2019, No.2, 9-18, 2019. (Summary) The growth of Internet communication sites such as weblogs and social networking sites brought younger people especially in teens and in their 20s to create new words and to use them very often. We prepared an emotion corpus by collecting weblog article texts including new words, analyzed the corpus statistically, and proposed a method to estimate emotions of the texts. Most slang words such as Youth Slang are too ambiguous in sense classification to be registered into the existing dictionaries such as thesaurus. To cope with these words, we created a large scale of Twitter corpus and calculated sense similarities between words. We proposed to convert unknown word to semantic class id so that we might be able to process the words that were not included in the learning data. For calculation similarities between words and converting the word into word cluster id, we used the word embedding algorithms such as word2vec, or GloVe. We defined this method as a method using Bag of Concepts as feature. As a result of the evaluation experiment using several classifiers, the proposed method was proved its robustness for unknown expressions. (Keyword) Youth Slang / Unknown Words / Bag of Concepts / Word Embedding / Unsupervised Clustering (Tokushima University Institutional Repository) ● Metadata: 113245 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1571980077723683712 (Tokushima University Institutional Repository: 113245, CiNii: 1571980077723683712)
20.	Kazuyuki Matsumoto, Fuji Ren, Masaya Matsuoka, Minoru Yoshida and Kenji Kita : Slang Feature Extraction by Analyzing Topic Change on Social Media, CAAI Transactions on Intelligence Technology, 2019. (Summary) Recently, we often see words such as youth slang, neologism and Internet slang on social networking sites (SNSs) that are not registered on dictionaries. Because the documents posted to SNSs include a lot of fresh information, they are thought to be useful for collecting information. It is important to analyze these words (hereinafter referred to as slang) and capture their features for the improvement of the accuracy of automatic information collection. This paper aims to analyze what features can be observed in slang by focusing on the topic. We construct topic models from document groups including target slang on Twitter by Latent Dirichlet Allocation (LDA). With the models, we chronologically analyze change of topics during a certain period of time to find out the difference in the features between slang and general words. Then, we propose a slang classification method based on the change of features. (Keyword) Slang / Topic analysis / Latent Dirichlet Allocation / Social media, / Tweet representation (Tokushima University Institutional Repository) ● Metadata: 113246 (Link to Publication Site) ● Publication site (DOI): 10.1049/trit.2018.1060 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1049/trit.2018.1060 (Tokushima University Institutional Repository: 113246, DOI: 10.1049/trit.2018.1060)
21.	Kazuyuki Matsumoto, Akira Fujisawa, Minoru Yoshida and Kenji Kita : ASCII Art Classification based on Deep Neural Networks Using Image Feature of Characters, Journal of Software, Vol.13, No.10, 559-572, 2018. (Summary) In recent years, a lot of non-verbal expressions have been used on social media. Ascii art (AA) is an expression using characters with visual technique. In this paper, we set up an experiment to classify AA pictures by using character features and image features. We try to clarify which feature is more effective for a method to classify AA pictures. We proposed four methods: 1) a method based on character frequency, 2) a method based on character importance value and 3) a method based on image features and 4) a method based on image features of characters. We trained neural networks by using these four features. In the experimental result, the best classification accuracy was obtained in the feed forward neural networks that used image features of characters. (Keyword) ASCII Art / classification / deep neural network / character feature (Tokushima University Institutional Repository) ● Metadata: 113248 (Link to Publication Site) ● Publication site (DOI): 10.17706/jsw.13.10.559-572 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.17706/jsw.13.10.559-572 (Tokushima University Institutional Repository: 113248, DOI: 10.17706/jsw.13.10.559-572)
22.	Mei Chen, Qingmei Xiao, Kazuyuki Matsumoto, Minoru Yoshida, Kenji Kita and Xin Luo : Similarity Analysis and Repeating Pattern Detection in Fingerprint Features, Journal of Computational Methods in Sciences and Engineering, Vol.18, No.3, 553-562, 2018. (Keyword) audio fingerprint / similarity analysis / repeating pattern detection / music retrieval (Link to Publication Site) ● Publication site (DOI): 10.3233/JCM-180781 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85059945684 (DOI: 10.3233/JCM-180781, Elsevier: Scopus)
23.	Kazuyuki Matsumoto, Kyosuke Akita, Fuji Ren, Minoru Yoshida and Kenji Kita : Intimacy Estimation of the Characters in Drama Scenario, Journal of Japan Society for Fuzzy Theory and Intelligent Informatics, Vol.30, No.3, 591-604, 2018. (Summary) Recently, a portable digital device equipped with voice guidance has been widely used, increasing the demand for the usabilityconscious dialogue system. One of the problems with the existing dialogue system is its immature application to non-task-orienteddialogue. Non-task-oriented dialogue requires some scheme that enables smooth and flexible conversations with a user. For example,it would be possible to provide topics related to a person who is familiar with a user or avoid providing topics related to a personwho is not a good friend with the user, by considering relationship with others in real life of the user. In this paper, we focus on thedialogue made by the two characters in a drama scenario, and tried to express their relationship with a scale of intimacy degree.There will be such various elements related to the intimacy degree as the frequency of response to the utterance and the attitude ofa speaker during the dialogue. We focus on the emotional state of the speaker during the utterance and try to realize an intimacyestimation with high accuracy. As the evaluation result, we achieved high accuracy in intimacy estimation than the existing methodbased on speech role. (Keyword) intimacy degree / scenario dialogue / emotional state (Tokushima University Institutional Repository) ● Metadata: 113247 (Link to Publication Site) ● Publication site (DOI): 10.3156/jsoft.30.3_591 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390564237991620096 ● Search Scopus @ Elsevier (DOI): 10.3156/jsoft.30.3_591 (Tokushima University Institutional Repository: 113247, DOI: 10.3156/jsoft.30.3_591, CiNii: 1390564237991620096)
24.	Fujino Naoya, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Emotion Estimation Adapted to Gender of User Based on Deep Neural Networks, International Journal of Advanced Intelligence (IJAI), Vol.10, No.1, 121-133, 2018. (Summary) In this study, we focus on Twitter as a representative SNS and target emotion estima-tion from tweets posted on Twitter by male and female users. Specically, we constructgender-based emotion estimation models assuming that there are different word usage ten-dencies between genders. By analyzing gender-specic differences in the use of emotion-related slang and emoji, we propose a method to improve emotion estimation based onneural networks using a different distributed representation model for each gender. Ourevaluation experiments show that training with Deep Convolutional Neural Networks us-ing word's distributed representation as the feature produced higher estimation accuracythan training with Feed Forward Neural Networks. (Keyword) emotion estimation / user's gender, / deep neural networks (Tokushima University Institutional Repository) ● Metadata: 113250 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050302172853082368 (Tokushima University Institutional Repository: 113250, CiNii: 1050302172853082368)
25.	Hirokimi Fukuda, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Index Generation of BGM Video Based on Distinctive Comments, International Journal of Advanced Intelligence (IJAI), Vol.10, No.1, 113-119, 2018. (Summary) The purpose of this study is to generate indexes for background music (BGM) based ondistinctive comments annotated to BGM videos posted on Nico Nico Videos. Our proposedmethod detects the end/start positions of the BGM considering the increase/decrease inthe number of comments, comments distinctive to the songs, and exclamatory expressionsannotated to the videos. The evaluation result indicated that the proposed method couldgenerate correct indexes within a 10 s error for BGM videos less than 40 min long andwith annotations of over 30,000 comments. (Keyword) Nico Nico Video / Index generation / BGM video for work (Tokushima University Institutional Repository) ● Metadata: 113249 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050020697876372224 (Tokushima University Institutional Repository: 113249, CiNii: 1050020697876372224)
26.	Kazuyuki Matsumoto, Seiji Tsuchiya, Miyake Takeshi, Minoru Yoshida and Kenji Kita : Flame Prediction Based on Harmful Expression Judgement Using Distributed Representation, International Journal of Technology and Engineering Studies, Vol.4, No.1, 7-15, 2018. (Summary) In recent years, flaming-that is, hostile or insulting interaction-on social media has been a problem. To avoid or minimize flaming, enabling the system to automatically check messages before posting to determine whether they include expressions that are likely to trigger flaming can be helpful. We target two types of harmful expressions: insulting expressions and expressions that are likely to cause a quarrel. We first constructed an original harmful expressions dictionary. To minimize the cost of collecting the expressions, we built our dictionary semi-automatically by using word distributed representations. The method used distributed representations of harmful expressions and general expressions as features, and constructed a classifier of harmful/general expressions based on these features. Anevaluation experiment found that the proposed method was able to extract harmful expressions with an accuracy of approximately 70%. The proposed method was also able to extract unknown expressions; however, it tended to wrongly extract non-harmful expressions. The method is able to determine unknown harmful expressions not included in the basic dictionary and can identify semantic relationships among harmful expressions. Although the method cannotpresently be applied directly to multi-word expressions, it should be possible to add such a capability by introducing time-series learning. (Keyword) flame prediction / harmful expression / distributed representation / support vector machine (Tokushima University Institutional Repository) ● Metadata: 113251 (Link to Publication Site) ● Publication site (DOI): 10.20469/ijtes.4.10002-1 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1572261552700328448 ● Search Scopus @ Elsevier (DOI): 10.20469/ijtes.4.10002-1 (Tokushima University Institutional Repository: 113251, DOI: 10.20469/ijtes.4.10002-1, CiNii: 1572261552700328448)
27.	Kazuyuki Matsumoto, Fuji Ren, Minoru Yoshida and Kenji Kita : Review Score Estimation Based on Transfer Learning of Different Media Review Data, International Journal of Advanced Intelligence (IJAI), Vol.9, No.4, 541-555, 2017. (Summary) We propose a model to classify reviews based on review data from different media sources. Recently, research has been actively conducted on transfer learning between different domains with various kinds of big data as the target. The fact that evaluation expressions often vary in different domains presents a barrier to reputation analysis. Users commonly use various linguistic expressions to refer to creative works, depending on the specific media form.For example, the terms or expressions used in anime to describe creative works within that medium are different from the expressions used in comics, or games or movies. These differences can be considered as features of each individual medium. We should expect, then, that there would be differences in evaluation expressions among the various media, as well. We analyze the effects of such differences on classification accuracy by conducting transfer learning between review data from different media and demonstrate compatibility between the original (pre-transfer) and target (post-transfer) media by constructing a review classification model. As a result of our evaluation experiments, we are able to more accurately estimate review scores without using SO-Scores for training review fragments based on Long Short-Term Memory (LSTM) rather than using a method based on SO-Scores. (Keyword) review classification / transfer learning / Long Short-Term Memory / different media (Tokushima University Institutional Repository) ● Metadata: 113252 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1570572702840072192 (Tokushima University Institutional Repository: 113252, CiNii: 1570572702840072192)
28.	Kazuyuki Matsumoto, Akira Fujisawa, Minoru Yoshida and Kenji Kita : Emotion Recognition of Emoticon Based on Character Embedding, Journal of Software, Vol.12, No.11, 849-857, 2017. (Summary) This paper proposes a method for emotion estimation of emoticon based on distributed representations of character meanings of the emoticon. The existing studies on emoticon have tried to extract emoticons from texts, estimate emotions of emoticons by breaking them up into constituent parts and by using their combination as feature. Based on the recently developed technique for word embedding, we try to realize versatile emotion estimation from emoticons by training the meanings of the characters constituting the emoticons and by using them as feature unit of emoticon based on the recent developed word embedding technique. The cross validation test was conducted for the emoticons emotion estimation model that was constructed based on the deep convolutional neural networks using distributed representations of characters as feature, and its result showed that our proposed method can estimate emotion of unknown emoticons with higher F1-score than the method based on the character n-gram appearance frequency vector as feature. (Keyword) emoticon / emotion recognition / character embedding / convolutional neural networks (Tokushima University Institutional Repository) ● Metadata: 113914 (Link to Publication Site) ● Publication site (DOI): 10.17706/jsw.12.11.849-857 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.17706/jsw.12.11.849-857 (Tokushima University Institutional Repository: 113914, DOI: 10.17706/jsw.12.11.849-857)
29.	Yunong Wu, Kenji Kita, Kazuyuki Matsumoto, XIN KANG and Fuji Ren : Constructing A Short Text Conversation system Based on the Relations Between Posts and Comments, International Journal of Advanced Intelligence (IJAI), Vol.9, No.3, 369-380, 2017.
30.	Zhang Guodong, Jiang Peiling, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : An Improvement of Pedestrian Detection Method with Multiple Resolutions, Journal of Computer and Communications, Vol.5, No.9, 102-116, 2017. (Summary) In object detection, detecting an object with 100 pixels is substantially different from detecting an object with 10 pixels. Many object detection algorithms assume that the pedestrian scale is fixed during detection, such as the DPM detector. However, detectors often give rise to different detection effects under the circumstance of different scales. If a detector is used to perform pedestrian detection in different scales, the accuracy of pedestrian detection could be improved. A multi-resolution DPM pedestrian detection algorithm is proposedin this paper. During the stage of model training, a resolution factor is added to a set of hidden variables of a latent SVM model. Then, in the stage of detection, a standard DPM model is used for the high resolution objects and a rigid template is adopted in case of the low resolution objects. In our experiments, we find that in case of low resolution objects the detection accuracy of a standard DPM model is lower than that of a rigid template. In Caltech, the omission ratio of a multi-resolution DPM detector is 52% with 1 false positive per image (1FPPI); and the omission ratio rises to 59% (1FPPI) as far as a standard DPM detector is concerned. In the large-scale sample set of Caltech, the omission ratios given by the multi-resolution and the standard DPM detectors are 18% (1FPPI) and 26% (1FPPI), respectively. (Keyword) Deformable Part Model / Pedestrian Detection / Multi-Resolution / Latent SVM (Tokushima University Institutional Repository) ● Metadata: 113913 (Link to Publication Site) ● Publication site (DOI): 10.4236/jcc.2017.59007 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.4236/jcc.2017.59007 (Tokushima University Institutional Repository: 113913, DOI: 10.4236/jcc.2017.59007)
31.	Kazuyuki Matsumoto, Satoshi Tanaka, Minoru Yoshida, Kenji Kita and Fuji Ren : Ego-state Estimation from Short Texts Based on Sentence Distributed Representation, International Journal of Advanced Intelligence (IJAI), Vol.9, No.2, 145-161, 2017. (Summary) Human personality multilaterally consists of complex elements. Egogram is a method to classify personalities into the patterns according to the combinations of the five levels of ego-states. With recent development of Social Networking Service (SNS), more researches have been trying to judge personality from the statements on SNS. However, there are several problems in the personality judgment based on the superficial information of the statements. Concretely, personality is not always reflected on every statement and the tendency of the statements with influence of personality changes with the times. It is also important to collect sufficient amount of statement data including the results of personality judgment. In this paper, to realize automatic egogram judgement, we focused on the short texts on the SNS, especially microblogs, and represented the comments on Twitter by distributed representation (sentence vector) in the pre-training. Then, we tried to create a model to estimate ego-state level of each user by using a deep neural network. The experimental result showed that our proposed method estimated ego-state with higher accuracy than the baseline method based on bag of words. To investigate the change of personality according to time, we also analyzed how the match rates of the estimation results changed before/after the egogram judgment. Moreover, we confirmed that the personality pattern classification was improved by adding a feature expressing the formal degree of the sentence. (Keyword) Egogram / Personality Estimation / Twitter / Social Networking Service / Distributed Representation (Tokushima University Institutional Repository) ● Metadata: 113455 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050583647829752704 (Tokushima University Institutional Repository: 113455, CiNii: 1050583647829752704)
32.	Kazuyuki Matsumoto, Seiji Tsuchiya, Imono Misako, Minoru Yoshida and Kenji Kita : Conversion of Japanese Slang into Standard Japanese Considering Sensibility, Transactions of the Japanese Society for Artificial Intelligence, Vol.32, No.1, WII-A_1-12, 2017. (Summary) With the recent spread of communication using social media, exchanging opinions each other on web has become more common irrespective of age and sex. On the other hand, a problem called as ``Internet flaming'' often occurs along with the increase of social network service users. One of the reasons might be that the users do not recognize meanings/intentions/emotions expressed by other users' words. In this study, we focused on slangs (Internet slangs) that are often used on SNS but are not registered in dictionaries, then tried to convert them into standard words. We also intended to output more appropriate candidates by considering not only semantic similarity but also affective similarity. The proposed method conducts filtering and re-ranking over the semantically similar candidates obtained based on distributed representations to detect the inappropriate candidates as standard word by focusing on two points: (1) features of slang/standard word and (2) affective similarity between the inputted word and the candidate words. In the evaluation experiment, the proposed method obtained a higher MRR than the baseline method. (Keyword) Internet slang / standard word / sensibility / concept of word (Tokushima University Institutional Repository) ● Metadata: 113254 (Link to Publication Site) ● Publication site (DOI): 10.1527/tjsai.WII-A (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1527/tjsai.WII-A (Tokushima University Institutional Repository: 113254, DOI: 10.1527/tjsai.WII-A)
33.	Zhang Guodong, Jiang Peilin, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Reidentification of Persons Using Clothing Features in Real-Life Video, Applied Computational Intelligence and Soft Computing, Vol.2017, 1-9, 2017. (Summary) Person reidentification, which aims to track people across nonoverlapping cameras, is a fundamental task in automated video processing. Moving people often appear differently when viewed from different nonoverlapping cameras because of differences in illumination, pose, and camera properties. The color histogram is a global feature of an object that can be used for identification. This histogram describes the distribution of all colors on the object. However, the use of color histograms has two disadvantages. First, colors change differently under different lighting and at different angles. Second, traditional color histograms lack spatial information. We used a perception-based color space to solve the illumination problem of traditional histograms. We also used the spatial pyramid matching (SPM) model to improve the image spatial information in color histograms. Finally, we used the Gaussian mixture model (GMM) to show features for person reidentification, because the main color feature of GMM is more adaptable for scene changes, and improve the stability of the retrieved results for different color spaces in various scenes. Through a series of experiments, we found the relationships of different features that impact person reidentification. (Keyword) Clothing Feature / Gaussian mixture mode / spatial pyramid matching / person reidentification / perception-based color space (Tokushima University Institutional Repository) ● Metadata: 113911 (Link to Publication Site) ● Publication site (DOI): 10.1155/2017/5834846 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1155/2017/5834846 (Tokushima University Institutional Repository: 113911, DOI: 10.1155/2017/5834846)
34.	Kazuyuki Matsumoto, Minoru Yoshida, Seiji Tsuchiya, Kenji Kita and Fuji Ren : Slang Analysis Based on Variant Information Extraction Focusing on the Time Series Topics, International Journal of Advanced Intelligence (IJAI), Vol.8, No.1, 84-98, 2016. (Summary) Recently, with increase of the number of users of Social Networking Sites, online communications have been more and more actively made, raising the possibility to use the big data on SNS for analyzing the diversity of language. Japanese language uses varieties of character types and such character types are combined and used for creating words and phrases. Therefore, it is difficult to morphologically analyze such words and phrases even though morphological analysis is a basic processing in natural language processing. Such words and phrases that are not registered in morphological analysis dictionaries are usually not defined strictly and semantic interpret for them seems to vary depending on deindividual. In this study, we chronologically analyze the topics related to slang on Twitter. In this paper, as a validation experiment, we conducted a topic analysis experiment chronogically by using the sequential tweet data, and discussed the difference of topic change according to the slang types. (Keyword) Slang / Topic analysis / Time-series analkysis (Tokushima University Institutional Repository) ● Metadata: 113910 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1571698602718115840 (Tokushima University Institutional Repository: 113910, CiNii: 1571698602718115840)
35.	Akira Fujisawa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : An Illustration Image Classification Focusing on Infrequent Colors, International Journal of Advanced Intelligence (IJAI), Vol.8, No.1, 72-83, 2016. (Summary) Illustration images used in comics or animation works have the style which is emotional feature. The style is represented by the various elements of the illustration image. There are a few researches which treat the style of illustration image is fewer. In this paper, we tried to model the style by using an image feature, and to classify an illustration image by the style. We thought that the color which appeared infrequently represent the style of illustration image. Therefore, we proposed the method for creating the color histogram which emphasize colors appearing infrequently. We called that as Infrequency histogram; IF-hist. To test effectiveness of the histogram, we experimented to classify the two styles which are For boys and For girls based on the IF-hist. As the result of the experiment, the case using the IF-hist, the precision of classification result by the style For girls is 93%. It is 50% higher than the case using existing normal color histogram. Moreover the precision of the other classification results have improved. (Keyword) Image Classification / Illustration Image / Color Histogram / Style / IF-hist (Tokushima University Institutional Repository) ● Metadata: 113909 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050583647829579648 (Tokushima University Institutional Repository: 113909, CiNii: 1050583647829579648)
36.	Zhang Guodong, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Pedestrian Re-identification using Color Feature in Multi Surveillance Video, International Journal of Advances in Electronics and Computer Science, Vol.3, No.3, 11-14, 2016. (Summary) In this paper we present a system to solve the problem of moving pedestrian re-identification in surveillancevideo. Surveillance video has low-resolution, high video noise and limited monitoring scope. Our proposed framework mustdeal with several problems such as variations of illumination conditions, poses and occlusions. How to extract the robustfeature that can adapt the problems have been the task. The people of global color approaches do not change in the processof monitoring. Our paper use the color histogram as feature descriptors and choose RGB HSV and UVW for color space.Traditional histogram method extract the global color approach as the feature. The object color structure information will beneglected. We use the SPM model supplement the structure information for the histogram. The results of a test from a realsurveillance system show that our method can provide a probability of matching. (Keyword) Pedestrian Re-identification / UWV Color space / Color Histogram / SPM (Tokushima University Institutional Repository) ● Metadata: 113906 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050865122806291456 (Tokushima University Institutional Repository: 113906, CiNii: 1050865122806291456)
37.	Sun Zhuoran, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : A Study on the Review Analysis for the Automatic Evaluation of the Product, International Journal of Management and Applied Science (IJMAS), Vol.2, No.3, 2016. (Summary) Nowadays, e-business has been never more popular and more convenient. Online merchants have flourished andmerchandize have diversified. However, its still difficult for users to quickly locate their desired products on online markets.Often times, an online shopper will evaluate ratings and comments from other users to speculate the current product. Inparticular, the easiest and the most intuitive way is to see the scores to determine the quality of the product. Therefore, if wecan automatically identify and analyze all the reviews and accurately and objectively give a score to a particular product, itwill be more efficient for users to search merchandize and for online business owners to manage the products they sell. (Keyword) e-business / Reviews / Analysis / Automatic Evaluation / management of technology (Tokushima University Institutional Repository) ● Metadata: 113907 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050020697876158592 (Tokushima University Institutional Repository: 113907, CiNii: 1050020697876158592)
38.	市川賢, Norihide Kitaoka, Satoru Tsuge, 武田一哉 and Kenji Kita : 種々のテキスト検索モデルの頑健性向上による音声ドキュメント検索の高精度化, Transactions of Information Processing Society of Japan, Vol.56, No.3, 1003-1012, 2015. (Summary) We apply modifications to typical text retrieval methods based on vector space model, query likelihood model, and relevance model, to make them robust to out-of-vocabulary words and misrecognition. We propose novel query expansion methods and combination methods of syllable recognition-based retrieval with word recognition-based retrieval, for these typical methods. We used NTCIR-9 SpokenDoc task to evaluate them. Each modified method achieved better result than baseline. The methods based on stochastic models like the query likelihood model and the relevance model achieved better performance than the vector space model. The performance of our proposed methods was better than the best result published in NTCIR-9 competition. (Keyword) 音声ドキュメント検索 / vector space model / クエリ尤度モデル / 適合モデル / 検索質問拡張 / 音節認識結果による検索 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050845762835865728 (CiNii: 1050845762835865728)
39.	Kazuyuki Matsumoto, Kyosuke Akita, Xielifuguli Keranmu, Minoru Yoshida and Kenji Kita : Extraction Japanese Slang from Weblog Data Based on Script Type and Stroke Count, Procedia Computer Science, Vol.35, No.2014, 464-473, 2014. (Summary) Young people commonly use slang in the texts for weblogs or Social Networking Sites. How to treat such slang words properly is one of the problems in the field of text mining. In this paper, we examined several methods to extract Japanese slang called Wakamono Kotoba, which is particularly used by young people, by focusing on its script type and stroke count. In the evaluation experiment, a high precision was obtained when we adopted script type for extraction. (Keyword) Japanese slang / Conditional Random Fields / unknown word / stroke count / script type (Link to Publication Site) ● Publication site (DOI): 10.1016/j.procs.2014.08.127 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84924135505 (DOI: 10.1016/j.procs.2014.08.127, Elsevier: Scopus)
40.	Satoru Tsuge, Hiromasa Ohashi, Ken Ichikawa, Norihide Kitaoka, Kazuya Takeda and Kenji Kita : 音声ドキュメント検索における種々の検討および線形補間係数を自動決定する検索質問拡張, 情報処理学会論文誌, Vol.55, No.6, 1625-1636, 2014. (Summary) 本論文では，2011年に開催されたNTCIR-9ワークショップのコアタスクSpokenDoc内のSDRサブタスクに参加した際に行った音声ドキュメント検索に対する種々の検討を報告する．ベクトル空間モデルによるテキスト検索において，索引語の単位，索引語重みづけ方法，ベクトル空間に射影する音声認識候補数，検索質問拡張方法などの比較を行い，各ベクトル空間で計算される類似度を統合して音声ドキュメント検索を行う手法を提案した． (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050001337904502272 (CiNii: 1050001337904502272)
41.	Yunong Wu, Kenji Kita and Kazuyuki Matsumoto : Three Predictions are Better than One: Sentence Multi-Emotion Analysis from Different Perspectives, IEEJ Transactions on Electrical and Electronic Engineering (TEEE), Vol.9, No.6, 642-649, 2014. (Summary) Emotion prediction has been a core task in affective computing, which aims at finding the thorough human mental states byanalyzing peoples activities. In this paper, we focus on predicting emotions in the public online blogs from different people, byextracting as many reasonable emotions for each blog sentence as possible. Concretely, we consider three different perspectivesfor analyzing the multiple emotions in a sentence: (i) predict sentence emotions by examining the emotion related topics in aglobal sense; (ii) predict the sentence emotions from the context-sensitive word emotions; and (iii) predict sentence emotions byconsidering the emotional significance in the local bag of words. We build different probabilistic models from each perspective, toseparately generate the sentence emotion probabilities. We then integrate these probabilistic models to jointly predict the emotionprobabilities. Because the component models are based on different emotional assumptions with distinct features, the integratedpredictions should predict emotions from more general perspectives and therefore yield better results. In the experiment, weemploy different evaluation criteria to compare the multi-emotion predictions from the single and integrated models. Comparedto the results in the baseline model, our bi-integrated models achieve 8.69% higher Micro F1 and 7.78% higher Macro F1 scores,on average. Moreover, our tri-integrated model acquires 10.00% higher Micro F1 and 9.19% higher Macro F1 scores than thebaseline results, which proves our assumption, and suggests interesting features in the different emotion perspectives. © 2014Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc. (Keyword) sentence emotion analysis / multi-emotion / L-LDA / CRF / LGR (Link to Publication Site) ● Publication site (DOI): 10.1002/tee.22020 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84907939400 (DOI: 10.1002/tee.22020, Elsevier: Scopus)
42.	Nakagawa Shohei, Motoyuki Suzuki, Kazuyuki Matsumoto and Kenji Kita : Emotion Estimation Method Based on Normalization of Prosodic Features by Using Synthesis Speech, The Transactions of the Institute of Electronics, Information and Communication Engineers D, Vol.J97-D, No.3, 533-539, 2014. (Summary) 本論文では，音声からの感情推定において特徴量を正規化して識別を行う方法を提案する．従来感情識別に用いられている平均パワーやピッチといった韻律的特徴量は，感情による変化だけではなく発話内容そのものによって大きく変化する．そのため，たとえ同じ感情で発話されたとしても発話内容が異なれば異なる韻律となり，感情推定の性能低下を招き得る．そこで本論文では，平静の感情で同じ発話内容を発話した音声からも特徴量を抽出し，それとの差分に注目することで感情による特徴量の変化のみを抽出する方法を提案する．使用している5種類の特徴量(平均パワー，ピッチ等)ごとに減算による正規化，除算による正規化，正規化なし，の三つの方法で正規化し，全ての組み合わせの中で最も性能が向上する組み合わせを探索した．その結果，最適な組み合わせによる識別性能は，正規化を行わない従来法と比較して5.98%向上した．正規化法に関する分析を行ったところ，平均パワーは正規化が必要(演算は減算でも除算でも大きな差はない)であり，一方ピッチとMFCCは正規化なし，残りの二つの特徴量についてはどちらでも性能は大きくは変化しないことがわかった． (Keyword) 感情識別 / 特徴量正規化 / 合成音声 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1520009407264043136 (CiNii: 1520009407264043136)
43.	Keranmu Xielifuguli, Akira Fujisawa, Yusuke Kusumoto, Kazuyuki Matsumoto and Kenji Kita : Pleasant/Unpleasant Filtering for Affective Image Retrieval Based on Cross-Correlation of EEG Features, Applied Computational Intelligence and Soft Computing, Vol.2014, 2014. (Summary) We propose a sensitivity filtering technique for discriminating preferences (pleasant/unpleasant) for images using a sensitivity image filtering system based on EEG. Using a set of images retrieved by similarity retrieval, we perform the sensitivity-based pleasant/unpleasant classification of images based on the affective features extracted from images with the maximum entropy method (MEM). In the present study, the affective features comprised cross-correlation features obtained from EEGs produced when an individual observed an image. However, it is difficult to measure the EEG when a subject visualizes an unknown image. Thus, we propose a solution where a linear regression method based on canonical correlation is used to estimate the cross-correlation features from image features. (Keyword) image processing / information retrieval system / electroencephalogram / sentience computer (Link to Publication Site) ● Publication site (DOI): 10.1155/2014/415187 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1155/2014/415187 (DOI: 10.1155/2014/415187)
44.	Qingmei Xiao, Saito Narumi, Kazuyuki Matsumoto, Luo Xin, Yokota Yasushi and Kenji Kita : Index Compression for Audio Fingerprinting Systems Based on Compressed Suffix Array, International Journal of Information and Education Technology (IJIET), Vol.3, No.4, 455-460, 2013. (Summary) In this paper, we propose a method for index compression using a compressed suffix array. Taking advantage of the fact that the repetitive characters occur frequently in higher bits of the sorted audio fingerprint data, the proposed method compresses the index by encoding the 8-bit data sequences by Run Length Encoding. Vertical Coding is also used to compress the array, wherein the positions of the sorted data are stored. (Link to Publication Site) ● Publication site (DOI): 10.7763/IJIET.2013.V3.317 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.7763/IJIET.2013.V3.317 (DOI: 10.7763/IJIET.2013.V3.317)
45.	Qingmei Xiao, Daito Yusaku, Kazuyuki Matsumoto, Motoyuki Suzuki and Kenji Kita : Fast Search Method for Audio Fingerprinting Systems Based on Query Multiplexing, IEEJ Transactions on Electronics, Information and Systems, Vol.132, No.9, 1481-1487, 2012. (Summary) In music information retrieval, a huge search space has to be explored because a query audio clip can start at any position of any music in the database, and also a query is often corrupted by highly significant noise and distortion. Audio fingerprints have attracted much attention recently for providing compact representation of the perceptually relevant parts of audio signals. In this paper, we propose an extremely fast method of exploring a huge hamming space for audio fingerprinting systems. The effectiveness of our method has been evaluated by experiments using databases of 8,740 real songs and 800 artificially corrupted and 268 real queries. (Keyword) music information retrieval / audio fingerprint / hamming space / query multiplexing (Link to Publication Site) ● Publication site (DOI): 10.1541/ieejeiss.132.1481 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390282679584466688 ● Search Scopus @ Elsevier (DOI): 10.1541/ieejeiss.132.1481 (DOI: 10.1541/ieejeiss.132.1481, CiNii: 1390282679584466688)
46.	Zhang Le, Kazuyuki Matsumoto, Motoyuki Suzuki and Kenji Kita : Skelton Pruning Based on the Total Bisector Angle of the End Branches, IEEJ Transactions on Electronics, Information and Systems, Vol.132, No.5, 750-758, 2012. (Summary) Skeletonization is an important technique for shape representation, but the skeleton's high sensitivity to boundary noise hampers its application to automatic shape matching. In this paper, we propose a linear-time skeleton pruning algorithm based on the total bisector angle of the end branches. The main idea of the proposed algorithm is to iteratively delete the end branches having the smallest total bisector angle. Since the total bisector angle considers both the local and the global skeleton information, the proposed algorithm removes all redundant branches generated by boundary noise and obtains the ideal skeleton. The experimental results show that the proposed algorithm is highly adaptable, which means that under the same threshold conditions, the pruned skeletons obtained by our proposed algorithm are in accord with human visual perception for most shapes in the MPEG-7 and Tari 56 datasets. Consequently, the pruned skeletons can be applied to automatic shape matching. (Keyword) Skelton / Skelton Pruning / Total Bisector Angle / End Branch (Link to Publication Site) ● Publication site (DOI): 10.1541/ieejeiss.132.750 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390282679586212736 ● Search Scopus @ Elsevier (DOI): 10.1541/ieejeiss.132.750 (DOI: 10.1541/ieejeiss.132.750, CiNii: 1390282679586212736)
47.	Kazuyuki Matsumoto, Kenji Kita and Fuji Ren : Emotional Vector Distance Based Sentiment Analysis of Wakamono Kotoba, China Communications, Vol.9, No.3, 87-98, 2012. (Summary) In this paper, we propose a method for estimating emotion in Wakamono Kotoba that were not registered in the system, by using Wakamono Kotoba example sentences as features. The proposed method applies Earth Mover's Distance(EMD) to vector of words. As a result ofthe evaluation experiment using 14,440 sentences, higher estimation accuracy is obtained by considering emotional distance between words - an approach that had not been used in the conventional research - than by using only word importance value. (Keyword) Wakamono Kotoba / sentiment analysis / emotion estimation / Earth Mover's Distance (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84861787581 (Elsevier: Scopus)
48.	Yunong Wu, Kenji Kita, Fuji Ren, Kazuyuki Matsumoto and XIN KANG : Exploring the Importance of Modification Relation for Emotional Keywords Annotation and Emotion Types Recognition, International Journal of Intelligent Engineering and Systems, Vol.4, No.4, 19-26, 2011. (Summary) In this study, we make a scheme to explore the importance of modification relation for the emotional keywords annotation and emotion types recognition. We extract three modification features which are degree words,negative words and conjunctions from the Chinese emotion corpus named Ren-CECps. Beside word and part-of speech,three modification relations are adopted as feature in this study. We have carried out eight experiments with different feature sets for emotional keywords annotation and emotion types recognition in sentence level. Eight basic emotion types have been selected and Conditional Random Fields have been employed as the algorithm. In the part of evaluation, we demonstrate the importance of the modification features and our experiment results show the effectiveness of the modification features for improving the performance of emotional keywords annotation and emotion types recognition. (Keyword) Emotional Keywords Annotation / Emotion types recognition / Modification relation / CRFs (Link to Publication Site) ● Publication site (DOI): 10.22266/ijies2011.1231.03 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84859614924 (DOI: 10.22266/ijies2011.1231.03, Elsevier: Scopus)
49.	Mei Chen, Kenji Kita and Xin Luo : Cluster-Based Reputation Model in Peer-to-Peer Network, International Journal of Machine Learning and Computing, Vol.1, No.4, 366-371, 2011. (Summary) This paper presents a cluster-based reputation model (CBRM) in Peer-to-Peer network. The model is consisted by reputation mechanism and cluster. In the model, we take the reputation mechanism for realizing the security transaction; and the network topology structure of CBRM adopts the cluster, so efficiency of reputation management is noticeably raised. (Link to Publication Site) ● Publication site (DOI): 10.7763/IJMLC.2011.V1.54 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.7763/IJMLC.2011.V1.54 (DOI: 10.7763/IJMLC.2011.V1.54)
50.	Wenbin Zhang, Haoze Lu, Yasuo Horiuchi, Satoru Tsuge, Kenji Kita and Shingo Kuroiwa : Text-Independent Speaker Identification Based on Reducing Intersession Variability of Speech Feature Using PCA Transformation, Journal of Signal Processing, Vol.15, No.4, 275-278, 2011. (Summary) In text-independent speaker identification, the phonetic variability and the intersession variability strongly afect the performance of speaker identification. In this paper, PCA transformation is utilized to reduce intersession variability of speech data in the observed space. The proposed method reduced the identification error rate by 42.6% compared with the conventional MFCCs and reduced the identification error rate by 27.2% compared with the MFB-PCA. (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1524232505592363648 (CiNii: 1524232505592363648)
51.	Masami Shishibori, Sangkon Lee and Kenji Kita : An E-mail Filtering Method Based on Multi-attribute Values of Users Profile, International Journal of Computer Applications in Technology, Vol.40, No.4, 273-279, 2011. (Summary) E-mail is one of the often used communication methods; however, a lot of e-mail documents including useless information are distributed to computers. To solve this problem, we propose a method to filter out unimportant e-mail documents by judging the content of each document. Our method constructs personal profiles. They consist of multi-attribute values such as 'Sender', 'Theme' and 'Type' of each document. By using the profile, this method can distinguish whether a new e-mail document is important or not for the users. It is found that the accuracy of this method is higher than that of the traditional methods. (Link to Publication Site) ● Publication site (DOI): 10.1504/IJCAT.2011.041656 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1504/IJCAT.2011.041656 (DOI: 10.1504/IJCAT.2011.041656)
52.	Masami Shishibori, Daichi Koizumi and Kenji Kita : Fast Retrieval Algorithm for Earth Mover's Distance Using EMD Lower Bounds and Skipping Algorithm, Advances in Multimedia, Vol.2011, No.Article ID 421820, 1-9, 2011. (Summary) The earth mover's distance (EMD) is a measure of the distance between two distributions, and it has been widely used in multimedia information retrieval systems, in particular, in content-based image retrieval systems. When the EMD is applied to image problems based on color or texture, the EMD reflects the human perceptual similarities. However, its computations are too expensive to use in large-scale databases. In order to achieve efficient computation of the EMD during query processing, we have developed fastEMD, a library for high-speed feature-based similarity retrievals in large databases. This paper introduces techniques that are used in the implementation of the fastEMD and performs extensive experiments to demonstrate its efficiency. (Link to Publication Site) ● Publication site (DOI): 10.1155/2011/421820 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1155/2011/421820 (DOI: 10.1155/2011/421820)
53.	Yoshihide Matsumoto, Tadashi Uemiya, Masami Shishibori and Kenji Kita : A Method for Detecting Subtitle Regions in Videos Using Video Text Candidate Images and Color Segmentation Images, International Journal of Advanced Intelligence (IJAI), Vol.2, No.1, 37-55, 2010. (Summary) In this paper, a method for detecting text regions in digital videos with telop, such as drama, movie and news programming, is proposed. The typical characteristics of telop are that it does not move, and that its edges are strong. This method takes advantage of these characteristics to produce video text candidate images.
54.	Tadashi Uemiya, Yoshihide Matsumoto, Koizumi Daichi, Masami Shishibori and Kenji Kita : Fast Multidimensional Nearest Neighbor Search Algorithm Based on Ellipsoid Distance, International Journal of Advanced Intelligence (IJAI), Vol.1, No.1, 89-107, 2009. (Summary) The nearest neighbor search in high-dimensional spaces is an interesting and important problem that is relevant for a wide variety of applications, including multimedia information retrieval, data mining, and pattern recognition. For such applications, the curse of high dimensionality tends to be a major obstacle in the development of efficient search methods. This paper addresses the problem of designing a new and efficient algorithm for high-dimensional nearest neighbor search based on ellipsoid distance. The proposed algorithm uses Cholesky decomposition to perform data conversion beforehand so that calculation by ellipsoid distance function can be replaced with calculation by Euclidean distance, and it improves efficiency by omitting an unnecessary operation. Experimental results indicate that our scheme scales well even for a very large number of dimensions.
55.	Xiaoying Tai, Lidong Wang, Qin Chen, Fuji Ren and Kenji Kita : A New Method of Medical Image Retrieval based on Color-Texture Correlogram and GTI Model, International Journal of Information Technology & Decision Making, Vol.8, No.2, 239-248, 2009. (Summary) This paper presents a method for endoscopic image retrieval based on color texture correlogram and Generalized Tversky's Index (GTI) model. First we define a new image feature named color texture correlogram, which is the extension of color correlogram. The texture image extracted by texture spectrum algorithm is combined with color feature vector, and then we calculate the spatial correlation of colortexture feature vector. Similarity metric is also the key technology during domain of image retrieval, GTI model is used in medical image retrieval for similarity metric, and the technique of relevance feedback is used in the algorithm to enhance the efficiency of retrieval. Experimental results show that the method discussed in this paper is much more effective. (Link to Publication Site) ● Publication site (DOI): 10.1142/S0219622009003363 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1142/S0219622009003363 (DOI: 10.1142/S0219622009003363)
56.	Kazuhide Kanenishi, Kenji Matsuura, Takahiro Oie, Masahiko Sano, Kenji Kita and Yoneo Yano : Design Issue and Operation Method of the University Information Portal Based on the Existing Web Based System, Academic Information Processing Environment Research, Vol.12, 34-42, 2009. (Summary) 現在，高等教育機関において，学内の情報化を推進するため大学ポータルの導入が進んでいる．本稿では，大学ポータルのモデルを提案する．大学ポータルとは，単一のWEBベースのシステムではなく，分散するデータベースやWEBシステムが統合したシステムであり，学内情報の基盤となるフレームワークのことである．徳島大学の大学ポータルを事例として，大学ポータルの構築，運用に関する問題を具体的に考察する．運用の様子を示し，考察をおこなう．最後に，大学ポータルの課題について考察する． (Keyword) Single Signe On / single signe on / Authentication infrastructure / authentication infrastructure / Federation / federation / Authorization / authorization / WEB Service / web service / 認証基盤 / ニンショウキバン / 連携 / レンケイ / 認可 / ニンカ / WEB サービス / web サービス (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1521136281064684160 (CiNii: 1521136281064684160)
57.	Masami Shishibori, Nobuki Nishikawa and Kenji Kita : A Method to Retrieve Telops based on the Distance of Character Image Features between Queries and Telops, International Journal of Computer Applications in Technology, Vol.35, No.2/3/4, 234-240, 2009. (Summary) Video scene retrieval methods based on telop characters can retrieve the corresponding scenes to the query from the huge video data. In this paper, a new video scene retrieval method based on telop characters is proposed. In order to specify the suitable scene, this method recognizes the only corresponding telop characters to the query keyword, not all the characters. This method calculates the distance between each image features of telop characters and template image features of query keyword. The number of distance calculations can decrease by indexing the multi-dimensional data for image features of telop characters. (Link to Publication Site) ● Publication site (DOI): 10.1504/IJCAT.2009.026600 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-67650324065 (DOI: 10.1504/IJCAT.2009.026600, Elsevier: Scopus)
58.	Masami Shishibori and Kenji Kita : Construction of an Image Knowledge Database on WWW Image Retrieval Systems, International Journal of Computer Applications in Technology, Vol.34, No.2, 108-114, 2009. (Summary) This paper proposes a method to construction of an image knowledge database to store feedback information on WWW image retrieval systems. When the query has not yet been registered in the database, the image features of the true image selected by the user are registered in the database. On registration, the similarity of the features is considered, and only representative features are registered. On retrieval, this database is used as the supervisor data for the query, and images closer to the supervisor data are presented with priority. (Link to Publication Site) ● Publication site (DOI): 10.1504/IJCAT.2009.023616 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1504/IJCAT.2009.023616 (DOI: 10.1504/IJCAT.2009.023616)
59.	Masami Shishibori, 宮本裕也, Satoru Tsuge and Kenji Kita : An Improved Method to Select Candidates on Metric Index VP-tree, The Transactions of the Institute of Electronics, Information and Communication Engineers D, Vol.J91-D-I, No.3, 677-686, 2008. (Summary) 本論文では,距離索引の一つであるVP-treeの改良法を提案する.VP-treeは検索時にルートから検索範囲に適合するノードをたどり,最終的にたどり着いたリーフノードに格納されたデータとの距離を算出する.しかし,リーフノードにおける距離計算が増大すると検索速度が遅くなる.そこで,リーフノードにおける絞込み法に着目し,その改良法として検索入力に対する最近傍点を三角不等式の基準点として用いる手法を提案する.実際の画像データを用いて評価実験を行った結果,従来手法に比べ検索時間を5∼12%削減することができた. (Keyword) マルチメディアデータベース / 距離索引 / VP-tree / 三角不等式 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1520853834232752000 (CiNii: 1520853834232752000)
60.	Shiro Ajioka, Satoru Tsuge, Masami Shishibori and Kenji Kita : Fast Multidimensional Nearest Neighbor Search Algorithm Using Priority Queue, Electrical Engineering in Japan, Vol.164, No.3, 69-77, 2008. (Summary) Nearest neighbor search in high dimensional spaces is an interesting and important problem which is relevant for a wide variety of applications, including multimedia information retrieval, data mining, and pattern recognition. For such applications, the curse of high dimensionality tends to be a major obstacle in the development of efficient search methods. This paper addresses the problem of designing an efficient algorithm for high dimensional nearest neighbor search using a priority queue. The proposed algorithm is based on a simple linear search algorithm and eliminates unnecessary arithmetic operations from distance computations between multidimensional vectors. Moreover, we propose two techniques, a dimensional sorting method and a PCA-based method, to accelerate multidimensional search. Experimental results indicate that our scheme scales well even for a very large number of dimensions. (Keyword) nearest neighbor search / high-dimensional space / fast linear search / priority queue / multimedia information retrieval (Link to Publication Site) ● Publication site (DOI): 10.1002/eej.20502 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1002/eej.20502 (DOI: 10.1002/eej.20502)
61.	西川伸紀, Masami Shishibori, Satoru Tsuge and Kenji Kita : 検索質問と字幕の文字画像特徴量間の距離に基づく字幕検索手法, Transactions of Information Processing Society of Japan, Vol.48, No.SIG20, 28-38, 2007. (Summary) In this paper, a new video telop retrieval method based on telop characters is proposed. This method calculates the distance between each image features of telop characters and template image features of query keyword. The number of distance calculations can decrease by indexing the multidimensional data for image features of telop characters. Experimental results, using 91 query keywords, show that the average precision of proposed method using 1-gram feature becomes 98.61%, and using 2-gram feature becomes 99.59%. Moreover, the retrieval time can be obtained in about 0.5 seconds when using 2-gram feature. (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050001337891836672 (CiNii: 1050001337891836672)
62.	Kazuma Hara, Shingo Kuroiwa, Kouji Tanaka, Satoru Tsuge, Fuji Ren, Masami Shishibori and Kenji Kita : Acoustic Model Adaptation Using Speech Synthesis for Codec Speech Recognition, The Transactions of the Institute of Electronics, Information and Communication Engineers D, Vol.J90-D, No.9, 2541-2549, 2007. (Summary) 現在,IP電話端末から音声認識システムを利用する場合,音声データ通信時における音声圧縮・伸張(符号化)の影響より固定電話と比較して音声認識精度が低下する問題がある.これを回避するため,符号化された多量の音声データを用いて音響モデルを再学習,及び適応する手法が提案されている.しかし,再学習や適応に必要となる多量の音声データを収集することは多大な労力と時間が必要となる.そこで本論文では,音声データの収集を必要としない音響モデル適応手法を提案する.本提案手法は,認識に用いる音響モデルから適応に必要な音声データを自己生成する手法である.本論文ではこの手法を実現するため,音響モデルの各分布より音素波形を合成し適応に用いる「合成音素波形による音響モデルの分布適応法」と連続音声を合成し適応に用いる「連続合成音声を用いた音響モデル適応法」を提案する.符号化方式G723.1を用いた音声認識実験結果より,連続合成音声を用いた音響モデル適応法は適応前の音響モデルの認識精度を改善することが分かった.これらの結果より,提案手法は適応データの収集を必要とせず,符号化音声を高精度に認識する手法として有効であるといえる. (Keyword) 符号化音声認識 / 音声符号化 / HMM音声合成 / 音響モデル適応 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1520853833503954304 (CiNii: 1520853833503954304)
63.	Xin Luo, Masami Shishibori, Fuji Ren and Kenji Kita : Incorporate feature space transformation to content-based image retrieval with relevance feedback, International Journal of Innovative Computing, Information and Control, Vol.3, No.5, 1237-1250, 2007. (Summary) In recent years, the employment of feedback information to improve image retrieval precision has become a hot subject in research field. But in the traditional relevance feedback methods, both relevant and non-relevant assigned information were required for the retrieval system. For the sake of practicality and convenience, this paper assumes that users only need to label the images from the previous query as relevant, which generates a new vector as feedback information. Through the feature space transformation, it is an adjustment in the spatial resolution of the feature space. The spatial resolution around relevant samples is contracted. Meanwhile, the analysis of the user's intension together with relevant forecast of the interested objects in the database make it posiible for the less similar vectors to get closer to the query vector and thus increasing query precision. Compared with the traditional relevance feedback approach, our method is shown to obviously improve the retrieval feedback performance. (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-63649127226 (Elsevier: Scopus)
64.	Atsushi Inouchi, Shuichi Shinohara, Hiroshi Inoue, Kenji Kita and Mitsuo Itakura : Identification of specific sequence motifs in the upstream region of 242 human miRNA genes., Computational Biology and Chemistry, Vol.31, No.3, 207-214, 2007. (Summary) We have identified novel over-represented and conserved motifs in the upstream regions of human and mouse miRNA stem-loop sequences by means of a new bioinformatic processing regimen. We observed sequence conservation -500 bp upstream in 189 human and mouse miRNAs declining with increasing distance from their putative miRNA stem-loop origin. We also found relatively GC-rich regions having more than 50% of guanine+cytosine (G+C) content at about -30 and -170 bp relative to human miRNA stem-loop sequence origin. To further identify specific sequence motifs that might be involved in the transcriptional regulation of miRNA precursors, we first searched 500 bp upstream sequences of 194 non-redundant human miRNA stem-loop sequences for frequently occurring motifs 5-15 bp long. We then found the comparable occurrences of the 20 most frequent motifs in the 2000 bp upstream regions of 242 human and 290 mouse miRNAs. The significantly reduced frequency of occurrence of all 20 motifs in the regions 2000 bp upstream of 23,570 human RefSeq genes demonstrated that these motifs were specific to the upstream miRNA sequences. The most frequently observed motif M1 (GTGCTTMTAGTGCAG), with a MEME E-value of 3.8e-57 was distributed within 500 bp upstream of stem-loop sequences and was also miRNA-specific. We suggest that these over-represented motif sites are good candidates for experimentally testing miRNA expression as well as possible interaction with regulatory factors. (Keyword) Animals / Base Composition / Base Sequence / Computational Biology / Consensus Sequence / CpG Islands / Databases, Genetic / GC Rich Sequence / Humans / Mice / MicroRNAs / Response Elements (Link to Publication Site) ● Publication site (DOI): 10.1016/j.compbiolchem.2007.03.011 (Link to Search Site for Scientific Articles) ● PubMed @ National Institutes of Health, US National Library of Medicine (PMID): 17499549 ● Summary page in Scopus @ Elsevier: 2-s2.0-34248596571 (DOI: 10.1016/j.compbiolchem.2007.03.011, PubMed: 17499549, Elsevier: Scopus)
65.	Satoru Tsuge, Shingo Kuroiwa, Masami Shishibori, Fuji Ren and Kenji Kita : Real-timeFrequency Characteristic Normalization for Distributed Speech Recognition, Transactions of Information Processing Society of Japan, Vol.48, No.2, 900-908, 2007. (Summary) 本論文では，分散音声認識における入力系の周波数特性の際による認識率劣化を抑制する周波数特性正規化手法として，複数参照ケプストラムを用いた実時間周波数特性正規化手法を提案した． (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050001337882493184 (CiNii: 1050001337882493184)
66.	Masami Shishibori, 大西泰代, Satoru Tsuge and Kenji Kita : Similar Music Retrieval for the Query-by-humming Using the Earth Mover's Distance, Transactions of Information Processing Society of Japan, Vol.48, No.1, 300-311, 2007. (Summary) In this paper, we propose a new similar music retrieval method based on humming query using the Earth Mover's Distance as the distance measure. Computing the EMD is based on a solution to the transportation problem, and the EMD is applied as the distance measure on similar image retrieval systems. In addition, we focus that the time complexity of the EMD is exponential worst case toward the number of notes, the improved method to decrease the number of notes in the music feature is also proposed. (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050845762812636800 (CiNii: 1050845762812636800)
67.	小泉大地, Masami Shishibori, 中川嘉之, Satoru Tsuge and Kenji Kita : WWW画像検索システムにおける有害画像フィルタリング手法, Transactions of Information Processing Society of Japan, Vol.47, No.SIG8, 147-156, 2006. (Summary) In this paper, we propose a new filtering method using partial URL-based weighing scheme. The method first gathers a lot of hazardous URLs from search results of conventional WWW image retrieval systems by hazardous keywords. The method next decomposes a full URL into several partial paths, then estimates hazardous score for each partial path based on frequencies of paths. And it filters out hazardous images by distinguishing partial URL that there is possibility of hazardous. Experimental results show that the proposed method can improve the accuracy of filtering out hazardous images from search results of conventional systems. (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050282812868568960 (CiNii: 1050282812868568960)
68.	味岡四郎, Satoru Tsuge, Masami Shishibori and Kenji Kita : Fast Multidimensional Nearest Neighbor Search Algorithm Using Priority Queue, IEEJ Transactions on Electronics, Information and Systems, Vol.126, No.3, 353-360, 2006. (Summary) Nearest neighbor search in high dimensional spaces is an interesting and important problem which is relevant for a wide variety of applications, including multimedia information retrieval, data mining, and pattern recognition. For such applications, the curse of high dimensionality tends to be a major obstacle in the development of efficient search methods. This paper addresses the problem of designing an efficient algorithm for high dimensional nearest neighbor search using a priority queue. The proposed algorithm is based on a simple linear search algorithm and eliminates unnecessary arithmetic operations from distance computations between multidimensional vectors. Moreover, we propose two techniques, a dimensional sorting method and a PCA-based method, to accelerate multidimensional search. Experimental results indicate that our scheme scales well even for a very large number of dimensions. (Keyword) Nearest neighbor search / High dimensional space / Fast linear search / Priority queue / Multimedia information retrieval / High dimensional space / Fast linear search / Priority queue / Multimedia information retrieval (Link to Publication Site) ● Publication site (DOI): 10.1541/ieejeiss.126.353 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390001204606139392 ● Summary page in Scopus @ Elsevier: 2-s2.0-33644796982 (DOI: 10.1541/ieejeiss.126.353, CiNii: 1390001204606139392, Elsevier: Scopus)
69.	Satoru Tsuge, Masami Shishibori, Fuji Ren, Kenji Kita and Shingo Kuroiwa : Specific Speaker's Japanese Speech Corpus over Long and Short Time Periods, Advances in Natural Language Processing Research in Computing Science, Vol.18, 115-124, 2006. (Summary) We have been collecting the Japanese speech corpus for investigating the relationship between intra-speaker speech variability and speech recognition performance. In this paper, we introduce our speech corpus. Our corpus consists of six speakers speech data. Each speaker read specific utterance sets three times a day, once a week. Using a specific female speakers speech data in this corpus, we conduct speech recognition experiments for investigating the relationship between intra-speaker speech variability and speech recognition performance. Experimental results show that the variability ofrecognition performance over different days is larger than variability of recognition performance within a day.
70.	Shingo Kuroiwa, Satoru Tsuge, Masami Shishibori, Fuji Ren and Kenji Kita : Simple PCAを用いたベクトル空間情報検索モデルの次元削減, IEEJ Transactions on Electronics, Information and Systems, Vol.125, No.11, 1773-1779, 2005. (Summary) In this paper, we propose to use the Simple Principal Component Analysis (SPCA) for dimensionality reduction of the vector space information retrieval model. The SPCA algorithm is a data-oriented fast method which does not require the computation of the variance-covariance matrix. In SPCA, principal components are estimated iteratively so we also propose a criteria to determine the convergence. The optimum number of iterations for each principal component can be determined using the criteria. Experimentally, we show that the SPCA-based method offers improvement over the conventional SVD-based method despite its small amount of computation. This advantage of SPCA can be attributed to its iterative procedure which is similar to clustering methods such as $k$-means clustering. On the other hand, the proposed method which orthogonalizes the basis vectors also achieved much higher accuracy than the conventional random projection method based on $k$-means clustering. (Keyword) 情報検索 / ベクトル空間モデル / 次元削減 / Simple PCA / クラスタリング / Information retrieval / Vector Space Model / Dimensionality Reduction / Simple PCA / clustering (Link to Publication Site) ● Publication site (DOI): 10.1541/ieejeiss.125.1773 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390282679580608000 ● Search Scopus @ Elsevier (DOI): 10.1541/ieejeiss.125.1773 (DOI: 10.1541/ieejeiss.125.1773, CiNii: 1390282679580608000)
71.	羅辛, 邰暁英, Masami Shishibori and Kenji Kita : Image Database Retrieval Based on Relevance Feedback by Feature Space Warping, 計算机科学, Vol.32, No.7, 235-238, 2005. (Summary) In recent years, the employment of the user's feedback information to improve the image retrieval precision has become a hot subject in research field. But in the traditional relevance feedback methods, both relevant and irrelevant user's assigned information was required for the retrieval system. For the sake of practicality and convenience, this paper advances that users only need to choose their inquired image files, which generate new index vector as relevance information. Through the feature vector space warping, the index is moved towards the user's inquiry intension. Compared with the traditional relevance feedback technique, the proposed method is shown to obviously improve the retrieval precision.
72.	持橋大地, 菊井玄一郎 and Kenji Kita : 言語表現のベクトル空間モデルにおける最適な計量距離, The Transactions of the Institute of Electronics, Information and Communication Engineers D-II, Vol.J88-D-II, No.4, 747-756, 2005. (Summary) 自然言語処理の手法にはコサイン距離, すなわち単純なユークリッド距離に依存しているものが多いが, この距離は素性間の相関を考慮しないこと, また素性に恣意的な重みづけを必要とするという問題をもっている.本論文ではこれに代わり, 訓練データより導かれる最適な計量距離を提案し, 二つの問題を同時に解決する.この計量は訓練データのクラスタ構造のひずみを最小化する二次最適化問題の解として導くことができる.提案する計量距離の効果を, 同義文検索, 文書検索, 及び機械学習用一般ベクトルデータのクラスタリングタスクによって確認し, ユークリッド距離に対し常に精度の改善が見られた.特にクラスタ化の強い同義文検索タスクにおいて精度の改善が大きく, ベースラインと比較して11点平均精度で33%の向上があった. (Keyword) tf.idf / コサイン距離 / 素性空間 / カーネル法 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1520572357970628608 (CiNii: 1520572357970628608)
73.	Satoru Tsuge, Shingo Kuroiwa, Masami Shishibori and Kenji Kita : Channel Compensation Method for ETSI Standard DSR Front-end, IEEJ Transactions on Electronics, Information and Systems, Vol.125, No.1, 120-127, 2005. (Summary) 本論文では，分散型音声認識方式における入力系の周波数特性の異なりによる音声認識精度の変動を抑制する手法として，平均一致化手法を提案した．提案手法は入力特徴量の平均と圧縮に用いるVQセントロイドとの平均を一致させ，入力系の周波数特性を正規化する．提案手法の有効性を検証するためETSIが勧告している分散型音声認識フロントエンドを用いた音声認識実験を行った結果，提案手法より周波数特性の変動による音声認識精度の低下を抑制することが可能であった． (Keyword) Distributed speech recognition / ETSI standard DSR front-end / Frequency characteristic of input device / VQ distortion / Frequency characteristic of input device / VQ distortion (Link to Publication Site) ● Publication site (DOI): 10.1541/ieejeiss.125.120 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390282679581397888 ● Search Scopus @ Elsevier (DOI): 10.1541/ieejeiss.125.120 (DOI: 10.1541/ieejeiss.125.120, CiNii: 1390282679581397888)
74.	Satoru Tsuge, Masami Shishibori, Shingo Kuroiwa and Kenji Kita : Information Retrieval Using Non-negative Matrix Factorization, IEEJ Transactions on Electronics, Information and Systems, Vol.124, No.7, 1500-1506, 2004. (Summary) ベクトル空間情報検索モデルを用いた全文検索を行う場合，各文書を表現する索引語ベクトルがスパースになることが問題となる．そこで，本論文ではベクトル空間モデルの次元削減により情報検索精度を向上させる．次元削減には，非負行列である文書行列に対する次元削減手法として，Non-negative Matrix Factorization (NMF)を適用した．NMFは非負行列を2つの非負行列へ分解を行う手法である．提案手法は，NMFを用い非負行列である索引語文書行列を分解し，その一方の行列へ索引語文書行列を射影することにより，次元削減を行う手法である．MEDLINEを用いた情報検索実験で提案手法の有効性を示した． (Keyword) information retrieval / vector space model / Non-negative Matrix Factorization / dimensionality reduction (Link to Publication Site) ● Publication site (DOI): 10.1541/ieejeiss.124.1500 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390282679580788480 ● Search Scopus @ Elsevier (DOI): 10.1541/ieejeiss.124.1500 (DOI: 10.1541/ieejeiss.124.1500, CiNii: 1390282679580788480)
75.	Masami Shishibori, 小泉大地, Satoru Tsuge and Kenji Kita : Development of a WWW Image Retrieval System Using the Image Knowledge Database, The Transactions of the Institute of Electronics, Information and Communication Engineers D-I, Vol.J87-D-I, No.2, 154-163, 2004. (Summary) ユーザからのフィードバック情報を格納した画像知識データベースを用いたWWW画像検索システムを開発した．本システムでは，既存システムの結果からユーザが選択した正解画像の画像的特徴量をフィードバック情報としてデータベースに登録することで，次回の検索の際，このデータベースを教師データとして，より類似している画像を優先的に提示することができる．従来システムとしてGoogle Image Searchを用い，130個の検索キーに対する評価実験を行った結果，11.6%の精度向上が確認できた． (Keyword) WWW画像検索システム / 類似画像検索 / 両像知識データベース / World Wide Web (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1520572358455627008 (CiNii: 1520572358455627008)
76.	Kenji Kita and Masami Shishibori : Efficient Multidimensional Indexing Using One-dimensional Self-Organizing Maps, Journal of Natural Language Processing, Vol.10, No.5, 41-54, 2003. (Summary) 高次元空間における最近傍検索は，マルチメディア·コンテンツ検索，データ·マイニング，パターン認識等の分野における重要な研究課題であるが，効率的な多次元インデキシング手法は，従来，存在しなかった．本論文では，自己組織化マップを用いて，高次元空間での近傍関係をできる限り保ちつつ，高次元データを1次元空間へ配置し，1次元マップから得られる情報を用いて探索範囲を限定することにより，きわめて高速な最近傍検索手法を提案した． (Keyword) Nearest neighbor search / Multidimensional indexing / Multimedia contents retrieval / Self-organizing map (Link to Publication Site) ● Publication site (DOI): 10.5715/jnlp.10.5_41 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1520853834973292928 ● Search Scopus @ Elsevier (DOI): 10.5715/jnlp.10.5_41 (DOI: 10.5715/jnlp.10.5_41, CiNii: 1520853834973292928)
77.	Tai Xiaoying, Fuji Ren and Kenji Kita : Improvement of vector space information retrieval by using long-term relevance feedback, Asian-Information-Science-Life, Vol.2, No.1, 83-96, 2003. (Summary) This paper proposes a new method to improve information retrieval performance of the vector space model (VSM) by long term preservation of user-supplied relevance information in the system. The proposed method incorporates user's relevance feedback information and information of original documents similarities into the retrieval model that is build by using a sequence of linear transformations. The results of our experiments show that the proposed method has better retrieval performance and provides an approach that enables preservation of user-supplied information in the system.
78.	Satoru Tsuge, Masami Shishibori, Shingo Kuroiwa and Kenji Kita : Relevance Feedback Using Support Vector Machine for Information Retrieval, Transactions of Information Processing Society of Japan, Vol.44, No.1, 59-67, 2003. (Summary) 本論文では，ベクトル空間情報検索モデルを用いた全文検索精度の向上を目指した．本論文では，サポートベクターマシンを用いた適合性フィードバック手法を提案した．提案手法は，利用者からの適合·不適合の情報をもとにサポートベクターマシンを学習させ，利用者が要求する検索文書を識別し，不適合文書を利用者に提示しない手法である．日本語全文検索テストコレクションを用い，提案手法の有効性を検証する実験を行い，提案手法が全文検索に有効であることを示した． (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1050564287837317504 (CiNii: 1050564287837317504)
79.	Tai Xiaoying, Fuji Ren and Kenji Kita : An information retrieval model based on vector space method by supervised learning, Information Processing & Management, Vol.38, No.6, 749-764, 2002. (Summary) This paper proposes a method to improve retrieval performance of the vector space model in part by utilizing user-supplied information of those documents that are relevant to the query in question. In addition to the user's relevance feedback information, information such as original document similarities is incorporated into the retrieval model, which is built by using a sequence of linear transformations. High-dimensional and sparse vectors are then reduced by singular value decomposition and transformed into a low-dimensional vector space, namely the space representing the latent semantic meanings of words. The method has been tested with two test collections, the Medline collection and the Cranfield collection. (Keyword) information retrieval / document / feedback information / vector space model (Link to Publication Site) ● Publication site (DOI): 10.1016/S0306-4573(01)00053-X (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1573950401491463680 ● Search Scopus @ Elsevier (DOI): 10.1016/S0306-4573(01)00053-X (DOI: 10.1016/S0306-4573(01)00053-X, CiNii: 1573950401491463680)
80.	台暁英, Masami Shishibori, Kenji Kita and Fuji Ren : 長期的適合性フィードバックによるベクトル空間モデルの精度改善, INFORMATION, Vol.5, No.2, 255-265, 2002. (Summary) 本論文では，ユーザからのフィードバック情報を，長期的にベクトル空間モデルに基づく検索システムに残す方法を提案する．提案するモデルでは，検索精度を高めるため，ユーザからの適合性の判定結果だけではなく，文書間距離などの情報も検索モデルに組み入れ，線形変換によりベクトル空間検索モデルを構築する．その後，ベクトル空間モデルの高次元行列を特異値分解により縮退し，低次元のベクトル空間，つまり索引語と文書の潜在的な意味構造を表した意味空間に変換する．提案手法により，実験システムを構築し，二つのテストコレクションを用いて，評価実験を行った．本論文で提案した手法の有効性を確かめることができた． (Keyword) ベクトル空間モデル / 文書間距離 / フィードバック情報 / 高次元行列
81.	Xiaoying Tai, Minoru Fukumi and Kenji Kita : 教師あり学習によるベクトル空間情報検索モデルの精度改善, IEEJ Transactions on Electronics, Information and Systems, Vol.121-C, No.10, 1647-1653, 2001. (Summary) 本論文では，教師あり学習によるベクトル空間情報検索モデルの精度改善手法について述べた．提案手法を用いて，線形変換により検索質問集合に対する適合性情報および文書間距離情報を検索モデルに組み入れることにより，ユーザからのフィードバック情報を長期的に検索システムに反映させることが可能となった．
82.	Minoru Sasaki and Kenji Kita : Analysis of the Characteristics and the Efficiency of Information Retrieval Systems by Statistical Method, Journal of Natural Language Processing, Vol.8, No.1, 85-99, 2001. (Summary) 本論文では，情報検索システムの平均適合率，再現率·適合率曲線を直線回帰させた傾きと切片が，システムで用いられている手法とどのような相関関係を持っているかを調査し，それぞれの手法がシステムの性能に与える大きさを統計的観点から分析した． (Keyword) Information Retrieval / IREX Workshop / Evaluation / Precision / Recall (Link to Publication Site) ● Publication site (DOI): 10.5715/jnlp.8.85 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390001204476116352 ● Search Scopus @ Elsevier (DOI): 10.5715/jnlp.8.85 (DOI: 10.5715/jnlp.8.85, CiNii: 1390001204476116352)
83.	Minoru Sasaki and Kenji Kita : Dimensionality Reduction of Vector Space Information Retrieval Model Based on Random Projection, Journal of Natural Language Processing, Vol.8, No.1, 5-19, 2001. (Summary) ベクトル空間モデルは，情報検索における代表的な検索モデルである．ベクトル空間モデルでは，文書を索引語の重みベクトルで表現するが，文書ベクトルは一般に要素数が非常に多く，スパースなベクトルになるため，検索時間の長さや必要なメモリ量が大きな問題となる．本論文では，この問題を解決するために，文書ベクトルの次元圧縮を行う手法としてランダム·プロジェクションを用いた検索モデルを提案した． (Keyword) Information retrieval / Vector space model / Random projection / Latent semantic indexing / Concept vector (Link to Publication Site) ● Publication site (DOI): 10.5715/jnlp.8.5 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390282679452831488 ● Search Scopus @ Elsevier (DOI): 10.5715/jnlp.8.5 (DOI: 10.5715/jnlp.8.5, CiNii: 1390282679452831488)
84.	Hiroki Oda and Kenji Kita : PPM^言語モデルを用いた日本語単語分割, Transactions of Information Processing Society of Japan, Vol.41, No.3, 689-700, 2000. (Summary) 日本語の単語分割の精度を向上させるために，データ圧縮アルゴリズムPPM (Prediction by Partial Matching)を用いた言語モデルを適用した単語分割手法を提案した．PPM*は，確率·統計的テキスト圧縮技報として優れており，無限長文脈を取り扱うことができるという利点を持っているため，通常の文字n-gramモデルよりも頑健な単語分割を行うことができるようになった．
85.	Hiroki Oda, Shinsuke Mori and Kenji Kita : 文字クラスモデルによる日本語単語分割, Journal of Natural Language Processing, Vol.6, No.7, 93-108, 1999. (Summary) 日本語処理において，単語の同定，すなわち文の単語分割は最も基本的かつ重要な処理である．本論文では，日本語文字のクラス分類により得られた文字クラスの確率モデルを用いた新しい単語分割手法を提案した．また，対話文データベースを用いた評価実験の結果，文字クラスモデルは単純な文字連鎖の確率モデルよりも，高い精度を達成できることを示した．
86.	Kenji Kita : Automatic Clustering of Languages Based on Probabilistic Models, Journal of Quantitative Linguistics, Vol.6, No.2, 167-171, 1999. (Summary) This paper proposes a novel method for automatically clustering languages. The basic idea of this method involves debveloping a probabilistic model for each language from the given linguistic data, and then computing the distance between language models according to the distance measure defined on the langugae models. The effectiveness of the proposed method has been confirmed by reconstructing language trees using multilingual texts from the ECI/MCI corpus.
87.	Hiroki Oda and Kenji Kita : 単語の位置情報に基づくコーパスからのコロケーションの自動抽出, Journal of Natural Language Processing, Vol.5, No.1, 79-99, 1998. (Summary) コーパス中の単語位置情報に基づき，連語表現(コロケーション)を自動抽出する手法を提案した．提案した手法では，コーパス中に各単語の位置情報を用いて，任意の文中のコロケーションを連続型・不連続型の別に抽出する．実験の結果，連続型・不連続型コロケーションともに重要な表現が抽出されており，提案した手法の有効性を示すことができた．
88.	Kenji Kita, Yoshikazu Fukui, Masaaki Nagata and Tsuyoshi Morimoto : 発話タイプ付きコーパスを用いた確率的対話モデルの自動生成, Journal of Natural Language Processing, Vol.4, No.4, 73-85, 1997. (Summary) 高次の言語情報である対話に対する確率的モデルを，エルゴード HMM および確率オートマトンを用いて，コーパスから自動的に生成するための方法について述べた．これらの確率的対話モデルを用いることにより，話者の交替や質問・応答・確認といった会話の基本的な構造を確率・統計的な観点からモデル化することができた．また，得られた確率的対話モデルを，情報理論的な観点から評価した．
89.	Kenji Kita : 確率的言語モデルに基づく多言語コーパスからの言語系統樹の再構築, Journal of Natural Language Processing, Vol.4, No.3, 71-82, 1997. (Summary) 各言語の言語データから確率的言語モデルを構築し，次に確率的言語モデルの間に導入した距離に基づき，元の言語に対するクラスタリングを実行する方法を提案した．提案した方法を用いることにより，言語系統樹の構築を自動的に行うことができることを示した．
90.	Kenji Kita and Hiroaki Ogata : Collocations in Language Learning: Corpus-based Automatic Compilation of Collocations and Bilingual Collocation Concordancer, Journal of Computer Assisted Language Learning, Vol.10, No.3, 229-238, 1997. (Summary) In this paper, we are primarily concerned with the acquisition of collocational knowledge from corpora to use it in the language learning process. We present an efficient method from extracting collocations from corpora which uses the cost criteria measure and a tree-based data structure. We also propose a bilingual collocation concordancer, a tool that provides language learners with collocation correspondences between a native language and a foreign language. (Link to Publication Site) ● Publication site (DOI): 10.1080/0958822970100303 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1080/0958822970100303 (DOI: 10.1080/0958822970100303)
91.	Kenji Kita : Mixture Probabilistic Context-Free Grammar: An Improvement of a Probabilistic Context-Free Grammar Using Cluster-Based Language Modeling, Journal of Natural Language Processing, Vol.3, No.4, 103-113, 1996. (Summary) This paper proposes an improved probabilistic CFG, called mixture probabilistic CFG, based on an idea of cluster-based language modeling. This model assumes that the language model parameters have different probability distributions in different topics or domains. The basic idea of this model involves clustering a training corpus into a number of subcorpora, and then training probabilistic CFGs from these subcorpora. The resulting probabilistic CFGs become context- or topic-dependent, and thus accurate language modeling would be possible.
92.	Kenji Kita, Yoneo Yano and Tsuyoshi Morimoto : One-Pass Search Algorithm for Continuous Speech Recognition Using Generalized LR Parsing: A CFG-Driven, Frame-Synchronous HMM-Based Approach, Transactions of Information Processing Society of Japan, Vol.36, No.5, 1252-1259, 1995. (Summary) This paper presents a novel continuous speech recognition algorithm that integrates three major technologies: (1) hidden Markov modeling for speech, (2) a generalized LR parser for handling context-free grammar (CFG) constraints, and (3) the one-pass search algorithm for achieving an efficient search. We also introduce three techniques that used in the development of the algorithm: (1) LR path merging, (2) the use of a shared tree-structured stack, and (3) LR-parser-based dynamic network generation.
93.	Kenji Kita, Yasuhiko Kato, Takashi Omoto and Yoneo Yano : A Comparative Study of Automatic Extraction of Collocations from Corpora: Mutual Information vs. Cost Criteria, Journal of Natural Language Processing, Vol.1, No.1, 21-33, 1994. (Summary) This paper is primarily concerned with the acquisition of collocational knowledge from corpora for use in language learning. First we discuss the importance of collocational knowledge in second langugae learning, and then take up two measures, mutual information and cost criteria, for automatically identifying or extracting collocations from corpora. (Keyword) corpus
94.	永井明人, Kenji Kita and 他9名 : HMMと一般化LR構文解析法を用いた実時間大語彙連続音声認識装置の実現, The Journal of the Acoustical Society of Japan, Vol.50, No.9, 723-729, 1994. (Summary) 大語彙の連続音声認識を実時間で処理するための，HMM-LR 連続音声認識装置の設計，開発及び処理性能について述べた．処理量が極めて大きな継続時間制御付きトレリス計算を伴うHMM-LR連続音声認識を実時間で実行するために，本装置は種々の高速化手法を33個のDSPを用いて実現した．その結果，連続文節発声の入力音声に対し，入力文の長さに依らずに，発声終了後から2∼3秒ですべての認識処理を終了する処理速度を達成した．
95.	Kenji Kita, Tsuyoshi Morimoto, Kazumi Ohkura, Shigeki Sagayama and Yoneo Yano : Spoken Sentence Recognition Based on HMM-LR with Hybrid Language Modeling, IEICE Transactions on Information and Systems, Vol.E77-D, No.2, 258-265, 1994. (Summary) This paper describes Japanese spoken sentence recognition using hybrid language modeling, which combines the advantages of both syntactical and stochastic language models. As the baseline system, we adopted the HMM-LR speech recognition system, with which we have already achieved good performance for Japanese phrase recognition tasks. Several improvements have been made to this system aimed at handling continuously spoken sentences, including (1) HMM training with continuous utterances as well as word utterances, (2) development of a sentential grammar, and (3) introduction of stochastic langugae models.
96.	Kenji Kita, 小倉健太郎, 森元逞 and Yoneo Yano : 仕事量基準を用いたコーパスからの定型表現の自動抽出, Transactions of Information Processing Society of Japan, Vol.34, No.9, 1937-1943, 1993. (Summary) 自然言語において頻繁に使用される定型的な表現をコーパスから自動的に抽出する基準として仕事量という概念を提案した．仕事量は，いくつかの単語を一まとまりの単位と考えることにより，各単語を別個に処理するよりも，どれだけの処理が削減できるかということを定量的に測る尺度である．また，仕事量基準を用いた定型表現の抽出方法について述べ，提案した方法が有効であることを実験的に示した．
97.	Akito Nagai, Shigeki Sagayama, Kenji Kita and Hideaki Kikuchi : Three Different LR Parsing Algorithms for Phoneme-Context-Dependent HMM-Based Continuous Speech Recognition, IEICE Transactions on Information and Systems, Vol.E76-D, No.1, 29-37, 1993. (Summary) This paper discusses three approaches for combining an efficient LR parser and phoneme-context-dependent HMMs and compare them through continuous speech recognition experiments. This paper also includes discussion of the results of recognition experiments, and a comparison and efficiency of the three algorithms.
98.	Kenji Kita, Tsuyoshi Morimoto and Shigeki Sagayama : LR Parsing with a Category Reachability Test Applied to Speech Recognition, IEICE Transactions on Information and Systems, Vol.E76-D, No.1, 23-28, 1993. (Summary) This paper proposes an extended LR parsing algorithm, called LR parsing with a category reachability test (the LR-CRT algorithm). The LR-CRT algorithm enables a parser to efficiently recognize those sentences that belong to a specified grammatical category. We apply the LR-CRT algorithm to improve a speech recognition system based on two-level LR parsing. The system was evaluated using sentential speech data uttered by phrase by phrase, and attained a word accuracy of 97.5% and a sentence accuracy of 91.2%.
99.	Kenji Kita, Terumasa Ehara and Tsuyoshi Morimoto : Processing Unknown Words in Continuous Speech Recognition, IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, Vol.E74, No.7, 1811-1816, 1991. (Summary) This paper proposes a continuous speech recognition method which accepts any utterance that might include unknown words. In this method, words not in the lexicon are transcribed as phoneme sequences, while words in the lexicon are recognized correctly.
100.	Kenji Kita, Toshiyuki Takezawa and Tsuyoshi Morimoto : Continuous Speech Recognition Using Two-Level LR Parsing, IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, Vol.E74, No.7, 1806-1810, 1991. (Summary) This paper describes a continuous speech recognition system using two-level predictive LR parsing. In Japanese, a sentence is composed of several phrases, thus two kinds of grammars, namely an intra-phrase grammar and an inter-phrase grammar are sufficient for recognizing sentences. Two-level predictive LR parsing makes it possible to use not only intra-phrase grammatical constraints but also inter-phrase grammatical constraints during speech recognition. The system is applied to Japanese sentence recognition where sentence were uttered phrase by phrase, and attains a word accuracy of 95.9% and a sentence accuracy of 84.7%. (Keyword) speech recognition
101.	Kenji Kita, Takeshi Kawabata and Toshiyuki Hanazawa : HMM Speech Recognition Using Stochastic Language Models, The Journal of the Acoustical Society of Japan, Vol.12, No.3, 99-105, 1991. (Summary) In this paper, the following three stochastic language models are investigated for an HMM continuous speech recognition system. (1) Trigram model of Japanese syllables, (2) Stochastic shift/reduce model in LR parsing, and (3) Trigram model of context-free rewriting rules. (Keyword) speech recognition
102.	花沢利行, Kenji Kita, 中村哲, 川端豪 and 鹿野清宏 : HMM-LR音声認識システムの性能評価, The Journal of the Acoustical Society of Japan, Vol.46, No.10, 817-823, 1990. (Summary) HMM 音韻認識と拡張 LR 構文解析法を統合化した HMM-LR 音声認識システムにセパレートベクトル量子化，HMM の継続時間長制御，VQ コードブック·マッピングに基づく話者適応を組み入れることによりシステムの改良を行った．日本語文節認識実験で評価を行い，システムの有効性を示した． (Keyword) 音声認識
103.	Kenji Kita, 川端豪 and 斎藤博昭 : HMM音韻認識と拡張LR構文解析法を用いた連続音声認識, Transactions of Information Processing Society of Japan, Vol.31, No.3, 472-480, 1990. (Summary) 高精度の音声認識を実現するためには，言語的な情報の利用が不可欠である．本論文では，文脈自由文法に基づく統語的(文法的)な制約を音声認識時に用いる方法について検討し，拡張 LR 構文解析法と HMM (Hidden Markov Model)を用いた音声認識手法を提案した．提案した手法では，音声認識と言語処理の間に音韻ラティス等の中間的なデータを介さないため，高精度のかつ効率的な認識処理系を構成することができるという特徴を持っている．
104.	Kenji Kita : 直観主義的解析学の諸原理, 科学基礎論研究, Vol.16, No.1,2, 69-75, 1983. (Summary) 本論文は，直観主義論理的論理に基づく解析学である直観主義的解析学での公理の意味を明らかかにすると共に，直観主義的解析学の公理系から，古典的解析学では成立しえない定理をいくつか導き，これらの定理の意味論的立場での解釈について論じた．

Academic Paper (Unrefereed Paper):

1.	Masahiko Sano and Kenji Kita : 徳島大学キャンパスネットワークシステムの更新について, Academic Information Processing Environment Research, 94-102, 2010. (Summary) 徳島大学で平成22年3月に行われたキャンパスネットワークの更新について述べた．ネットワークのさまざまな問題点を調査し，可用性向上，帯域拡大，高機能化，低運用コスト化等の課題について検討し，更新を行った． (Keyword) キャンパスネットワーク更新 / キャンパスネットワークコウシン / ネットワークの可用性 / ネットワークノカヨウセイ / 運用コスト削減 / ウンヨウコストサクゲン / Update of campus network / update of campus network / Availability of network / availability of network / Operation cost reduction / operation cost reduction (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1521417756041793408 (CiNii: 1521417756041793408)
2.	Xin Luo, Masami Shishibori and Kenji Kita : Distance Metric Learning for Content-based Image Retrieval, 広西師範大学学報, Vol.25, No.2, 186-189, 2007. (Summary) The performance of a content-based image retrieval system is inherently constrained by the low-level features adopted to represent the images in the databases. Machine learning and statistical modeling approach is a simple yet effective technique for image retrieval, but the retrieval performance is often dependent on a large number of training data. The method presents an optimal metric distance function based on a simple cluster and a little training data by learning a distance metric that respects these relationships. Experiments show that the learned metrics can be used to significantly improve image retrieval performance.

Academic Letter:

1.	Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Analyzing Transformers via Value Matrices, Transactions of the Japanese Society for Artificial Intelligence, Vol.38, No.2, 2023. (Keyword) Transformer / value matrix / word embedding (Link to Publication Site) ● Publication site (DOI): 10.1527/tjsai.38-2_C-MB7 (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1390013795251121280 ● Search Scopus @ Elsevier (DOI): 10.1527/tjsai.38-2_C-MB7 (DOI: 10.1527/tjsai.38-2_C-MB7, CiNii: 1390013795251121280)
2.	Masami Shishibori, 山本一晴, 小泉大地 and Kenji Kita : URLの類似性に着目したWWW空間からの関連語自動収集手法, 日本データベース学会Letters, Vol.6, No.1, 57-60, 2007. (Summary) 本論文では，シーズとなる単語群(基底単語群)の関連語をWWW空間から効率的に自動収集する手法を提案する．基底単語群が存在するページのURL集合と関連語が存在するURL集合間に類似性があると予測される．そこで本手法では，基底単語群を既存のWWW検索システムに入力することで得られる検索結果のURLをデータベース化し，基底単語群のURLデータベースと類似性を計算することで，関連語を効率的に収集する． (Link to Search Site for Scientific Articles) ● CiNii @ National Institute of Informatics (CRID): 1520290884798639104 (CiNii: 1520290884798639104)

Review, Commentary:

1.	Fujisawa Akira, Kazuyuki Matsumoto, Okumura Noriyuki, Minoru Yoshida and Kenji Kita : Challenge to ASCII Art - An Image Feature-Based Approach -, Journal of Japanese Society for Artificial Intelligence, Vol.32, No.3, 364-370, May 2017. (Keyword) ASCII arat / image feature / non-verbal expression / emoticon / histograms of oriented gradients (Tokushima University Institutional Repository) ● Metadata: 113912 (Tokushima University Institutional Repository: 113912)
2.	Kenji Kita and 山川義介 : 類似画像検索技術を用いた商品推薦システム『SUDACHI』の開発とその応用, Image Lab, Vol.19, No.8, 11-15, Aug. 2008.
3.	Kenji Kita and Mikio Yamamoto : 確率·統計情報を用いた文法研究, 日本語学, Vol.19, No.4, 202-212, Apr. 2000. (Summary) 言語データ(コーパス)を重視する経験主義的な方法論の有効性と，言語データをもとにした文法の研究について，確率文脈自由文法と確率木置換文法を例にとり説明した．
4.	Kenji Kita and Mikio Yamamoto : 確率·統計的言語モデル, Computer Today, No.3, 28-33, Mar. 2000. (Summary) 確率·統計的言語モデルの基本的な考え方，および，応用例として統計的機械翻訳などについて述べた．
5.	Kenji Kita, 中村哲 and 永田昌明 : 確率·統計モデルの音声言語処理への応用, Journal of Japanese Society for Artificial Intelligence, Vol.10, No.2, 25-32, Mar. 1995. (Summary) 音声言語の確率·統計的モデルが，実際の音声認識や自然言語処理でどのように用いられているのかを中心に説明した．また，確率文脈自由文法の弱点である文脈情報や語彙情報の記述能力の不足を解消する試みとして，履歴に基づく文法や決定木パーザなどの新しい構文解析モデルについても解説した．
6.	中村哲, Kenji Kita and 永田昌明 : 音声言語の確率モデル, Journal of Japanese Society for Artificial Intelligence, Vol.10, No.2, 17-24, Mar. 1995. (Summary) 音声言語処理におけるマルコフモデルを基本とした確率モデル，および，最尤推定法やEMアルゴリズムを基本としたモデルのパラメータ推定法について述べた．

Proceeding of International Conference:

1.	Shohei Noguchi, Kazuyuki Matsumoto, Minoru Yoshida, Ryota Nishimura and Kenji Kita : Extraction of keywords from disease-fighting blogs by fine-tuning the T5 model, 2022 6th International Conference on Natural Language Processing and Information Retrieval, Dec. 2022. (Summary) To measure lifestyle-related diseases such as heart disease and diabetes, which are on the increase, information on patients' lifestyles is necessary. With the growth of the Internet culture in recent years, a great deal of valuable information about patients' long fight against the disease, including lifestyle and condition changes, is now being written in the form of blogs and tweets. In this study, keywords related to changes in health status are extracted from "disease-fighting blogs" using Named Entity Recognition system, which is fine-tuned the "T5 Model", one of the latest deep learning and natural language processing models. This method can be applied to a system that informs patients and health care professionals how their lifestylehabits affect their health status. (Keyword) disease-fighting blog / named entity recognition / T5
2.	Daiki Mori, Kazuyuki Matsumoto, Manabu Sasayama, Minoru Yoshida and Kenji Kita : Emotion and personality estimation of utterances in a dialogue model reflecting personality, Proceedings of 4th International Conference on Technological Competency as Caring in Nursing and Health Sciences 2022, 105, Online, Dec. 2022. (Summary) One of the factors that enable humans to interact with AI in a friendlier manner is personality, and since the personality required of AI varies depending on the situation and context, a mechanism is needed that allows AI to switch between different personalities depending on the person with whom he or she is interacting. In this study, we use a Japanese language Transformer Encoder-decoder dialogue model (Japanese-dialog-transformers) and aim to construct a dialogue model that reflects the personality of the target person by fine-tuning the interview dialogue data. By analyzing the output of the proposed dialogue models using the MBTI estimation model and the sentiment analysis model, we evaluate the consistency of the personality and discuss the differences and similarities among the dialogue models. (Keyword) natural language processing / dialogue model / artificial intelligence / deep learning
3.	Reishi Amitani, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Affective Analysis and Visualization from Posted Text, Replies, and Images for Analysis of Buzz Factors, Frontiers in Artificial Intelligence and Applications (FAIA), Fuzzy Systems and Data Mining VIII, Vol.358, 191-203, Nov. 2022. (Summary) In this study, we propose a method to visualize the factors that contribute to the buzz phenomenon triggered by Twitter posts. The analysis included tweets, images, and replies. Replies are after-the-fact responses posted in response to a posted tweet and therefore cannot be used to predict buzz phenomena. Therefore, they cannot be used to predict the buzz phenomena. In this study, the tweet body, images, and reply text were feature vectors, and an affective analysis model was constructed. Visualization of the relationship between the sensibility features output from this model and the number of RTs and likes (echo index), which represent the scale of the buzz, will be useful for analyzing the factors behind the popularity. Consequently, the subjective sensibility information with the most likes also tended to have a higher degree of similarity among the sensibility vectors. (Keyword) Tweet analysis / natural language processing / trend analysis (Link to Publication Site) ● Publication site (DOI): 10.3233/FAIA220384 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85142130952 (DOI: 10.3233/FAIA220384, Elsevier: Scopus)
4.	Motonobu Yoshida, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : A System to Correct Toxic Expression with BERT, Proceedings of the 14th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management - KEOD, 92-97, Valletta, Oct. 2022. (Summary) This paper describes a system for converting posts with toxic expression on social media, such as those containing slander and libel, into less-toxic sentences. In recent years, the number of social media users as well as the cases of online flame wars has been increasing. Therefore, to prevent flaming, we first use a prediction model based on Bidirectional Encoder Representations from Transformers (BERT) to determine whether a sentence is likely to be flamed before it is posted. The highest classification accuracy recorded 82% with the Japanese Spoken Language Field Adaptive BERT Model (Japanese Spoken BERT model) as a pre-trained model. Then, for sentences that are judged to be toxic, we propose a system that uses BERTs masked word prediction to convert toxic expressions into safe expressions, thereby converting them into sentences with mitigated aggression. In addition, the BERTScore is used to quantify whether the meaning of the converted sentence has changed in meaning compared to the original sentence and evaluate whether the modified sentence is safe while preserving the meaning of the original sentence. (Keyword) Toxic Expression / BERT / classification / Text Crrection / Flame War (Link to Publication Site) ● Publication site (DOI): 10.5220/0011586100003335 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85146200806 (DOI: 10.5220/0011586100003335, Elsevier: Scopus)
5.	Daichi Kawahara, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Music Video Search System Based on Comment Data and Lyrics, Advances in Natural Computation, Fuzzy Systems and Knowledge Discovery(Lecture Notes on Data Engineering and Communications Technologies; LNDECT), Vol.153, 1125-1135, Jul. 2022. (Summary) Many people currently use video-sharing services such as YouTube. Keyword search is prevalent in these services. It can be hard to find a video that matches the users' interests using keyword search unless appropriate words are used. In this study, we propose a method for retrieving music videos with similar impressions by analyzing comment data from YouTube viewers and music lyrics. The proposed method converts comments and lyrics into vectors using Word2Vec, and music videos with similar impressions are retrieved using fuzzy c-means clustering According to the mean reciprocal rank (MRR) scores, it was clear that the output of the music videos had the same impression within the top three songs. (Keyword) artificial intelligence / music video search / lyric / fuzzy c-means (Link to Publication Site) ● Publication site (DOI): 10.1007/978-3-031-20738-9_122 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85147845007 (DOI: 10.1007/978-3-031-20738-9_122, Elsevier: Scopus)
6.	Naomichi Tabuchi, Kazuyuki Matsumoto, Minoru Yoshida, Ryota Nishimura and Kenji Kita : Analysis of Lifestyle Habits from Weblogs of Patients with Intractable Diseases Using Deep Learning, Proceedings of International Conference on Electrical, Computer and Energy Technologies (ICECET), Jul. 2022. (Summary) In recent years, patient-centered medicine has become widely accepted. To keep a record of their struggles,patients sometimes publish their diaries (disease blogs) on weblogs to be read by an unspecified number of people. In particular, writing a blog may be helpful in the case of intractable diseases of which there are few cases, and patients often feel isolated because they know few people with the same disease. In this study, we used deep learning technology to analyze the blogs written by patients with intractable diseasesand to determine how they live their daily lives and their feelings. If their lifestyles are understood, the relationship between the physical condition and mental changes of people fighting intractable diseases can be determined, and a system that shows behavioral guidelines for new users fighting similar intractable diseases can be built. (Keyword) lifestyle habits / intractable disease / deep learning (Link to Publication Site) ● Publication site (DOI): 10.1109/ICECET55527.2022.9872887 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85138876019 (DOI: 10.1109/ICECET55527.2022.9872887, Elsevier: Scopus)
7.	Naomichi Tabuchi, Kazuyuki Matsumoto, Minoru Yoshida, Ryota Nishimura and Kenji Kita : Risk Analysis of Developing Lifestyle-related Diseases based on the Content of Social Networking Service Posts, Innovation in Medicine and Healthcare, Part of the Smart Innovation, Systems and Technologies book series, Vol.308, 63-72, Jun. 2022. (Summary) Japans leading causes of death are cancer, cardiovascular disease, and diabetes, which account for approximately 60% of all deaths. All these diseases are closely related to lifestyle habits such as diet, exercise, and sleep and are called lifestyle-related diseases. Lifestyle-related diseases are characterized by a lack of subjective symptoms, making it difficult to notice the deterioration of the disease. In this study, we analyzed tweets related to diet, exercise, physical condition, and mental state from the contents posted by users on social networking services (SNS) to construct a system that can prevent lifestyle-related diseases more simply. We found that the results of this work can lead to the construction of a system than can detect risk factors leading to the development of lifestyle-related diseases, from the contents of users posts. (Keyword) risk analysis / social networking service / lifestyhle-related diseases (Link to Publication Site) ● Publication site (DOI): 10.1007/978-981-19-3440-7_6 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85135061563 (DOI: 10.1007/978-981-19-3440-7_6, Elsevier: Scopus)
8.	Taiga Kirihara, Kazuyuki Matsumoto, Manabu Sasayama, Minoru Yoshida and Kenji Kita : Topic Segmentation for Interview Dialogue System, Proceedings of 2021 5th International Conference on Natural Language Processing and Information Retrieval, Online, Dec. 2021. (Summary) In this study, topic segmentation was performed by referring to the interview dialogue corpus. Utterance intention tags were added to the existing interview dialogue corpus, and uttered sentences were vectorized using BERT, Sentence BERT, and Distil BERT. In addition, topic classification was performed using the utterance intention tags and the features of the preceding and following uttered sentences. Consequently, the greatest accuracy was achieved when the utterance intention tag was used with DistilBERT (Keyword) interview dialogue / neural networks / speech intention / DistilBERT (Link to Publication Site) ● Publication site (DOI): 10.1145/3508230.3508237 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1145/3508230.3508237 (DOI: 10.1145/3508230.3508237)
9.	Reishi Amitani, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Prediction of Number of Likes and Retweets based on the Features of Tweet Text and Images, Proceedings of 2021 5th International Conference on Natural Language Processing and Information Retrieval, Online, Dec. 2021. (Summary) The current study aimed to investigate social media trends and propose an analysis method to explore the factors underpinning the buzz phenomenon on Twitter. As it is not always possible to determine the cause of the buzz phenomenon from the text content alone posted on Twitter, we limited the analysis to tweets with attached images and devised an analysis method using both text and images. We investigated whether there is a relationshipbetween the features of both tweet text and its attached images, and how the relationship between these features is related to the number of likes and retweets (RTs) receivedthat is, indicators of popularity. We trained a multi-task neural network that takes the features extracted from the images and text as input, and then outputs the number of likes and RTs before extracting the feature vectors of the same dimension from the two inputs (images and text, respectively) from the middle layer. By calculating the distance between these feature vectors, we analyzed the relationship between the number of likes and RTs. The results revealed that the average vectors of BERT and inceptionresnetv2 served as predictors of the number of likes and RTs. We also found that tweet text with a low number of likes and RTs was short and simple. (Keyword) Twitter / multi-task learning / neural networks / trend ananlysis (Link to Publication Site) ● Publication site (DOI): 10.1145/3508230.3508244 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1145/3508230.3508244 (DOI: 10.1145/3508230.3508244)
10.	Ryota Kishima, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Construction of MBTI Personality Estimation Model Considering Emotional Information, Proceedings of the 35th Pacific Asia Conference on Language, Information and Computation (PACLIC35), 262-269, Shanghai, Nov. 2021. (Summary) The widespread use of social networking services (SNSs) such as Twitter and Facebook in recent years have rendered it easy to interact with various people all over the world.Analysis of the user personality and emotion using features such as the text information and image information on SNSs has been researched.This study analyzes the effectiveness of using emotional information as a feature when building a personality estimation model for Twitter users. The extensively used Myers-Briggs type indicator (MBTI) is used for analysis. A method is proposed for estimating the MBTI personality pattern by extracting emotional information from Twitter and adding it to the characteristics of the distributed expressions of sentences and words.Through evaluation experiments, it is determined whether the emotional information is effective in constructing the MBTI personality estimation model. The obtained results demonstrate that the accuracy is improved when emotional information is included as a feature, compared to the case without emotional information. (Keyword) MBTI / personality analysis / emotion / Twitter
11.	Kazuyuki Matsumoto, Matsunaga Takumi, Minoru Yoshida and Kenji Kita : Construction of Emotional Similarity Word Embedding Model for Sentiment Analysis, 20th Mexican International Conference on Artificial Intelligence (MICAI2021), Oct. 2021. (Summary) We propose a method for constructing a dictionary of emotional expressions, which is an indispensable language resource for sentiment analysis in the Japanese. Furthermore, we propose a method for constructing a language model that reproduces emotional similarity between words, which to date has yet not been considered in conventional dictionaries and language models. In the proposed method, we pre-trained sentiment labels for the distributed representations of words. An intermediate feature vector was obtained from the pre-trained model. By learning an additional semantic label on this feature vector, we can construct an emotional semantic language model that embeds both emotion and semantics. To confirm the effectiveness of the proposed method, we conducted a simple experiment to retrieve similar emotional words using the constructed model. The results of this experiment showed that the proposed method can retrieve similar emotional words with higher accuracy than the conventional word-embedding model. (Keyword) affective computing / emotional embedding / DistilBERT / deep neural networks / emotion
12.	Kishima Ryota, Kazuyuki Matsumoto, Hirobayashi Tomoki, Seiji Tsuchiya, Minoru Yoshida and Kenji Kita : The Relationship between Personality Patterns and Harmfulness: Analysis and Prediction based on Sentence Embedding, 3rd Technological Competency as Caring in the Health Sciences 2021, Aug. 2021. (Summary) In this paper, we propose a method for detecting harmful contents in social media. In the Social Networking Service (SNS) the articles and messages posted on these services can be viewed by an unspecified number of unknown people from worldwide. Recently, increasing the number of users who have been forced to close their SNS accounts or have been suffering from mental problems because of blew up SNS triggered by hate speech or bullying remarks in social media. Even if the management of the SNS takes countermeasures, it is difficult to deal with all types of harmful expressions. In this research, we focus on the characteristic that harmful statements need to be judged from the entire sentence, and extract features of harmful expressions using a general-purpose language model. (Keyword) harmful expression / sentence embedding / Internet flaming / MBTI / personality analysis
13.	Kazuyuki Matsumoto, Ryu Mopuaa, Minoru Yoshida and Kenji Kita : Lifestyle Analysis via a Corpus of Disease-Fighting Weblogs, Proceedings of 2021 IEEE 2nd International Conference on Pattern Recognition and Machine Learning (PRML 2021), 266-272, Chengdu, Jul. 2021. (Summary) In recent years, the population of pre-diabetics in Japan has been increasing year by year. Type 2 diabetes is a type of lifestyle-related disease that can be prevented to a certain extent by correcting lifestyle habits. However, by the time we realize that there is a problem with our lifestyle, it may be too late. Therefore, early detection of risk factors for lifestyle-related diseases is important. In this study, we collected the blogs of lifestyle-related disease fighters, set up multiple keyword categories that are considered to be related to risk factors, and constructed a corpus of disease fighting blogs with labels for each category. The results of the evaluation experiments show that our proposed method can be applied to a wide range of topics. As a result of evaluation experiments, our proposed method achieves categorization of keywords and sentences with higher accuracy than the simple method. (Keyword) diabetic patients / lifestyle-related disease / category classification / machine learning (Link to Publication Site) ● Publication site (DOI): 10.1109/PRML52754.2021.9520697 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1109/PRML52754.2021.9520697 (DOI: 10.1109/PRML52754.2021.9520697)
14.	Minoru Yoshida, Kohno Shogo, Kazuyuki Matsumoto and Kenji Kita : Visualization of the Artist Relations Using Twitter User Profiles, Frontiers in Artificial Intelligence and Applications (FAIA), Fuzzy Systems and Data Mining VI,, Vol.331, 682-689, Nov. 2020. (Link to Publication Site) ● Publication site (DOI): 10.3233/FAIA200746 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85101594489 (DOI: 10.3233/FAIA200746, Elsevier: Scopus)
15.	Fujisawa Akira, Kazuyuki Matsumoto, Kazuki Ohta, Minoru Yoshida and Kenji Kita : ASCII Art Classification Model by Transfer Learning and Data Augmentation, Frontiers in Artificial Intelligence and Applications (FAIA), Fuzzy Systems and Data Mining VI, Vol.331, 608-618, Nov. 2020. (Summary) In this study, we propose an ASCII art category classification method based on transfer learning and data augmentation. ASCII art is a form of nonverbal expression that visually expresses emotions and intentions. While there are similar expressions such as emoticons and pictograms, most are either represented by a single character or are embedded in the statement as an inline expression. ASCII art is expressed in various styles, including dot art illustration and line art illustration. Basically, ASCII art can represent almost any object, and therefore the category of ASCII art is very diverse. Many existing image classification algorithms use color information; however, since most ASCII art is written in character sets, there is no color information available for categorization. We created an ASCII art category classifier using the grayscale edge image and the ASCII art image transformed from the image as a training image set. We also usedVGG16, ResNet-50, Inception v3, and Xceptions pre-trained networks to fine tune our categorization. As a result of the experiment of fine tuning by VGG16and data augmentation, an accuracy rate of 80% or more was obtained in the human category. (Keyword) Text art / ASCII art / transfer learning / fine tuning / data augmentation (Tokushima University Institutional Repository) ● Metadata: 115914 (Link to Publication Site) ● Publication site (DOI): 10.3233/FAIA200738 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.3233/FAIA200738 (Tokushima University Institutional Repository: 115914, DOI: 10.3233/FAIA200738)
16.	Kirihara Taiga, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Keyword Extraction from TV Program Viewers Tweet Based on Neural Embedding Model, Frontiers in Artificial Intelligence and Applications (FAIA), Fuzzy Systems and Data Mining VI, Vol.331, 360-369, Nov. 2020. (Summary) In recent years, young people have not been watching television (TV) as much as they used to. This is mainly because a number of TV programs are very long and/or have limited viewing times. Recently, individuals have been actively posting live-action tweets on Twitter to comment on TV content while watching programs in real time. In this study, we propose a method for extracting key phrases related to the event scenes of TV programs using live tweets, and we propose a scene search system that aims at efficient TV program viewing. The experimental results indicated that the program contents were estimated with an error of approximately 5% to 10% with respect to the program time. In addition, the extracted key phrases were visualized for each event scene category using the t-SNE algorithm. (Keyword) natural language processing / TV Program / key phrase extraction / Twitter analysis (Link to Publication Site) ● Publication site (DOI): 10.3233/FAIA200714 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85101584659 (DOI: 10.3233/FAIA200714, Elsevier: Scopus)
17.	Kohei Torii, Minoru Yoshida, Kazuyuki Matsumoto, Jiro Tsuruki, Kobayashi Kenichiro, Eiichi Honda and Kenji Kita : Towards Automatic Dental Diagnosis System Based on Deep Learning, Proc. IEEE International Conference on Natural Language Processing and Knowledge Engineering, 64-65, Online, Oct. 2020.
18.	Mawoneke Felix Kudzai, Luo Xin, Shi Youqin and Kenji Kita : Reverse Image Search for the Fashion Industry Using Convolutional Neural Networks, 2020 IEEE 5th International Conference on Signal and Image Processing (ICSIP), Oct. 2020. (Keyword) Reverse Image Search / Convolutional Neural Network / Fashion (Link to Publication Site) ● Publication site (DOI): 10.1109/ICSIP49896.2020.9339350 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85101075568 (DOI: 10.1109/ICSIP49896.2020.9339350, Elsevier: Scopus)
19.	Ryu Mopuaa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Construction of Annotated TOBYO Blog Corpus for Lifestyle Disease Analysis of Diabetic Patient, Proceedings of The 15th International Conference on Natural Language Processing and Knowledge Engineering (NLP-KE20), 77, Oct. 2020. (Summary) In this study, we target to analyze weblogs authored by diabetes patients describing their struggle against their disease. In order to build a corpus for analysis, we annotate tags to the important keywords/key phrases such as blood glucose level, diet and emotions in collected weblog data. We aim at constructing a labeled lifestyle disease weblog corpus (called as TOBYO weblog corpus) to extract useful information from patients' weblogs of against illness. For example, to improve our lifestyle, it would be very useful to know how we should live every day, what kind of diet we should eat, how much exercise is appropriate, and so on. In our study, we organize the tagged data and take statistics on part-of-speech combinations. We set the top N thresholds in the rank of the frequency of the part-of-speech and create a system to exclusively pre-process the key phrases that are ranked as more frequent part of speech combinations. The extracted key phrase information is vectorized by a model of BERT (bidirectional encoder representation from a transformer), then the key phrases are sorted out by making a model for category prediction that uses two inputs (BERT vector and character frequency vector). To evaluate our proposed method, we conduct cross validation test and open test by using several machine learning algorithms. In comparison with the baseline classification method, which is a simple and ordinal neural network method, our method can achieve higher performance. (Keyword) Tobyo corpus / neural networks / weblog / lifestyle-related disease / diabetes
20.	Liu Maobo, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Construction of lifestyle disease weblog corpus and evaluation of keyword classification model, International Forum on Advanced Technologies 2020 (IFAT2020), Mar. 2020. (Keyword) lifestyle-related disease / weblog corpus / keyword classification
21.	Kazuyuki Matsumoto, Manabu Sasayama, Minoru Yoshida and Kenji Kita : Emotional State Estimation by Dialogue History and Sentence Distributed Representation, Proceedings of 2019 6th IEEE International Conference on Cloud Computing and Intelligence Systems(CCIS2019), 31-35, Singapore, Dec. 2019. (Summary) The paper presents a novel approach to predict humans emotion in the dialogue. We target the scenario dialogue corpora. The corpora are annotated by some subjects with emotion tag. We use sentence embedding by Bidirectional Encoder Representations from Transformers as versatile and rich feature. By acquiring multi-context information from dialogue text, the accuracy of emotion prediction can be improved. As an experiment by machine learning algorithm such asdeep neural networks for the proposed method, we achieved higher accuracy than the method using embedding feature from utterance itself. (Keyword) dialogue sentences / dialogue emotion estimation / sensibility robotics (Link to Publication Site) ● Publication site (DOI): 10.1109/CCIS48116.2019.9073750 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85085005205 (DOI: 10.1109/CCIS48116.2019.9073750, Elsevier: Scopus)
22.	Jin Qiuzi, Luo Xin, Shi Youqun and Kenji Kita : Image Generation Method Based on Improved Condition GAN, International Conference on Systems and Informatics (ICSAI2019), 1290-1294, Shanghai, Nov. 2019. (Link to Publication Site) ● Publication site (DOI): 10.1109/ICSAI48974.2019.9010120 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85081970301 (DOI: 10.1109/ICSAI48974.2019.9010120, Elsevier: Scopus)
23.	Yuko Sakuta, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Kansei Mapping of Color Images Using Earth Mover's Distance, International Conference on Software Technology and Engineering (ICSTE2019), Oct. 2019.
24.	Akira Fujisawa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : An Approach for Conversion of Japanese Emoticons into Emoji Based on Character-Level Neural Autoencoder, Frontiers in Artificial Intelligence and Applications, Vol.320, 635-644, Oct. 2019. (Summary) In this paper, we propose a method for converting Japanese emoticons into emoji. The method creates a model that translates text into emoji using training neural networks with a character-based feature; the model determines the positive/negative/neutral polarity of the emoji from the text. By extracting a feature vector from the hidden layer of the model, we calculate the similarity between the input sentence and the text annotated with emoji in the database; the conversion candidates come from the emoticon in the input text. In comparison tests of the proposed method and a word-level feature method that uses a word distributed representation vector, the emoji emotion polarity-based model achieved a maximum accuracy rate of 90.0%, representing an 8% improvement over the word-level feature method. (Keyword) emoji / emoticon / neural autoencoder (Link to Publication Site) ● Publication site (DOI): 10.3233/FAIA190231 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.3233/FAIA190231 (DOI: 10.3233/FAIA190231)
25.	Kazuyuki Matsumoto, Mopaa Ryu, Minoru Yoshida and Kenji Kita : Emotion Analysis on Weblog of Lifestyle Diseases, Proceedings of the 2019 International Symposium on Signal Processing Systems (SSPS2019), 142-148, Beijing, Sep. 2019. (Summary) In this paper, we investigate the relation between patient lifestyle and emotional state transition by analyzing weblog articles written by patients with diabetes. To conduct our analysis, we use blood sugar level and body weight to indicate the disease condition, together with meal keywords and sentences that express feelings in the selected weblogs. A corpus annotated with a tag to word units in the target articles is constructed and a method of keyword classification based on the constructed corpus is proposed and evaluated. Finally, we analyze whether the indicated emotion is positive or negative in sentences expressing emotion in the corpus and investigate the relation between the emotion and a change in blood sugar level. (Keyword) text mining / diabetes / lifestyle diseases / emotion analysis / machine learning (Link to Publication Site) ● Publication site (DOI): 10.1145/3364908.3365294 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85077207726 (DOI: 10.1145/3364908.3365294, Elsevier: Scopus)
26.	Kazuyuki Matsumoto, Yuta Hada, Minoru Yoshida and Kenji Kita : Analysis of Reply-Tweets for Buzz Tweet Detection, Proceedings of 33rd Pacific Asia Conference on Language, Information and Computation (PACLIC33), 138-146, Sep. 2019. (Summary) In this study, we propose a method for predicting whether a tweet will create a buzz on the Internet by examining tweeted replies posted by others. We also investigate the distinguishing characteristics of replies to buzz tweets by analyzing feature amounts. Our proposed method first converts each reply tweet into a vector expression using a word distributed representation or some other vectorization method. We then apply a machine learning method for binary classification todetermine whether the reply is to a buzz tweet or a non-buzz tweet. We classify the target tweet into buzz or non-buzz categories by comparing the total buzz and non-buzz scores produced by the classifier. The proposed method using StarSpace achieved 93.1% F1-score, while an approach that used number of retweets and number of favors (likes) achieved 77.8% F1-score. Wealso found that there are a number of words that are characteristic of buzz tweet replies anda number of words that are characteristic of non-buzz tweet replies. (Keyword) text mining / Twitter / buzz tweet / StarSpace / Deep learning (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85084953625 (Elsevier: Scopus)
27.	Kazuyuki Matsumoto, Manabu Sasayama, Minoru Yoshida and Kenji Kita : Personality Analysis on Dialogue System for Emotional Interaction Robot, Conference Proceedings of Asia Pacific Society for Computing and Information Technology 2019 Annual Meeting (APSCIT 2019 Annual Meeting), Jul. 2019. (Summary) In this study, we proposed the personality analysis method for dialogue system towards realization of emotional interaction robot. To achieve smooth communication between human and robot in dialogue, it is necessary that overcome feeling of strangeness by generating dialogue utterance based on consistent personality. Firstly, we extract the wording feature from utterance texts on social media. To obtain relation between the wording feature and personality ego-state level, we use sentence level embedding feature and create state level estimator for each ego-state. Next, we evaluate the ego-state level estimator with an evaluation test set. Finally, we use this estimator for analysis of chat-bot dialogue. By comparing the existing chat-bot dialogue and user utterance, we consider about non-reality of current chat-bot dialogue generation. (Keyword) personality analysis / dialogue system / egogram / emotional interaction robot
28.	Kojima Takumi, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Research on statements in tweets and sleep time of classification, the 3rd International Conference on Natural Language Processing and Information Retrieval, Tokushima, Japan, Jun. 2019.
29.	Hidetoshi Nakao, Masakazu Imaoka, Kenji Oka, Mitsumasa Hida, Takeshi Morifuji, Masashi Hashimoto, Kazuyuki Matsumoto and Kenji Kita : Factor Affecting Low-Arch and High-Arch of Foot, 13th International Society of Physical and Rehabilitation Medicine (ISPRM 2019), No.P3-447, Jun. 2019.
30.	Hidetoshi Nakao, Masakazu Imaoka, Kenji Oka, Mitsumasa Hida, Takeshi Morifuji, Masashi Hashimoto, Kazuyuki Matsumoto and Kenji Kita : Body Factors of Low Arch of Foot for Middle and Senior Age, World Confederation for Physical Therapy (WCPT2019), No.PO-E-30-SUN2, May 2019.
31.	Horikawa Haruki, Minoru Yoshida, Kenji Kita and Kazuyuki Matsumoto : Extended Dataset for Food Image Classification Based on Twitter, Hong Kong International Conference on Engineering and Applied Science (HKICEAS 2018), Dec. 2018.
32.	Feng Zhihua, Luo Xin, Yang Tao and Kenji Kita : An object Detection System Based on YOLOv2 in Fashion Apparel, 2018 IEEE 4th International Conference on Computer and Communications (ICCC), Dec. 2018. (Link to Publication Site) ● Publication site (DOI): 10.1109/CompComm.2018.8780944 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85070807662 (DOI: 10.1109/CompComm.2018.8780944, Elsevier: Scopus)
33.	Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Modeling Relations Between Profiles and Texts, Proceedings of the 14th the Asia Information Retrieval Societies Conference (AIRS 2018), LNCS 11292, 103-109, Taipei, Nov. 2018. (Summary) We propose a method to model Twitter texts and user profiles simultaneously by considering the relations between the texts and profiles to obtain the distributed representations of the words in both. (Link to Publication Site) ● Publication site (DOI): 10.1007/978-3-030-03520-4_10 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85057273674 (DOI: 10.1007/978-3-030-03520-4_10, Elsevier: Scopus)
34.	Akira Fujisawa, Kazuyuki Matsumoto, Ohta Kazuki, Minoru Yoshida and Kenji Kita : ASCII Art Category Classification based on Deep Convolutional Neural Networks, Proceedings the 5th IEEE International Conference on Cloud Computing and Intelligence Systems (CCIS), 345-349, Nanjing, Nov. 2018. (Summary) In text-based communications on social media, emoticons, emojis, and stamps provide important ways to convey emotion and facilitate smoother communication in a short text. As non-verbal expressions, the characters in ASCII Art have no inherent meaning. However, by using a characters position or shape, emotion or intention can be expressed visually. Until now, it has been difficult to classify ASCII Art using the ASCII Art characters. In this paper, we propose a method for classifying ASCII Art by converting ASCII Art into images and using deep learning. In an evaluation experiment, the proposed method was able to achieve a classification success rate as high as 97%. (Keyword) ASCII Art / classification / Deep learning / Convolutional Neural Networks (Link to Publication Site) ● Publication site (DOI): 10.1109/CCIS.2018.8691245 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85064980913 (DOI: 10.1109/CCIS.2018.8691245, Elsevier: Scopus)
35.	Kazuyuki Matsumoto, Manabu Sasayama, Minoru Yoshida, Kenji Kita and Fuji Ren : Transfer Learning Based on Utterance Emotion Corpus for Lyric Emotion Estimation, Proceedings of 2018 5th IEEE International Conference on Cloud Computing and Intelligence Systems (CCIS), 699-703, Nanjing, Nov. 2018. (Summary) In prior research, there are various emotion estimation approaches that are corpus-based, dictionary-based, or rule-based. However, it is necessary to prepare language resources suitable to each domain, which will incur substantial cost. In this paper, we propose an approach to estimate emotion of lyric phrase by using transfer learning. We applied the existing utterance emotion corpus to the lyric emotion estimation task. As the result of an evaluation experiment, we obtained higher accuracy than the baseline method using a simple bag of words feature tf-idf weighted vector and logistic regression model. (Keyword) Transfer learning / Emotion estimation / Lyric / Word embedding (Link to Publication Site) ● Publication site (DOI): 10.1109/CCIS.2018.8691207 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85064985706 (DOI: 10.1109/CCIS.2018.8691207, Elsevier: Scopus)
36.	Bando Koji, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Twitter User's Hobby Estimation Based on Sequential Statements Using Deep Neural Networks, 2018 10th International Conference on Software Technology and Engineering, Oct. 2018. (Keyword) Twitter / hobby estimation / Neural Networks / random forests / support vector machine
37.	Yunong Wu, XIN KANG, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Emoticon-based emotion analysis for Weibo articles in sentence level, Lecture Notes in Computer Science, Vol.11248, 104-112, Oct. 2018. (Keyword) Emotion Analysis / Emoticon (Link to Publication Site) ● Publication site (DOI): 10.1007/978-3-030-03014-8_9 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85057082718 (DOI: 10.1007/978-3-030-03014-8_9, Elsevier: Scopus)
38.	Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Classification of Emoji Categories from Tweet Based on Deep Neural Networks, Proceedings of the 2nd International Conference on Natural Language Processing and Information Retrieval (NLPIR2018), 17-25, Bangkok, Sep. 2018. (Summary) In this paper, we describe the sentiment analysis method from tweets based on emojis category. Many of existing study about sentiment analysis focused on the emotional expressions included in sentence. However, because there are various kinds of emotional expressions, such as Internet slang, it cannot be constructed that the fixed emotional expression dictionary. The most of the methods based on corpus and machine learning, its performance is quite depended on the quality of annotation. Therefore, we attempt to use categories which are expressed by emoji as sentiment label instead of manually annotated labels. Our proposed method uses automatically annotated category label by emoji which is annotated to sentence, and train word embedding feature by deep neural networks. As the result of the experiment, our proposed method overcome the simple word feature based method. (Keyword) Emoji / Twitter / deep neural networks (Link to Publication Site) ● Publication site (DOI): 10.1145/3278293.3278306 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1145/3278293.3278306 (DOI: 10.1145/3278293.3278306)
39.	Nakao Hidetoshi, Jirela Dilixiati, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Emotion Recognition from Emoticons using Convolutional Neural Networks, Proceedings of 2nd Technological Competency as Caring in the Health Sciences 2018, 72, Aug. 2018. (Summary) In recent years, text communication has been grown steadily so that all kinds of daily contacts could be conducted now through Internet based text messaging services. As an advantage of text-based communication via the Internet, we can quickly interact with anyone wherever we are. At the same time, since it is difficult to express emotions with each other only with text, the chance of adding non-verbal information such as photos to text is increasing. Among the non-verbal information, there are emoticons in expressions that have been developed by Internet bulletin boards, chat, e-mail, etc. In particular, the types of emoticons used in Japan are diverse, even now; various types of emoticons continue to increase. There have been studies on these emoticons until now, which have become increasingly important in recent years. In this paper, we propose a method to classify facial expressions expressed from emoticons by letting deep convolutional neural networks. When recognizing emotions from emoticons, it is not known that seeing emoticons in character units is very effective. Information such as role positioning and coexistence roles is necessary. Characters in emoticons differ depending on font it also gives different impressions depending on context. In order to realize semantic analysis / emotion analysis considering emoticons, it is necessary to provide a corpus that records various kinds of emoticons and a database that gives information such as meanings and emotions expressed by emoticons. However, including all the emoticons is difficult, and since the emoticons are more vague than ordinary language expressions, that is why the construction of the dictionary is delayed. In this paper, we propose a method to classify the emotions of the emoticons by letting the deep convolutional neural network learn the features of the emoticons. By treating the emoticon as an image, the visual features can be captured and it is considered that natural recognition can be performed instead of processing like a character string. In addition, since the emoticon characterizes the facial expression with a character string, there is a problem that the impression is different only when the used fonts are different from each other, and therefore, when different fonts are used, how the classification accuracy is changed, we consider whether to do so. (Keyword) Emoticon / Convolutional Neural Networks / Emotion recognition / Data augmentation
40.	Yunong Wu, XIN KANG, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Sentence Emotion Classification for Intelligent Robotics Based on Word Lexicon and Emoticon Emotions, Proceedings of the International Conference of Intelligent Robotic and Control Engineering, 38-41, Aug. 2018. (Keyword) robotics / natural language processing (Link to Publication Site) ● Publication site (DOI): 10.1109/IRCE.2018.8492969 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85056756985 (DOI: 10.1109/IRCE.2018.8492969, Elsevier: Scopus)
41.	Ryusei Matsumoto, Minoru Yoshida, Kazuyuki Matsumoto, Hironobu Matsuda and Kenji Kita : Visualization of the occurrence trend of infectious diseases using Twitter, Proceedings of the 11th edition of the Language Resources and Evaluation Conference (LREC 2018), (poster), 511-514, May 2018. (Summary) We propose a system for visualizing the epidemics of infectious diseases. We apply factuality analysis both to disease detection and location estimation for accurate visualization. We tested our methods for several infectious diseases, and show that our method performs well on various diseases. (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85059887520 (Elsevier: Scopus)
42.	Jirela Dilixiati, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Predicting facial expression from emoticon based on convolutional neural networks, 4th International Forum on Advanced Technologies (IFAT 2018), Mar. 2018. (Summary) Text communication has been developed in recent years, all kinds of daily contacts are through the SNS regardless of age, gender and active. As an advantage of text-based communication via the Internet. There is a point that you can quickly respond to anyone wherever you are. In this paper, by letting the deep convolution neural network learn the image features of emoticons. We propose a method to classify by emotion expressed by emoticons. By treating emoticons as images, it is possible to capture visual features and think that natural identification can be made rather than handling as character strings. (Keyword) emoticon / convolutional neural network / emotion / facial expression
43.	Kenji Kita, Yunong Wu, Kazuyuki Matsumoto and Minoru Yoshida : The Effectiveness of Image Data Augmentation for Deep Learning, Conference Abstract of the 3rd International Conference on Multimedia and Image Processing, 19, Mar. 2018. (Keyword) Deep Learning / Data Augmentation / image processing
44.	Akira Fujisawa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Visual Similarity for ASCII Arts Using Image Features, Proceedings of the 2018 2nd International Conference on Machine Vision and Information Technology (CMVIT 2018), 17, Hong Kong, Feb. 2018. (Summary) Communication with various types of non-verbal expressions has been increasing on Social Networking Services. Among them, ASCII art is a character-based design technique to express human emotions. In many cases, ASCII arts are generated from original images by mapping some region of pixels to a certain character. In the field of natural language processing, ASCII arts are treated as sequences of characters. In addition, there have been few studies that aim to analyze or classify ASCII arts. In this paper, we extract characteristics of ASCII arts using image shape features. Our image shape-based features provide visual similarity between ASCII arts that are invariant across image scale and characters used in ASCII arts. Moreover, the proposed method enable us to compare visual similarity between ASCII arts and original images. To extract shape features from ASCII arts, we use image features of original images by converting ASCII arts to images. In the experiment, we compared visual similarity between ASCII arts and original images by using Histograms of Oriented Gradients(HOG). Our experimental result shows that the proposed method is effective to compare visual similarity between ASCII arts and original images. (Keyword) visual similarity / ascii art / image feature
45.	Yunong Wu, Kenji Kita, Kazuyuki Matsumoto, XIN KANG and Fuji Ren : Exploring the Relations Between Posts and Comments for Short Text Conversation, Proceedings of The 12th International Conference on Natural Language Processing and Knowledge Engineering (NLP-KE17), 228-238, Chengdu, Dec. 2017.
46.	Kazuyuki Matsumoto, Fuji Ren, Minoru Yoshida and Kenji Kita : Review Score Estimation Based on Transfer Learning of Different Media Review Data, Proceedings of The 12th International Conference on Natural Language Processing and Knowledge Engineering(NLP-KE'17), 185-199, Dec. 2017. (Summary) In this paper, we proposed a model to classify reviews based on review data of different media. Recently, researches have been actively made on transfer learning between different domains with various kinds of big data as a target. Evaluation expressions are usually different in different domains, and this becomes a barrier for reputation analysis. Users usually use different linguistic expressions to refer to the creative works of each media. For example, media such as "anime", "comics", "games" and "movies" have different terms or expressions to be described their creative works. These differences can be considered showing features of each medium and such differences should be found in other expressions as well as in evaluation expressions. We analyzed what effect such differences would cause to classication accuracy by conducting transfer learning between review data ofdifferent media. In this paper, we proved compatibility between original (pre-transferring) and target (post-transferring) media for each medium of the creative works by constructing a review classication model. As a result of evaluation experiments, we could more accurately estimate review score without using SO-Score for training of review fragment based on Long Short Term Memory (LSTM) than the SO-Score based method. (Keyword) review classification / transfer learning / Long Short-Term Memory / different media
47.	Hirokimi Fukuda, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Index Generation of BGM Video Based on Distinctive Comments, Proceedings of The 12th International Conference on Natural Language Processing and Knowledge Engineering(NLP-KE'17), 179-184, Dec. 2017. (Summary) The purpose of this study is to generate indexes to the BGMs based on the distinctive comments annotated to the BGM videos posted on Nico Nico Videos. Our proposed method detects the end/start positions of the BGMs considering the increase/decrease of the number of the comment data, the comments distinctive to the songs and the exclamatory expressions annotated to the videos. The evaluation result showed that the proposed method could generate correct indexes within 10 seconds error for the BGM videos with length of less than 40 minutes and with annotation of over 30,000 comments. (Keyword) Nico Nico Video / index generation / BGM video for work
48.	Naoya Fujino, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Emotion Estimation Adapted to Gender of User Based on Deep Neural Networks, Proceedings of The 12th International Conference on Natural Language Processing and Knowledge Engineering(NLP-KE'17), 166-178, Dec. 2017. (Summary) In this study, we focus on a representative SNS, Twitter and target emotion estimation from tweets posted on Twitter. we construct emotion estimation models applicable to each gender. It is considered that there would be different word usage tendencies depending on gender. Therefore, we analyze what difference exists in usage tendency of slangs or emoji, which are thought to be related to emotion, depending on gender. Then, we propose a method to improve emotion estimation based on neural networks by usingdistributed representation models for each gender. As the result of the evaluation experiments, the higher estimation accuracy was realized by training with Deep Convolutional Neural Networks using word's distributed representation as feature than by training with Feed Forward Neural Networks. (Keyword) emotion estimation / user's gender / deep neural networks
49.	Yunong Wu, XIN KANG, Kenji Kita and Fuji Ren : TUA1 at NTCIR-13 Short Text Conversation 2 Task, Proceedings of the 13th NTCIR Conference on Evaluation of Information Access Technologies, 211-214, Tokyo, Dec. 2017.
50.	Akira Fujisawa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Facial Expression Classification Based on Shape Feature of Emoticons, Proceedings of 1st International Conference on Machine Learning and Data Engineering (iCMLDE2017), 29-34, Sydney, Nov. 2017. (Summary) Emoticons are used in the situation of textual communication such as web mails and internet forums. Many of the existing studies dealing with classification or extraction of emoticons regard emoticons as a kind of character string and focus on what characters constitute the emoticons or how they are lined up. However, emoticons are used to express human facial expressions, and characters constituting them represent various facial parts such as eyes, nose, mouth, etc. Such characters can be identified as different facial parts depending on their positions, and facial expressions are thought to be represented by the combinations of their shape features. In this study, we classified the facial expressions of emoticons by focusing on the shape features of those emoticons. To deal with shape features of emoticons, we converted emoticons, which are text data, to image data. Emoticons are mainly formed by line segments of characters, and use only black and white colors. Therefore, other factors such as colors and shades were not considered as the feature to classify the facial expressions. In the experiments, we used image features that did not require color information. As the result of comparative experiment with the 1-nearest neighbor method using character features, the facial expression recognition rate is 52% when using the Histograms of Oriented Gradients(HOG) was used as image feature. By this result, proposed method improved recognition rate by 2 % than using baseline. (Keyword) facial expression classification / emoticon / image feature
51.	Takeshi Kusumoto, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Gender Estimation from Facial Images by Deep Neural Network, Proceedings of Annual Conference on Engineering and Applied Science, Nov. 2017. (Summary) In this paper, we collect new learning data and learned this by the existing neural network in order to exceed the accuracy of the model of the existing research. As a result, we succeeded in improving the accuracy by increasing the learning data. However, we discovered that performance declined due to biased learning data. The future task is to further improve accuracy and Research the relationship between bias and accuracy of learning data. (Keyword) gender estimation / facial image / deep neural network / image processing / deep learning
52.	Kazuyuki Matsumoto, Akira Fujisawa, Minoru Yoshida and Kenji Kita : Emotion Recognition of Emoticon Based on Character Embedding, 9th International Conference on Software Technology and Engineering (ICSTE2017), Hong Kong, Oct. 2017. (Summary) This paper proposes a method for emotion estimation of emoticon based on distributed representations of character meanings of the emoticon. The existing studies on emoticon have tried to extract emoticons from texts, estimate emotions of emoticons by breaking them up into constituent parts and by using their combination as feature. Based on the recently developed technique for word embedding, we try to realize versatile emotion estimation from emoticons by training the meanings of the characters constituting the emoticons and by using them as feature unit of emoticon based on the recent developed word embedding technique. The cross validation test was conducted for the emoticons emotion estimation model that was constructed based on the deep convolutional neural networks using distributed representations of characters as feature, and its result showed that our proposed method can estimate emotion of unknown emoticons with higher F1-score than the method based on the character n-gram appearance frequency vector as feature. (Keyword) emoticon / emotion recognition / character embedding / convolutional neural networks
53.	Kenji Kita, Kazuyuki Matsumoto and Minoru Yoshida : Face Identification Based on Deep Convolutional Neural Networks, 9th International Conference on Graphic and Image Processing, Oct. 2017. (Keyword) Face Identification / Deep Neural Networks
54.	Kazuyuki Matsumoto, Miyake Takeshi, Seiji Tsuchiya, Minoru Yoshida and Kenji Kita : Flame Prediction Based on Harmful Expression Judgement Using Distributed Representation, Book of Abstracts Proceeding of Engineering & Technology, Computer, Basic and Applied Sciences (ECBA-2017), Vol.3, No.16, 11, Taipei, Sep. 2017. (Summary) In recent years, flaming on social media has been a problem. To avoid flaming, it is useful for the system to automatically check the sentences whether they include the expressions that are likely to trigger flaming or not before posting messages. In this research, we target two harmful expressions. There are insulting expressions and the expressions that are likely to cause a quarrel. Firstly, we constructed an original harmful expression dictionary. Because a large cost requires to collect the expressions manually, we constructed the dictionary semi-automatically by using word distributed representations. The proposed method used distributed representations of the harmful expressions and general expressions as features, and constructed a classifier of harmful/general based on those features. An evaluation experiment found that the proposed method could extract harmful expressions with accuracy of approx. 70%.On the other hand, it was found that the proposed method could also extract unknown expressions, however, it wrongly extracted non-harmful. (Keyword) Flame Prediction / Harmful Expression / Distributed Representation / Support Vector Machine
55.	Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Distributed Representations for Words on Tables, Proceedings of the 21st Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2017), Part-I (LNAI 10234), 135-146, May 2017. (Summary) We consider a problem of word embedding for tables, and we obtain distributed representations for words found in tables. We propose a table word-embedding method, which considers both horizontal and vertical relations between cells to estimate appropriate word embedding for words in tables. We propose objective functions that make use of horizontal and vertical relations, both individually and jointly. (Link to Publication Site) ● Publication site (DOI): 10.1007/978-3-319-57454-7_11 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85018412327 (DOI: 10.1007/978-3-319-57454-7_11, Elsevier: Scopus)
56.	Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Acceleration of Similar Word Search in Distributed Representation, Proceedings of the 18th International Conference on Computational Linguistics and Intelligent Text Processing (CICLING 2017) (poster), Apr. 2017. (Summary) We report our study on accelerating search for similar words in distributed representation using cosine similarity measure.Our main idea is to convert similarity value to Euclidean distance, and apply the branch-and-bound approach to distance calculation itself, which means to terminate distance calculation when the current value exceeds the current N-th best similarity value when N is the size of final list of words similar to the query.We also apply other techniques for speeding up such as converting vector representation using PCA and reduction of adding and multiplying operations in distance calculation.
57.	Kenji Kita, Kazuyuki Matsumoto, Minoru Yoshida and Yunong Wu : A Comparison of Face Recognition Features Derived from Convolutional Neural Networks, Conference Abstracts of the 2017 International Conference on Frontiers of Image Processing, 17, Kathmandu, Nepal, Mar. 2017.
58.	Kazuyuki Matsumoto, Tanaka Satoshi, Minoru Yoshida, Kenji Kita and Fuji Ren : Ego-state Estimation from Short Texts Based on Sentence Distributed Representation, The 11th International Conference on Natural Language Processing and Knowledge Engineering (NLP-KE'16), Dec. 2016. (Keyword) ego-state / egogram / deep neural network / sentence distributed representation
59.	Akira Fujisawa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Improvement of the Histogram for Infrequent Color-Based Illustration Image Classification, Fuzzy Systems and Data Mining II, Frontiers in Artificial Intelligence and Applications, Vol.293, 299-305, Dec. 2016. (Keyword) Illustration image / Color histogram / Image classification / Style (Link to Publication Site) ● Publication site (DOI): 10.3233/978-1-61499-722-1-299 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.3233/978-1-61499-722-1-299 (DOI: 10.3233/978-1-61499-722-1-299)
60.	Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Table Topic Models for Hidden Unit Estimation, Proceedings of the 12th Asia Information Retrieval Societies Conference (AIRS2016), LNCS 9994, 302-307, Dec. 2016. (Summary) We propose a method to estimate hidden units of numbers written in tables. We focus on Wikipedia tables and propose an algorithm to estimate which units are appropriate for a given cell that has a number but no unit words. We try to estimate such hidden units using surrounding contexts such as a cell in the first row. To improve the performance, we propose the table topic model that can model tables and surrounding sentences simultaneously. (Link to Publication Site) ● Publication site (DOI): 10.1007/978-3-319-48051-0_23 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-85007107751 (DOI: 10.1007/978-3-319-48051-0_23, Elsevier: Scopus)
61.	Kazuyuki Matsumoto, Minoru Yoshida, Kenji Kita, Yunong Wu and Iwasa Fumihiro : Effect of Users Attribute on Emotion Estimation from Twitter, Proceedings of the 2nd IEEE International Conference on Computer and Communications, Vol.3, 1186-1190, Oct. 2016. (Summary) This study focused on emotion estimation fromutterances on Twitter and analyzed the differences caused forestimation according to the attributes of users. SNS userspossess various user attributes. If we could clarify the tendencyfor each attribute to involve emotion estimation, we would beable to create an emotion estimation model suitable for eachattribute. This study considered two attributes, sex and job, andinvestigated the tendency of emotion estimation based on anemotion estimation model constructed by machine learning. Inthis paper, as a feature for machine learning, we used asentence vector that was obtained by the summation of theword distributed expressions. As a result of the evaluationexperiment using a method based on the nearest neighbormethod, we obtained 0.4 higher estimation accuracies thanusing a baseline method based on a simple Bag of Words model. (Keyword) user attribute / emotion recognition / Twitter (Link to Publication Site) ● Publication site (DOI): 10.1109/CompComm.2016.7924892 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1109/CompComm.2016.7924892 (DOI: 10.1109/CompComm.2016.7924892)
62.	Kazuyuki Matsumoto, Seiji Tsuchiya, Minoru Yoshida and Kenji Kita : Judgment of Slang Based on Character Feature and Feature Expression Based on Slangs Context Feature, Communications in Computer and Information Science, Vol.652, 280-288, Kuala Lumpur, Sep. 2016. (Summary) This paper proposes a method to distinguish standardwords from slang words using information from the characters comprising thecharacter string. An experiment testing the effectiveness of our method showeda 30 % or more improvement in classification accuracy compared to the baselinemethod. We also use a contextual feature related to emotion to expand theunregistered slang word in the training data into other expressions and proposean emotion estimation method based on the expanded expressions. In our experiment,successful emotion estimation was obtained in nearly 54 % of the cases, anotably higher rate than with the baseline method. Our proposed method wasshown to have validity. (Keyword) Slang / Character feature / Context feature / Unknown expression (Link to Publication Site) ● Publication site (DOI): 10.1007/978-981-10-2777-2_25 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84989337887 (DOI: 10.1007/978-981-10-2777-2_25, Elsevier: Scopus)
63.	Zhang Guoliang, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Face Identification based on Deep Learning Using Low Pixel Images, International Conference on Engineering & Technology, Computer, Basic & Applied Sciences, Jul. 2016.
64.	Kazuyuki Matsumoto, Manabu Sasayama, Kohsuke Miyauchi, Minoru Yoshida and Kenji Kita : Lyric Retrieval System Based on Word's Intimacy Value, Advances in Engineering Research, Vol.24, 84-87, May 2016. (Keyword) Lyric retrieval / Query expansion / Word intimacy value
65.	Kazuyuki Matsumoto, Minoru Yoshida, Kenji Kita and Fuji Ren : An Approach to Refine Translation Candidates for Emotion Estimation in Japanese-English Language, Proceedings of the 7th International Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management, Vol.2, 74-83, Nov. 2015. (Summary) Researches on emotion estimation from text mostly use machine learning method. Because machine learning requires a large amount of example corpora, how to acquire high quality training data has been discussed as one of its major problems. The existing language resources include emotion corpora; however, they are not available if the language is different. Constructing bilingual corpus manually is also financially difficult. Wepropose a method to convert a training data into different language using an existing Japanese-English parallel emotion corpus. With a bilingual dictionary, the translation candidates are extracted against every word of each sentence included in the corpus. Then the extracted translation candidates are narrowed down into a set of words that highly contribute to emotion estimation and we used the set of words as training data. As theresult of the evaluation experiment using the training data created by our proposed method, the accuracy of emotion estimation increased up to 66.7% in Naive Bayes. (Keyword) Emotion Corpus / Emotion Estimation / Bilingual Tagged Corpus
66.	Kazuyuki Matsumoto, Minoru Yoshida, Seiji Tsuchiya, Kenji Kita and Fuji Ren : Slang Analysis Based on Variant Information Extraction Focusing on the Time Series Topics, the 10th International Conference on Natural Language Processing and Knowledge Engineering, Sapporo, Japan, Oct. 2015.
67.	Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Sensibility Estimation Method for Youth Slang by Using Sensibility Co-occurrence Feature Vector Obtained from Microblog, Proceedings of the 2015 IEEE International Conference on Computer and Communications (ICCC2015), Oct. 2015. (Summary) Social networking sites such as Twitter provide more opportunities to express what people think or intend in short text. In short text, abbreviations such as ASAP or joinus and emoticons are often used. Because these expressions are not registered into the existing dictionaries, these are analyzed as unknown expressions. That can be a bottleneck for improving accuracy of reputation analysis in text mining. To use context for unknown word clustering is a major method, however, it usually requires word segmentation process and it has weakness for split errors of unknown expressions such as youth slang. In this paper, we proposed a method to obtain the appropriate context even though unknown expressions cause split errors and estimate sensibility expressed in the text. Because the dimensions of the obtained context vector were enormous, we also proposed a method to create a feature vector based on the co-occurrence of the sensibility words as simple expression with low dimension. As an evaluation experiment, the proposed method showed certain accuracy even with the small training data. (Keyword) slang / microblog / Twitter / affective computing / sensibility classification (Link to Publication Site) ● Publication site (DOI): 10.1109/CompComm.2015.7387618 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84963939887 (DOI: 10.1109/CompComm.2015.7387618, Elsevier: Scopus)
68.	Kazuyuki Matsumoto, Kyosuke Akita, Minoru Yoshida, Kenji Kita and Fuji Ren : Estimate the Intimacy of the Characters Based on Their Emotional States for Application to Non-Task Dialogue, Proceedings of the 6th International Conference on Affective Computing and Intelligent Interaction (ACII2015), 327-333, Xi'an, Sep. 2015. (Summary) Non-task-oriented dialogue requires someschemes that enable smooth and flexible conversations with auser. For example, it would be possible to go beyond the closedrelationship between the system and the user by consideringthe users relationship with others in real life. In this paper,we focused on the dialogue made by the two characters in adrama scenario, and tried to express their relationship with ascale of intimacy degree. There will be such various elementsrelated to the intimacy degree as the frequency of response to theutterance and the attitude of a speaker during the dialogue. Wefocused on the emotional state of the speaker during the utteranceand tried to realize intimacy estimation with higher accuracy. Asthe evaluation result, we achieved higher accuracy in intimacyestimation than the existing method based on speech role. (Keyword) Intimacy Estimation / Scenario Corpus / Emotion Recognition (Link to Publication Site) ● Publication site (DOI): 10.1109/ACII.2015.7344591 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84964027376 (DOI: 10.1109/ACII.2015.7344591, Elsevier: Scopus)
69.	Kazuyuki Matsumoto, Minoru Yoshida, Qingmei Xiao, Luo Xin and Kenji Kita : Emotion Recognition for Sentences with Unknown Expressions based on Semantic Similarity by Using Bag of Concepts, Proceedings of the 12th International Conference on Fuzzy Systems and Knowledge Discovery (FSKD'15), 1428-1433, Aug. 2015. (Summary) In studies of emotion estimation from text, varietiesof methods have been attempted such as emotional expressiondictionary or sentence structure dictionary and emotion corpus.However, most of these methods targeted the expressions includedin the existing morphological analysis dictionaries, as a result,they did not pay enough attention to unknown words, especiallynewly coined words. In Japan, the growth of Internet communicationsites such as weblogs and social networking sites broughtyounger people especially in teens and in their 20s to create newwords and use them very often. We prepared an emotion corpusby collecting weblog article texts including new words, analyzedthe corpus statistically, and proposed a method to estimateemotions of the texts. Most slang such as Youth Slang is tooambiguous in sense classification to be registered into the existingdictionaries such as thesaurus. To cope with these words, wecreated a large scale of Twitter corpus to calculate sense similaritybetween words. We proposed to convert unknown to sense classid to process the words that were not included in learning data.We defined this as a method using Bag of Concepts as feature. Asa result of the evaluation experiment using several classifies, theproposed method was proved robustness for unknown expression. (Keyword) Bag of concepts / emotion recognition / unknown expression / semantic similarity (Link to Publication Site) ● Publication site (DOI): 10.1109/FSKD.2015.7382148 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84966470618 (DOI: 10.1109/FSKD.2015.7382148, Elsevier: Scopus)
70.	Minoru Yoshida, Kazuyuki Matsumoto, Qingmei Xiao, Xielifuguli Keranmu, Kenji Kita and Hiroshi Nakagawa : Extracting Corpus-Specific Strings by Using Suffix Arrays Enhanced with Longest Common Prefix, Proceedings of the 10th Asia Information Retrieval Society Conference (AIRS 2014), LNCS 8870, 360-370, Kuching, Dec. 2014. (Summary) We propose a new term extraction algorithm that considers all of the substrings as term candidates.Our algorithm uses a suffix array as the data structure that emulates the suffix tree of the corpus.We use two scoring functions, one of which is used to detect good substring boundaries as linguistic chunksand the other is to find domain-specific phrases and combine them with a re-ranking approach.Experiments show that the proposed all-substring term extraction algorithm shows good performancefor highly-frequent terms compared with the baseline algorithm that uses a morphological analyzer in the preprocessing step. (Link to Publication Site) ● Publication site (DOI): 10.1007/978-3-319-12844-3_31 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84912135895 (DOI: 10.1007/978-3-319-12844-3_31, Elsevier: Scopus)
71.	Kazuyuki Matsumoto, Sasayama Manabu, Qingmei Xiao, Fujisawa Akira, Minoru Yoshida and Kenji Kita : Reranking the Search Results for Lyric Retrieval Based on the Songwriters' Specific Usage of Words, Electronics, Communications and Networks IV Proceedings of the 4TH Internationa Conference on Electronics, Communications and Networks, 12 15 December 2014, Beijing, China, 1045-1052, Beijing, Dec. 2014. (Summary) In this paper, we proposed a method to search lyrics by considering tendency of the words used by the songwriters, and aimed to improve the search accuracy. The existing methods of query expansion for document retrieval tended to focus on the words co-occurrence relation obtained from large amount of text data or to focus on synonymy relation of words to expand query. However, lyrics have different characteris-tics from business documents or weblog articles. For example, lyrics contain a lot of expressions made with different combinations of words compared to other text styles. Therefore, it is not always effective to focus on co-occurrence relation or semantic relation on thesaurus. In this paper, we were on the assumption that the tendency of the words used in the lyrics are different among songwriters. And, we constructed a data of the important words for each songwriter. We used this important word list made for each songwriter to add weight on the search results by n-gram. According to the weights, we rearranged the rank order of the search result aiming to improve the accuracy. We conducted an evaluation experiment using the artificially-created fragment of lyrics including errors as search query, and found that the proposed method improved MRR compared to the baseline method. (Keyword) lyric retrieval / query expansion
72.	Kazuyuki Matsumoto, Fuji Ren, Qingmei Xiao, Minoru Yoshida and Kenji Kita : Emotion Predicting Method Based on Emotion State Change of Personae according to the Other's Utterance, Proceedings of the 3rd IEEE International Conference on Cloud Computing and Intelligence Systems(CCIS2014), 427-432, Nov. 2014. (Summary) It is a very difficult task to predict others emotion from dialogue flow for dialogue system or robot, however, that is a natural task for human.As the reason, because of the emotion of human changes at all times when he / she have a conversation with other people, these changes would be considered that oc-curred by responding internal emotional state to stimulus from external world .In the researches about dialogue system, most of them are superficial feature of utterance for question-answering system, thus, internal emotional state and stimulus from external world.In this paper, we focused on the scenario dialogue sentences. We annotated the various information to the corpora for helping analyze dialogue flow or human relationships and constructed the scenario emotion corpus. We discussed that the others utterances or emotions how effect to emotion changes by analyzing this cor-pora.As the method to predict the others emotion changes from self-emotion, we proposed the predicting method based on the emotion predicting weight. As the result of evaluation experiment, the effectiveness of the proposed method could be confirmed. (Keyword) emotion prediction / scenario corpus / emotion corpus (Link to Publication Site) ● Publication site (DOI): 10.1109/CCIS.2014.7175773 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84988227783 (DOI: 10.1109/CCIS.2014.7175773, Elsevier: Scopus)
73.	Minoru Yoshida, Kazuyuki Matsumoto, Kenji Kita and Hiroshi Nakagawa : Unsupervised Analysis of Web Page Semantic Structures by Hierarchical Bayesian Modeling, Proceedings of the Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD) 2014, Part II, LNAI 8444, 572-583, Tainan, May 2014. (Summary) We propose a Bayesian probabilistic modeling of the semantic structures of HTML documents.We assume that HTML documents has logically hierarchical structures and model them as links between blocks.These links or dependency structures are estimated via sampling methods. We use hierarchical Bayesian modeling where each block is given labels like ``heading''or ``contents'', and words and layout features (i.e., symbols and HTML tags) are generated simultaneouslybased on these labels. (Link to Publication Site) ● Publication site (DOI): 10.1007/978-3-319-06605-9_47 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84901278613 (DOI: 10.1007/978-3-319-06605-9_47, Elsevier: Scopus)
74.	Akira Fujisawa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Identifying who drew the illustration focusing on the eyes of the characters, Proceedings of 20th Korea-Japan Joint Workshop on Frontiers of Computer Vision, 82-89, Feb. 2014. (Summary) In this paper, we aimed to automatically identify who draw the of illustration image. Designs of illustration image differ depending on the illustrators. Also, it is difficult to find out common characteristics from the several illustrations drawn by illustrator. We tried to identify who drew the illustration by detecting distinctive habits of the illustrator from his/her works and by using them as the feature. To obtain the feature, we focused on the area of eyes of the illustrated characters. We obtained the edge information from there and calculated percentage of the edges. We call this percentage a "Line Percent" and considered it as the unique feature quantity of the illustrator. From an experiment, the average recall was 35 % and some of the recalls were over 90%. This result suggested that we could identify who draw the illustration images to some extent by using Line Percent as the feature. (Keyword) image processing / Identifying / Illustration / Edge / Canny Edge Detection
75.	Qingmei Xiao, Chen Mei, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Improvement of Indexing Methods for Audio Fingerprinting Systems, International Conference Data Mining, Civil and Mechanical Engineering (ICDMCME'2014), 65-71, Feb. 2014. (Summary) In music retrieval methods based on audio fingerprints, a large database is required in order to compare the fingerprints extracted from the query. It is necessary to develop an efficient search method. In this paper, we propose a method for index compression using a compressed suffix array. The proposed method compresses the index by encoding the 8-bit data sequences by Run Length Encoding. We also use Fibonacci Hashing function which can adjust the size of hash table according to the capacity of memory, provide a good distribution of hash value, and save memory.
76.	Mei Chen, Qingmei Xiao, Kazuyuki Matsumoto, Minoru Yoshida, Xin Luo and Kenji Kita : A Fast Retrieval Algorithm Based on Fibonacci Hashing for Audio Fingerprinting Systems, Proceedings of the 2013 International Conference on Advanced Information Engineering and Education Science, 219-222, Beijing, Dec. 2013. (Summary) In audio fingerprinting system, the database consists of hundreds of millions of sub-fingerprints. How to find out the most similar audio in the shortest time and use less memory in huge repository of sub-fingerprints is a hot topic of research. Recently, Philips introduced an effective search method based on hash table. In this paper, a fast retrieval algorithm based on the Fibonacci Hashing is proposed as an extension of Philipss method. The algorithm uses Fibonacci Hashing function which can adjust the size of hash table according to capacity of memory, provide a good distribution of hash value, and save memory. The performance of presented algorithm has been evaluated by experiments.
77.	Motoyuki Suzuki, Shohei Nakagawa and Kenji Kita : Emotion recognition method based on normalization of prosodic features, Proceedings of the APSIPA Annual Summit and Conference (ASC) 2013, 1-5, Nov. 2013. (Summary) Emotion recognition from speech signals is one of the most important technologies for natural conversation between humans and robots. Most emotion recognizers extract prosodic features from an input speech in order to use emotion recognition. However, prosodic features changes drastically depending on the uttered text. In order to solve this problem, we have proposed the normalization method of prosodic features by using the synthesized speech, which has the same word sequence but uttered with a neutral emotion. In this method, all prosodic features (pitch, power, etc.) are normalized. However, nobody knows which kind of prosodic features should be normalized. In this paper, all combinations of with/without normalization were examined, and the most appropriate normalization method was found.
78.	Ken Ichikawa, Satoru Tsuge, Norihide Kitaoka, Kazuya Takeda and Kenji Kita : Spoken document retrieval using both word-based and syllable-based document spaces with latent semantic indexing, Proceedings of the APSIPA Annual Summit and Conference (ASC) 2013, Oct. 2013. (Summary) In this paper, we propose a spoken document retrieval method using vector space models in multiple document spaces. First we construct multiple document vector spaces, one of which is based on continuous-word speech recognition results and the other on continuous-syllable speech recognition results. Query expansion is also applied to the word-based document space. We proposed to apply latent semantic indexing (LSI) not only to the word-based space but also to the syllable-based space, to reduce dimensionality of the spaces using implicitly defined semantics. Finally, we combine the distances and compare the distance between the query and the available documents in various spaces to rank the documents. In this procedure, we propose to model the document by hyperplane. To evaluate our proposed method, we conducted spoken document retrieval experiments using the NTCIR-9 SpokenDoc data set. The results showed that using the combination of the distances, and using LSI on the syllable-based document space, improved retrieval performance.
79.	Satoru Tsuge, Ken Ichikawa, Norihide Kitaoka, Kazuya Takeda and Kenji Kita : Spoken Content Retrieval Using Distance Combination and Spoken Term Detection Using Hash Function for NTCIR10 SpokenDoc2 Task, Proceeding of the 10th NTCIR Conference, 597-603, Jun. 2013. (Summary) In this paper, we describe a spoken content retrieval (SCR) and a spoken term detection (STD) which were used in the 2nd round of the IR (Information Retrieval) for Spoken Documents (SpokenDoc2) task. The SCR method maps the target documents into multiple vector spaces, which are the word-based vector space constructed by the word-based speech recognition results and the syllable-based vector space constructed by the syllable-based speech recognition results, by a vector space model. In the STD method, the sub-sequences extracted from target document are converted into the bit sequences by using the hash function. A query is also converted into a bit sequence by the same way. The candidates are detected by calculating hamming distance between the bit sequence of the query and that of target documents.
80.	Yunong Wu, Kenji Kita, Kazuyuki Matsumoto and Kang Xing : A Joint Prediction Model for Multiple Emotions Analysis in Sentences, 14th International Conference on Intelligent Text Processing and Computational Linguistics, 149-160, Mar. 2013. (Summary) In this study, we propose a scheme for recognizing people's multiple emotions from Chinese sentence. Compared to the previous studies which focused on the single emotion analysis through texts, our work can better reflect people's inner thoughts by predicting all the possible emotions. We first predict the multiple emotions of words from a CRF model, which avoids the restrictions from traditional emotion lexicons with limited resources and restricted context information. Instead of voting emotions directly, we perform a probabilistic merge of the output words' multi emotion distributions to jointly predict the sentence emotions, under the assumption that the emotions from the contained words and a sentence are statistically consistent. As a comparison, we also employ the SVM and LGR classifiers to predict each entry of the multiple emotions through a problem transformation method. Finally, we combine the joint probabilities of the multiple emotions of sentence generated from the CRF-based merge model and the transformed LGR model, which is proved to be the best recognition for sentence multiple emotions in our experiment.
81.	Satoru Tsuge, Hiromasa Ohashi, Norihide Kitaoka, Kazuya Takeda and Kenji Kita : Spoken document retrieval using combinational use of distances of multiple vector spaces and query expansion with optimized weight parameters, Proc. NCSP'13, 9-12, Mar. 2013. (Summary) We propose a distance combination method for spoken document retrieval. The proposed distance combination method maps the target documents, which are transcription of speech made by a speech recognizer, to multiple vector spaces. We, then, calculate the distance between a query vector and multiple versions of the documents on each vector space and combine these distances for retrieval. In addition, we propose a novel weight parameter determination method for query expansion, to minimize an angle between an expanded query vector and a document vector. The proposed methods achieved the best retrieval performance on a SpokenDoc task in an NTCIR-9 workshop.
82.	Yunong Wu, Kenji Kita and Kazuyuki Matsumoto : Recognition of Multiple Emotion States of Words using Multi-engine System, 11th International Conference on Computer Applications 2013, 201-206, Feb. 2013. (Summary) In this paper, we propose a multi-engine voting system to recognize the multiple emotion states expressed in the Chinese words according to the word identity and the context information, which can better reflect the people's real emotion states compared to previous works focused on single emotion analysis. Conditional Random Field and Maximum Entropy are employed to learn two probabilistic models to predict the possible emotions and assign the corresponding probability scores to the eight basic emotion types as well as the no-emotion type. To solve the problem of imbalanced data source, we introduce a threshold to improve the recognition of no-emotion and other word emotions. Our experiment achieves promising results which prove that multiple emotion states can be better recognized by the multi-engine system rather than either Conditional Random Field or Maximum Entropy (Keyword) Emotion Estimation / Maximum Entropy / Conditional Random Fields
83.	Yunong Wu, Kenji Kita and Kazuyuki Matsumoto : Recognition of Multiple Emotion States of Words using Multi-engine System, 11th International Conference on Computer Applications 2013, 201-206, Feb. 2013.
84.	Kazuyuki Matsumoto, Kenji Kita and Fuji Ren : Emotion Estimation from Sentence Using Relation between Japanese Slangs and Emotion Expressions, Proceedings of 26th Pacific Asia Conference on Language,Information and Computation, 377-384, Nov. 2012. (Summary) Most of Japanese slang words such as Wakamono Kotoba are analyzed as unknown word or segmented wrongly by the morphological analysis system. These problems are causing negative effect on sentiment analysis in text. These words generally have many varieties of notations and conjugations, and they lack versatility. As a result, many of them are not registered in the dictionaries, making morphological analysis more difficult. In this paper, we aimed to decrease such negative effects of Wakamono Kotoba for the accuracy of emotion estimation from sentence and proposed a method to increase the accuracy by using a classification method based on machine learning. In this method we used emotional expressions which had high relevance with Wakamono Kotoba as feature. As a result, the proposed method obtained 20% higher accuracy than the method only using morpheme Ngram as feature. (Keyword) Emotion Corpus / Japanese Slang / Out of Vocabulary
85.	Kazuyuki Matsumoto, Yoshioka Nobuhiro, Kenji Kita and Fuji Ren : Judgment of Depressive Tendency from Emotional Fluctuation in Weblog, Proceedings of the 16th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems - KES2012, 545-554, Sep. 2012. (Summary) Some people who have tendency toward depression describe their inner thoughts vividly in their weblogs. Such descriptions are sometimes made consciously but at other times unconsciously by the authors. If such unconscious authors could become aware of their own symptom of the depression, that would contribute to early detection and treatment of the sickness. To achieve this, it is essential to find out the tendency of depression from the descriptions in weblog. Because the symptom of depression is not static, we attempted to detect the tendency of depression focusing on the temporal fluctuation of emotion of the authors. As many weblogs describe what happened or what the authors felt in daily life like a personal journal, we thought that weblogs were ideal for obtaining the time-varying emotion of the authors. In this paper, we proposed the method to detect the tendency of depression in weblog. Our method is based on the temporal fluctuation of emotion described by the authors. (Keyword) Weblog / Emotional Fluctuation / Depressive Tendency
86.	Luo Xin, Pan Qiao, Chen Mei and Kenji Kita : Learning Distance Metrics with Feature Space Performance for Image Retrieval, 2012 Third International Conference on Mechanic Automation and Control Engineering (MACE '12), 3579-3583, Jul. 2012.
87.	Le Zhang and Kenji Kita : Skeleton Extraction and Pruning based on Multi-scale Turning Angle Descriptor, 18th Korea-Japan Joint Workshop on Frontiers of Computer Vision, 256-262, Feb. 2012. (Summary) The skeleton is an important feature in the representation of a shape in image analysis. This paper includes two parts. First, we propose an Euclidean distance-ordered thinning algorithm for skeleton extraction. Second, since the skeleton is sensitive to an object's boundary noise, we propose multi-scale turning angle descriptor to locate the salient skeleton points on the boundary, and we could cunduct the skeleton pruning by the salient points.
88.	Luo Xin and Kenji Kita : Region-Based Image Annotation Using Gaussian Mixture Model, 2012 International Conference on Information Technology and Software Engineering, 503-510, 2012.
89.	Wu Yunong, Kenji Kita, Fuji Ren, Kazuyuki Matsumoto and Kang Xin : Exploring Emotional Words for Chinese Document Chief Emotion Analysis, Proceedings of 25th Pacific Asia Conference on Language, Information and Computation(PACLIC2011), 597-606, Dec. 2011. (Summary) In this study, we develop a document emotion analysis model by making use of the function of emotional words annotation. Eight basic emotion categories have been selected, including Expect, Joy, Love, Surprise, Anxiety, Sorrow, Anger and Hate, for both the word and the document level emotion analysis. We introduce two parameters, term relevance and term frequency, to evaluate the relations between the word emotions and the document emotions. Promising experimental results reveal the effectiveness of our document emotion analysis model under different emotion situations. (Keyword) emotional words annotation / document emotion analysis / term relevance / term frequency
90.	Yunong Wu, Kenji Kita, Fuji Ren, Kazuyuki Matsumoto and XIN KANG : Exploring Emotional Words for Chinese Document Chief Emotion Analysis, The 25th Pacific Asia Conference on Language, Information, and Computation (PACLIC25), 597-606, Dec. 2011.
91.	Wu Yunong, Kenji Kita, Fuji Ren, Kazuyuki Matsumoto and XIN KANG : Exploring Emotional Words for Chinese Document Chief Emotion Analysis, The 25th Pacific Asia Conference on Language, Information, and Computation (PACLIC25), 597-606, Dec. 2011.
92.	Kazuyuki Matsumoto, Kenji Kita and Fuji Ren : Emotion Estimation of Wakamono Kotoba Based on Distance of Word Emotional Vector, Proc. IEEE International Conference on Natural Language Processing and Knowledge Engineering, 214-220, Tokushima, Nov. 2011. (Summary) Recently, there are many research that aim at estimating emotion from text. The meaning of language expression that we use in daily life varies depending on the context used. In that sense, language expression conveys ambiguous information. Wakamono Kotoba are words used by young people, and they also have semantic ambiguity. Most of the Wakamono Kotoba are not registered in the existing dictionaries, making it difficult for a system to recognize the meaning of Wakamono Kotoba. In this paper, we propose a method for estimating emotion in Wakamono Kotoba that were not registered in the system, by using Wakamono Kotoba example sentences as features. The proposed method applied Earth Movers Distance to vector of words. As a result of the evaluation experiment, higher estimation accuracy was obtained by considering emotional distance between words -- an approach that had not been used in the conventional research -- than by using only word importance value. (Link to Publication Site) ● Publication site (DOI): 10.1109/NLPKE.2011.6138197 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-84857271107 (DOI: 10.1109/NLPKE.2011.6138197, Elsevier: Scopus)
93.	Yunong Wu, Kenji Kita, Fuji Ren, Kazuyuki Matsumoto and XIN KANG : Modification Relations based Emotional Keywords Annotation using Conditional Random Fields, The 4th International Conference on Intelligent Networks and Intelligent Systems (ICINIS2011), 81-88, Nov. 2011.
94.	Wu Yunong, Kenji Kita, Fuji Ren, Kazuyuki Matsumoto and XIN KANG : Modification Relations based Emotional Keywords Annotation using Conditional Random Fields, The 4th International Conference on Intelligent Networks and Intelligent Systems (ICINIS2011), 81-88, Nov. 2011.
95.	Qingmei Xiao, Motoyuki Suzuki and Kenji Kita : Fast Hamming Space Search for Audio Fingerprinting Systems, Proc. International Conference on Music Information Retrieval, 133-138, Oct. 2011. (Summary) In music information retrieval, a huge search space has to be explored because a query audio clip can start at any position of any music in the database, and also a query is often corrupted by significant noise and distortion. Audio fingerprints have recently attracted much attention in music information retrieval, for they provide a compact representation of the perceptually relevant parts of audio signals. In this paper, we propose an extremely fast method of exploring a huge Hamming space for audio fingerprinting systems. The effectiveness of the proposed method has been evaluated by experiments using a database of 8,740 songs.
96.	Wu Yunong, Kenji Kita, Fuji Ren, Kazuyuki Matsumoto and Kang Xin : Modification Relations Based Emotional Keywords Annotation Using Conditional Random Fields, Proceedings of 2011 Fourth International Conference on Intelligent Networks and Intelligent Systems, 81-84, Oct. 2011. (Summary) In this study, we demonstrate the effectiveness of modification features for emotional keywords annotation on Chinese sentences extracted from a Chinese emotion corpus. Eight basic word emotion categories have been selected and Conditional Random Fields is employed in eight experiments with different sets of features. Our result shows that modification features make the contribution to emotional keywords annotation. (Keyword) modification features / emotional keywords annotation / CRFs
97.	Masami Shishibori, Sangkon Lee and Kenji Kita : An Improved Method to Select Candidates on Metric Index VP-tree, Proceedings of the International Conference on Knowledge Discovery and Information Retrieval (KDIR 2011), Oct. 2011. (Summary) VP-tree is an efficient metric space indexing method, however the number of distance calculations at leaf nodes tends to increase. In this paper, we propose an efficient algorithm to reduce the number of distance calculations at leaf nodes of the VPtree. The conventional VP-tree uses the triangle inequality at the leaf node in order to reduce the number of distance calculations. At this point, the vantage point of the VP-tree is used as a reference point of the triangle inequality. The proposed algorithm uses the nearest neighbor (NN) point for the query instead of the vantage point as the reference point. By using this method, the selection range by the triangle inequality becomes small, and the number of distance calculations at leaf nodes can be cut down.
98.	Qingmei Xiao, Satoru Tsuge and Kenji Kita : Music Retrieval Method Based on Filter-bank Feature and Earth Mover's Distance, 7th International Conference on Natural Computation, 1872-1876, Jul. 2011. (Summary) In this paper, we propose a novel music retrieval method using a filter bank feature and earch mover's distance (EMD). In the proposed method, we use MFCCs for acoustic features and EMD for a distance measure. Evaluation experimental results show that the accuracy of the proposed method achieves 96.73%.
99.	Chen Mei, Luo Xin, Wu Guowen, Wu Guowen, Tan Yang and Kenji Kita : Security Clustering Algorithm Based on Reputation in Hierarchical Peer-to-Peer Network, 4th IEEE International Conference on Computer Science and Inforymation Technology, 222-225, Jun. 2011.
100.	Luo Xin, Chen Mei, Pan Qiao, Wu Guowen and Kenji Kita : Optimum Distance Metric Based on Image Feature Space, 4th IEEE International Conference on Computer Science and Information Technology, 258-263, Jun. 2011.
101.	Zhang Wenbin, Lu Haoze, Horiuchi Yasuo, Satoru Tsuge, Kenji Kita and Shingo Kuroiwa : Text-Independent Speaker Identification Based on Reducing Inter-Session Variability of Speech Feature Using PCA Transformation, 2011 International Workshop on Nonlinear Circuits, Communication and Signal Processing, 421-424, Mar. 2011. (Summary) In text-independent speaker identification, the phonetic variability and the intersession variability strongly afect the performance of speaker identification. In this paper, PCA transformation is utilized to reduce intersession variability of speech data in the observed space. The proposed method reduced the identification error rate by 42.6% compared with the conventional MFCCs and reduced the identification error rate by 27.2% compared with the MFB-PCA.
102.	Masami Shishibori and Kenji Kita : Instance Search Experiments for TRECVID 2010, Proceedings of TREC Retrieval Evaluation 2010 (TRECVID2010), Mar. 2011. (Summary) On TRECVID 2010, we have participated in the instance search (INS) pilot task. For the INS task, our main focus was to apply SIFT-based image retrieval method with facial images. The overview of our retrieval system is described as the below. At the registration phase, facial images are extracted from each cut frame using the Haar-like operator. Next, noise images in the extracted facial images are eliminated using the SVM. Finally, SIFT features are detected from true facial images, and these features are registered in the facial database. At the retrieval phase, the facial area of the query image is extracted using the same operator as the registration phase. Next, SIFT features are detected from the query facial image. After that, the similarity between query facial image and the facial database is calculated based on 128 dimensional SIFT features.
103.	Masami Shishibori, Sangkon Lee and Kenji Kita : A Method to Improve Metric Index VP-tree for Multimedia Databases, Proc. of 17th Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV2011), 55-60, Feb. 2011.
104.	Chen Mei, Wu Guowen, Luo Xin, Kenji Kita and Yang Liqin : First Coming Clustering Algorithm Based on Reputation in Peer-to-Peer Network, 2010 WASE International Conference on Information Engineering, Aug. 2010.
105.	Zhang Le, He Qing, Shin-ichi Ito and Kenji Kita : Euclidean Distance-Ordered Thinning for Skeleton Extraction, Proc. of the 2nd International Conference on Education Technology and Computer, ICETC'10, Vol.1, 311-315, Shanghai, Jun. 2010. (Summary) The skeleton is an important feature for the representation of a shape in image analysis. In this paper, we propose a novel Euclidean distance-ordered thinning algorithm for skeleton extraction. The experiment results show that the algorithm is able to extract the connected and one-pixel wide skeleton that can correctly preserve the topology of the object. Furthermore, the extracted skeleton locates on the accurate position and it is insensitive to boundary noise.
106.	Zhang Le, He Qing, Luo Xin and Kenji Kita : A Skeleton Extraction and Pruning Algorithm Based on Bisector Angle, 16th Korea-Japan Joint Workshop on Frontiers of Computer Vision, 327-335, Feb. 2010. (Summary) The skeleton is an important feature in the representation of a shape in image analysis. This paper includes two parts. First, we propose a fast, efficient and accurate skeleton extraction method based on the criterion which can be used to determine whether a given pixel is a skeleton point independently. Second, since the skeleton algorithm is sensitive to an object's boundary noise, we propose a pruning method based on bisector angle, which can completely remove redundant protrusions and obtain accurate skeleton and preserve the original topological property of a shape.
107.	Le Zhang, Zhao Zuyun, Luo Xin, Zeng Peifeng and Kenji Kita : A Research of Distance Transform based Peeling Algorithm for Overlapped Fibers Separation, International Conference on Computational Intelligence and Software Engineering, Dec. 2009. (Summary) Overlapped fiber separation is one of main procedures in cross-sectional fiber analysis. In the proposed distance transform based peeling algorithm, a separation point is determined by removing points with the least distance. The overlapped fibers in cross-section are separated by the separation point. Experiments show that the proposed algorithm precisely separates overlapped fibers of various shapes.
108.	Masami Shishibori, Koizumi Daichi and Kenji Kita : A Fast Retrieval Algorithm for the Earch Mover's Distance Using EMD Lower Bounds and the Priority Queue, IEEE NLP-KE 2009, Dalian, China, Sep. 2009. (Summary) The earth mover's distance (EMD) is a measure of the distance between two distributions, and it has been widely used in multimedia information retrieval systems, in particular, in content-based image retrieval systems. When the EMD is applied to image problems based on color or texture, the EMD reflects the human perceptual similarities. However, its computations are too expensive to use in large-scale databases. In order to achieve efficient computation of the EMD during query processing, we have developed fastEMD, a library for high-speed feature-based similarity retrievals in large databases. This paper introduces techniques that are used in the implementation of the fastEMD and demonstrates the efficiency in extensive experiments.
109.	Q. Zhang, X-Y. Tai, Y. Dong, S. Pan, X. Luo and Kenji Kita : Two-Level Content-based Endoscope Image Retrieval, Fourth International Conference on Natural Computation, 208-212, Oct. 2008. (Summary) Based on the analysis of endoscope image, in this paper, a new color quantification method is proposed to extract improved CCV and the V component shape invariant moment achieving image feature base.
110.	Masami Shishibori, Satoru Tsuge, Le Zhang, Sasaki Minoru, Uemura Yoshiki and Kenji Kita : A Fast Retrieval Algorithm for the Earch Mover's Distance Using Lower Bounds, IEEE International Conference on Information Reuse and Integration, 445-450, Jul. 2008. (Summary) The Earth Mover's Distance (EMD) was introduced in Computer Vision as better distance measure between two distributions, and has been widely used in multimedia databases. Rubner et al. have been proposed CBIR model using the EMD, and its high quality for image similarity search has been demonstrated. Formally, the Earth Mover's Distance is defined as a Linear Programming problem which can be solved using the simplex method. Its computation, however, is too complex to apply the EMD in interactive multimedia database systems. In order to achieve the efficient retrieval, we propose a fast retrieval algorithm for the EMD. This method uses the lower bounding distance. The lower bounding distance of the EMD is simple approximation computationally.
111.	Nobuki Nishikawa, Masami Shishibori, Satoru Tsuge and Kenji Kita : A Method to Retrieve Telops Based on the Distance of Character Image Features between Query and Telop, Proceedings of 14th Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV2008), 25-30, Oita, Jan. 2008. (Summary) In this paper, a new video scene retrieval method based on telop characters is proposed. This method calculates the distance between each image features of telop characters and template image features of query keyword. The number of distance calculations can decrease by indexing the multidimensional data for image features of telop characters. Experimental results, using 91 query keywords, show that the average precision of proposed method using 1-gram feature becomes 98.61%, and using 2-gram feature becomes 99.59%. Moreover, the retrieval time can be obtained in about 0.5 seconds when using 2-gram feature.
112.	Masami Shishibori, Masayuki Muto and Kenji Kita : A Fast Retrieval Method for the Earth Mover's Distance in Multimedia Databases, Proceedings of 14th Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV2008), 35-40, Oita, Jan. 2008.
113.	Mitsuhiro Ozawa, Satoru Tsuge, Masami Shishibori, Kenji Kita, Minoru Fukumi, Fuji Ren and Shingo Kuroiwa : Automatic Utterance Segmentation Tool for Speech Corpus, IEEE NLP-KE2007, 401-406, Beijing, Aug. 2007. (Summary) We collect the speech data for investigating an intra-speakers' speech variability over a short and long time. In general, to reduce the load of speakers, the speech data are collected as one file from collecting start to collecting end. Hence, there are some noises, non-speech sections and mistaken sections in this file. Consequently, we must segment this file into individual utterances and select the useful utterances. This process requires a lot of time and efforts. In this paper, we propose an automatic utterance segmentation tool for dividing the collected speech data. The proposed tool is composed of four processes, which are a voice activity detection, speech recognition, a DP matching, and a correct of speech section. For evaluating the proposed tool, we conduct the evaluation experiments using a female speaker's speech data in our corpus. Experimental results show that the proposed method can reduce a filing time by 90% compared to a manual filing. In This paper, first, we introduced the large speech corpus. This speech corpus contains is the speech data collected by specific speaker over long and short time periods. And, we explained the automatic utterance segmentation tool which we made in the case of corpus build. And inspected the validity. As a result, it was demonstrated that the automatic utterance segmentation tool was high-performance. Furthermore, it was demonstrated that speech corpus build became simple by using the automatic utterance segmentation tool. (Link to Publication Site) ● Publication site (DOI): 10.1109/NLPKE.2007.4368062 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-47749136580 (DOI: 10.1109/NLPKE.2007.4368062, Elsevier: Scopus)
114.	Satoru Tsuge, Keiji SEIDA, Masami Shishibori, Kenji Kita, Fuji Ren, Minoru Fukumi and Shingo Kuroiwa : Analysis of Variation on Intra-Speakers Speech Recognition Performances, IEEE NLP-KE2007, 387-392, Beijing, Aug. 2007. (Summary) Even if a speaker uses a speaker-dependent speech recognition system, speech recognition performance varies. However, the relationships between intra-speaker's speech variability and speech recognition performance are not clear. To investigate these relationships, we have been collecting speech data since November 2002. In this paper, we analyze the relationships between intra-speaker's speech variability and the phoneme accuracy by a correlation analysis. Analyzed results showed the strong negative correlation between the phoneme accuracy and the speaking rate. The correlation coefficient indicated -0.77. Moreover, we can see that the phoneme accuracy is correlated with the temperature in the recording room and the humidity difference. (Link to Publication Site) ● Publication site (DOI): 10.1109/NLPKE.2007.4368060 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1109/NLPKE.2007.4368060 (DOI: 10.1109/NLPKE.2007.4368060)
115.	Masaki Naito, Kazumori Matsumoto, Masami Shishibori, Kenji Kita, Marco Cuturi, Tomoko Matsui, Shin'ichi Sato, Keiichiro Hoashi, Fumiaki Sugaya and Yasuyuki Nakajima : Shot Boundary Detection and High-Level Feature Extraction Experiments for TRECVID 2006, Proceedings of TREC Retrieval Evaluation 2006 (TRECVID2006), Washington, D.C., Mar. 2007.
116.	Satoru Tsuge, Minoru Fukumi, Masami Shishibori, Fuji Ren, Kenji Kita and Shingo Kuroiwa : Study of Relationships Between Intra-Speaker's Speech Variability and Speech Recognition Performance, 2006 International Symposium on Intelligent Signal Processing and Communication Systems (ISPACS 2006), 41-44, Tottori, Japan, Dec. 2006. (Summary) Even if a speaker uses a speaker-dependent speech recognition system, speech recognition performance varies. For this reason, speech quality is varied by some factors, including emotion, background noise, and so on, even though the speaker and utterance remain constant. However, the relationships between intra-speaker's speech variability and speech recognition performance are not clear. Hence, we focus on the intra-speaker's speech variability which affects the speech recognition performances. To investigate these relationships, we have been collecting speech data since November 2002. Using a part of the speech corpus, we conducted speech recognition experiments. In this paper, we analyze the relationships between intra-speaker's speech variability and the phoneme accuracy by using the correlation analysis. For factors of the correlation analysis, we use a number of errors, a speaking rate, a likelihood. Analysis results show a strong correlation between the number of the substitution errors and the phoneme accuracy although the correlations of the number of the deletion and the insertion errors are low. Therefore, it is considered that there are overlaps between phonemes since the feature parameters vary at each speaking rate. For improving the phoneme accuracy, it is needed that we study a method which discriminates phonemes. On the other hand, although the correlation between the phoneme accuracy and the speaking rate seems to be low, a strong correlation between the speaking rate and the number of deletion errors and insertion errors are found. Since the number of the insertion errors and the number of the deletion errors were in the counterbalance relation, the correlation between the speaking rate and the phoneme accuracy was low. However, we consider that it is needed to normalize the speaking rate because the speaking rate influences on the number of the deletion and the insertion errors.
117.	Satoru Tsuge, Masami Shishibori, Kenji Kita, Fuji Ren and Shingo Kuroiwa : Study of Intra-Speaker's Speech Variability over Long and Short Time Periods for Speech Recognition, Proc. of 2006 IEEE International Conference on Acoustic, Speech, and Signal Processing (ICASSP2006), Vol.1, 397-400, Toulouse, France, May 2006. (Summary) In this paper, we describe a Japanese speech corpus collected for investigating the speech variability of a specific speaker over short and long time periods and then report the variability of speech recognition performance over short and long time periods. Although speakers use a speaker-dependent speech recognition system, it is known that speech recognition performance varies pending when the utterance was uttered. This is because speech quality varies by occasion even if the speaker and utterance remain constant. However, the relationships between intra-speaker speech variability and speech recognition performance are not clear. Hence, we have been collecting speech data to investigate these relationships since November 2002. In this paper, we introduce our speech corpus and report speech recognition experiments using our corpus. Experimental results show that the variability of recognition performance over different days is larger than variability of recognition performance within a day. (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-33947625663 (Elsevier: Scopus)
118.	Kazumori Matsumoto, Masaru Sugano, Masaki Naito, Keiichiro Hoashi, Haruhisa Kato, Masami Shishibori, Kenji Kita, Fumiaki Sugaya and Yasuyuki Nakajima : Shot Boundary Detection and Low-Level Feature Extraction Experiments for TRECVID 2005, Proceedings of TREC Retrieval Evaluation 2005 (TRECVID2005), 155-167, Washington, D.C., Mar. 2006.
119.	Kazumori Matsumoto, Keiichiro Hoashi, Masaki Naito, Masami Shishibori and Kenji Kita : Report on TRECVID 2005, Proceedings of 12th Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV2006), 65-70, Tokushima, Feb. 2006.
120.	Nobuki Nishikawa, Masami Shishibori, Satoru Tsuge and Kenji Kita : A Method to Retrieve Video Scenes based on the Distance of Character Image Features between Query and Telop, Proceedings of 12th Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV2006), 211-216, Tokushima, Feb. 2006. (Summary) In this paper, a new video scene retrieval method based on telop characters is proposed. We pay attention to the kind of telop characters to be recognized for the video scene retrieval method. In order to specify the suitable scene, we have to recognize the only corresponding telop characters to the query not all characters in the video data. This method compares between each image features of telop characters and template image features of query.
121.	Masami Shishibori, Takayuki Minamimoto, Kazumori Matsumoto, Masaki Naito, Satoru Tsuge and Kenji Kita : Estimation of the Camera Motion based on Movement of Interest Points between Images, Proceedings of 12th Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV2006), 145-150, Tokushima, Feb. 2006. (Summary) In this paper, we propose the efficient algorithm to estimate the camera motion by using the movement information of interest points extracted from each frame image in a video. First, some interest points are extracted from the frame image by using the Harris interest operator, and then all correspondences of each interest point between two neighboring frames are decided. Based on these movement data of interest points, camera motion parameters (zoom, pan, tilt) are calculated. In the learning phase, ``shot CM vectors'' are generated based on the three CM parameters in a shot and input to the support vector machines. In the test phase, ``window CM vectors'' are generated within each window and input to the SVM to discriminate shots which contain the camera motion.
122.	Masami Shishibori, Daichi Koizumi, Kazuaki Ando, Satoru Tsuge and Kenji Kita : Feedback Retrieval using the Image Knowledge Database on WWW Image Retrieval Systems, Proceedings of 12th Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV2006), 205-210, Tokushima, Feb. 2006. (Summary) WWW image retrieval systems can retrieve the corresponding images to the query keyword from WWW, however every system cannot retrieve suitable images with high precision. In this paper, the feedback retrieval method using the image knowledge database is proposed. If the query keyword is not registered in the database, the user must select the suitable ones from image data retrieved by the conventional system, and then features of selected images are registered into the database as the supervised data.
123.	Daichi Koizumi, Masahiro Oonishi, Satoru Tsuge, Masami Shishibori and Kenji Kita : Automatic Meta-Data Annotation of Image Region, Proceedings of 2005 IEEE International Conference on Natural Language Processing and Knowledge Engineering (IEEE NLP-KE'05), 835-838, Wuhan, Oct. 2005. (Link to Publication Site) ● Publication site (DOI): 10.1109/NLPKE.2005.1598852 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-33847307648 (DOI: 10.1109/NLPKE.2005.1598852, Elsevier: Scopus)
124.	Masami Shishibori, Daichi Koizumi, Kazuaki Ando, Satoru Tsuge and Kenji Kita : Development of a WWW Image Retrieval System Using the Image Knowledge Database, Proceedings of 2005 IEEE International Conference on Natural Language Processing and Knowledge Engineering (IEEE NLP-KE'05), 506-510, Wuhan, Oct. 2005. (Summary) In this paper, a new WWW image retrieval system using the image knowledge database is proposed. This system can show more suitable images by filtering retrieval results of the conventional system. If the query keyword is not registered in the database, the user must select the suitable ones from image data retrieved by the conventional system, and then features of selected images are registered into the database as the supervised data. If the query keyword is the registered one, more similar images to the supervised data in the database can be indicated in the top order. The experimental results show that the average precision of this system becomes 11.6 % better than the conventional system. (Link to Publication Site) ● Publication site (DOI): 10.1109/NLPKE.2005.1598790 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-33847298544 (DOI: 10.1109/NLPKE.2005.1598790, Elsevier: Scopus)
125.	Satoru Tsuge, Masami Shishibori, Fuji Ren, Kenji Kita and Shingo Kuroiwa : Data Collection for Investigating Speech Variability in a Specific Speaker Over Long and Short Time Periods, Proceedings of 2005 IEEE International Conference on Natural Language Processing and Knowledge Engineering (IEEE NLP-KE'05), 152-157, Wuhan, Oct. 2005. (Summary) In this paper, we describe a Japanese speech corpus collected for investigating the speech variability of a specific speaker over short and long time periods. Although speakers use a speaker-dependent speech recognition system, it is known that speech recognition performance varies pending when the utterance was uttered. This is because speech varies even if the speaker utters a specific sentence. However, the relationship between intra-speaker speech variability and speech recognition performance is not clear. We have not seen a corpus of Japanese speech data of a specific speaker over a long time period. Hence, since 2002, we have been collecting speech data for investigating the relationships between speech variability and speech recognition performance. In this paper, we introduce our speech corpus and conduct speech recognition experiments. Experimental results show that the variability of recognition performance over different days is larger than variability of recognition performance within a day. (Link to Publication Site) ● Publication site (DOI): 10.1109/NLPKE.2005.1598725 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-33847334792 (DOI: 10.1109/NLPKE.2005.1598725, Elsevier: Scopus)
126.	Satoru Tsuge, Masami Shishibori, Fuji Ren, Kenji Kita and Shingo Kuroiwa : Frequency Characteristic Normalization Method using Blind Equalization Technique with Multiple References for DSR, Proceedings of 10th International Conference SPEECH and COMPUTER (SPECOM2005), Vol.1, 103-106, Patras, Greece, Oct. 2005. (Summary) In this paper, we propose a blind equalization method with multiple references for ETSI standard Distributed Speech Recognition (DSR) front-end. The proposed method compensates for acoustic mismatch caused by input devices. In ETSI advanced DSR front-end, the blind equalization method is introduced to compensate for acoustic mismatch. This method estimates the bias, which compensates for the mismatch, using one reference vector. If the input speech is short or contains many similar phonemes, there is concern that this method might not estimate the accurate bias. The proposed method estimates the bias using multiple references instead of one reference. Using multiple references, the proposed method estimates the bias more accurately. In addition, we represent the references by combining the VQ centroids used in the data compression process of ETSI standard DSR front-end. This limits increases in memory size and computation costs on the front-end. As a result, we can carry out this method in real-time on the DSR front-end. Experimental results on a Japanese newspaper dictation task indicate that the proposed method gave better performance under acoustic mismatched conditions than the conventional blind equalization method.
127.	Xin Luo, Shiro Ajioka, Masami Shishibori and Kenji Kita : Image Feedback Retrieval Based on Vector Space Model Transformation, Information Retrieval Technology, 616-625, Oct. 2005. (Link to Publication Site) ● Publication site (DOI): 10.1007/11562382_58 (Link to Search Site for Scientific Articles) ● Search Scopus @ Elsevier (DOI): 10.1007/11562382_58 (DOI: 10.1007/11562382_58)
128.	Satoru Tsuge, Shingo Kuroiwa, Masami Shishibori, Kenji Kita and Fuji Ren : Evaluation of frequency characteristic normalization method with multiple reference cepstrum on the Japanese newspaper article sentences speech corpus, Proceedings of the third International Conference on Information, 199-202, Tokyo, Nov. 2004. (Summary) 本論文では，分散型音声認識システムにおける入力系の周波数特性の違いによる音声認識精度の低下を抑制するため，複数の参照ケプストラムを用いた周波数特性正規化手法を提案した．提案手法は，フレーム同期で駆動し，各フレームで最近の参照ケプストラムを探し，周波数特性を正規化するバイアスを計算する手法である．提案手法の有効性を検証するため日本語連続音声認識実験を行った結果，提案手法は入力系の周波数特性が変動しても，音声認識精度の低下は少なく，有効であることがわかった．
129.	Satoru Tsuge, Shingo Kuroiwa, Kazuma Hara and Kenji Kita : Evaluation of ETSI Advanced Front-end and Bias Removal Method on the Japanese Newspaper Article, Proceedings of EUROSPEECH2003, 2145-2148, Geneva, Sep. 2003. (Summary) 本論文では，分散型音声認識システムにおける入力系の周波数特性の違いによる音声認識精度の低下を抑制するため，入力系の周波数特性を正規化する手法を提案した．提案手法は，実時間で駆動し，各フレームで最近の平均音素パラメータを探し，周波数特性を正規化するバイアスを計算する手法である．提案手法の有効性を検証するため日本語連続音声認識実験を行った結果，提案手法は入力系の周波数特性が変動しても，音声認識精度の低下は少なく，有効であることがわかった．
130.	Li Taihao, Fuji Ren, Shingo Kuroiwa and Kenji Kita : To define the feature function in extracting Japanese-Chinese bilingual word pairs using maximum entropy modeling, IEEE International Conference on Systems, Man and Cybernetics, Vol.3, 291-294, Tunisia, Oct. 2002. (Summary) In this paper, we present an approach to construct a translation dictionary used in multilingual natural language processing such as machine translation. We use a learning and extracting method of bilingual word pairs from parallel corpora using the maximum entropy modeling. We have conducted the experiments in Japanese and Chinese parallel corpora. (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-0038345949 (Elsevier: Scopus)
131.	Tai Xiaoying, Fuji Ren and Kenji Kita : Long-term Relevance Feedback using Simple PCA and Linear Transformation, 13th International conference and workshop on Database and Expert Systems Applications, 261-265, France, Sep. 2002. (Summary) This paper proposes a new method to improve information retrieval performance of the vector space model by preserving user-supplied relevance information for a long term in the system.
132.	Satoru Tsuge, Shingo Kuroiwa, Masami Shishibori, Fuji Ren and Kenji Kita : Robust Feature Extraction in a Variety of Input Devices on the Basis of ETSI Standard DSR Front-end, 7th International Conference on Spoken Language Processing, 2221-2224, Denver, USA, Sep. 2002. (Summary) This paper reports an evaluation ETSI standard DSR (Distributed Speech Recognition) front-end through continuous word recognition on a Japanese speech corpus and proposes a method, the basis removal method, that reduces the distortion between feature vector and VQ codebook.
133.	Daichi Koizumi, Satoru Tsuge, Masami Shishibori and Kenji Kita : Development of Image Keyword Registration System for Cross-media Information Retrieval between Texts and Images, International Conference on Information-2002, Series of Information & Management Sciences, Vol.3, 166-171, Beijing, Jul. 2002. (Summary) It is difficult to give keywords to huge image databases by human hands. In this paper, we introduce a keywords tagging support system, which has the following characteristics: (1) several users can work jointly through the network, (2) sensibility words and free descriptions can be attached in addition to system pre-defined keywords, (3) we can quickly and efficiently retrieve target images from the keyword.
134.	Shingo Kuroiwa, Satoru Tsuge, Hironori Tani, Masami Shishibori, Fuji Ren and Kenji Kita : Latent Semantic Indexing Based on Simple Principal Component Analysis, International Conference on Information-2002, Series of Information & Management Sciences, Vol.3, 172-177, Beijing, Jul. 2002. (Summary) In this paper, we propose to use the Simple Principal Component Analysis (SPCA) for dimensionality reduction of the vector space information retrieval model. The SPCA algorithm is a data-oriented fast method which does not require the computation of the variance-covariance matrix. In SPCA principal components are estimated iteratively so we also propose a criteria to determine the convergence. The optimum number of iterations for each principal component can be detemined using the criteria. Experimentally, we show that the SPCA-based method offers improvement over the conventional SVD-based method despite its small amount of computation.
135.	Satoru Tsuge, Shingo Kuroiwa, Masami Shishibori, Fuji Ren and Kenji Kita : An Evaluation of Japanese Speech Recognition Using ETSI Standard DSR Front-end, International Conference on Information-2002, Series of Information & Management Sciences, Vol.3, 372-376, Beijing, Jul. 2002. (Summary) This paper reports an evaluation of ETSI standard DSR (Distributed Speech Recognition) front-end through continuous word recognition on a Japanese speech corpus. Experimental results show that using non-quantized features in acoustic model training procedure can improve the recognition performance of DSR front-end features.
136.	Li Taihao, Fuji Ren and Kenji Kita : Extracting Japanese-Chinese Bilingual Word Paris with Maximum Entropy Model, Series of Information & Management Sciences, Vol.3, 33-37, Beijing, Jul. 2002. (Summary) In this paper,, we present an approach to construct a translation dictionary used in multilingual natural language processing such as machine translation. We use a learning and extracting method of bilingual word pairs from parallel corpora using the maximum entropy modeling. We are doing the experiment in Japanese and Chinese parallel corpora.
137.	Satoru Tsuge, Masami Shishibori, Shingo Kuroiwa and Kenji Kita : Dimensionality Reduction Using Non-negative Matrix Factorization for Information Retrieval, Natural Language Processing and Knowledge Engineering Mini Symposium, IEEE SYSTEMS, MAN, AND CYBERNETICS 2001 (NLPKE), 960-965, Tucson, Oct. 2001. (Summary) ベクトル空間情報検索モデルを用いた全文検索を行う場合，各文書を表現する索引語ベクトルがスパースになることが問題となる．そこで，本論文ではベクトル空間モデルの次元削減により情報検索精度を向上させる．次元削減には，非負行列である文書行列に対する次元削減手法として，Non-negative Matrix Factorization (NMF)を適用した．NMFは非負行列を2つの非負行列へ分解を行う手法である．提案手法は，NMFを用い非負行列である索引語文書行列を分解し，その一方の行列へ索引語文書行列を射影することにより，次元削減を行う手法である．MEDLINEを用いた情報検索実験で提案手法の有効性を示した．また，この検証結果より同一次元では従来の手法SVDを用いた手法と同等の検索精度を示すことがわかった． (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-0035725021 (Elsevier: Scopus)
138.	Fuji Ren, Li Shigang and Kenji Kita : Automatic Abstracting Important Sentences of Web Articles, IEEE International Conference on System, Man, and Cybernetics, Vol.3, 1705-1710, Tucson, Oct. 2001. (Summary) Being increasingly popular, the Internet greatly changes our live. We can conveniently receive and send information via the Internet. With the information explosion in Web, it is becoming crucial to develop means to automatically extract important sentences from the Web articles. In this paper, we propose a method which uses both statistical and structural information in sentence extraction. In addition, following the analysis of human's extractions, several heuristic rules are added to filter out non-important sentences and to prevent similar sentences from being extracted. Our experimental results proved the effectiveness of these means. In particular, once the heuristic rules being added, a significant improvement has been observed. (Link to Publication Site) ● Publication site (DOI): 10.1109/ICSMC.2001.973531 (Link to Search Site for Scientific Articles) ● Summary page in Scopus @ Elsevier: 2-s2.0-0035720436 (DOI: 10.1109/ICSMC.2001.973531, Elsevier: Scopus)
139.	Minoru Sasaki and Kenji Kita : Information Retrieval System Using Concept Projection based on PDDP Algorithm, Proceedings of PACLING2001 (Pacific Association for Computational Linguistics 2001), 243-249, Fukuoka, Oct. 2001. (Summary) In the vector space information retrieval model, text documents are often represented as high-dimensional and sparse vectors. In this paper, we propose to use the technique of dimensionality reduction using concept vectors based on PDDP algorithm as a way of solving these problems.
140.	Shingo Kuroiwa, Satoru Tsuge, Hironori Tani, Xiaoying Tai, Masami Shishibori and Kenji Kita : Dimensionality reduction of vector space model based on Simple PCA, Proc. Knowledge-Based Intelligent Information Engineering Systems & Allied Technologies (KES), Vol.2, 362-366, Osaka, Sep. 2001. (Summary) In this paper, we propose to use the Simple Principal Component Analysis (SPCA) for dimensionality reduction of the vector space information retrieval model. The SPCA algorithm is a data-oriented fast method which does not require the computation of the variance-covariance matrix. In SPCA principal components are estimated iteratively so we also propose a criteria to determine the convergence. The optimum number of iterations for each principal component can be detemined using the criteria.
141.	Satoru Tsuge, Masami Shishibori, Shingo Kuroiwa, Takashi Hirai and Kenji Kita : Dimensionality Reduction of Vector Space Information Retrieval Model based on Non-negative Matrix Factorization, Proc. Knowledge-Based Intelligent Information Engineering Systems & Allied Technologies (KES), 367-371, Osaka, Sep. 2001. (Summary) In this paper, we use Non-negative Matrix Factorization (NMF) for dimensionality reduction of the vector space model. The NMF decomposes a non-negative matrix, i.e., term-by-document matrix, into two non-negative matrices. One of the decomposed matrix can be regarded as the basis vectors. The dimensionality reduction can be performed by projecting the document vectors onto the lower dimensional space which is formed by these basis vectors.
142.	Satoru Tsuge, Masami Shishibori, Shingo Kuroiwa, Yasuhito Tanaka, Takashi Hirai, Ryu Okamoto and Kenji Kita : Relevance Feedback with Support Vector Machine for Information Retrieval, Proceedings of International Conference on Computer Processing of Oriental Languages (ICCPOL), 35-40, Seoul, May 2001. (Summary) Relevance feedback is a technique of modifying the query to improve retrieval performance based on relevance judgements from the user. We propose a relevance feedback method using support vector machine. Experimental results on Japanese test collection show that the proposed method is useful feedback method compared to the conventional feedback method.
143.	Minoru Sasaki and Kenji Kita : Vector Space Information Retrieval Using Concept Projection, Proceedings of the 19th International Conference on Computer Processing of Oriental Languages, 73-76, Seoul, May 2001. (Summary) Text documents are often represented as high-dimensional and sparse vectors using words as features in a multidimensional space. These vectors require a large number of computer resources and it is difficult to capture underlying concepts referred to by the terms. In this paper, we propose to use the technique of random projection for dimensional space reduction of vector space information retrieval model as a way of solving these problems.
144.	Yasuhito Tanaka, Takashi Hirai and Kenji Kita : 以媒体概念作為基礎的多媒体教育課程的考察(以 CD-ROM 為基礎的教育), Proceedings of the Second International Conference on New Technologies in Teaching and Learning Chinese, 147-151, Guilin, China, Dec. 2000.
145.	Yasuhito Tanaka, Takashi Hirai and Kenji Kita : 以媒体概念作為基礎的多媒体教育課程的考察(以 CD-ROM 為基礎的教育), Proceedings of the Second International Conference on New Technologies in Teaching and Learning Chinese, 147-151, Guilin, Dec. 2000.
146.	Satoru Tsuge, Toshiaki Fukada and Kenji Kita : Frame-Period Adaptation for Speaking Rate Robust Speech Recognition, Proceedings of the 6th International Conference on Spoken Language Processing, 718-721, Beijing, Oct. 2000. (Summary) This paper describes a frame-period adaptation method for speaking rate robust speech recognition. The proposed method determines an appropriate frame-period for each phrase by measuring its speaking rate or computing the acoustic likelihood with a set of frame-periods. Experimental results on spontaneous speech recognition show that the proposed method is effective for slower utterance.
147.	Xiaoying Tai, Minoru Sasaki, Yasuhito Tanaka and Kenji Kita : Improvement of Vector Space Information Retrieval Model based on Supervised Learning, Proceedings of the Fifth International Workshop on Information Retrieval with Asian Languages, 69-74, Hong Kong, Sep. 2000. (Summary) This paper proposes a method to improve retrieval performance of the vector space model by utilizing usre-supplied information of those documents that are relevant to the query in question.
148.	Xiaoying Tai and Kenji Kita : Vector Space Information Retrieval Model Using Supervised Learning, 7th Joint International Computer Conference, 442-448, China, 2000. (Summary) In this paper, a method is proposed to improve retrieval performance based on the vector space model by using the information about that which documents are right answers to the small number of questions.
149.	Yasuhito Tanaka and Kenji Kita : Multilingual Parallel Corpus of Major East Asian Languages, Proceedings of the MAL'99 (Multi-lingual Information Processing and Asian Language Processing), 125-134, Beijing, Nov. 1999. (Summary) This paper describes the development of a multilingual parallel corpus of major east Asian languages, including Japanese, Chinese, Korean, as well as English. The multilingual corpus here not only contains the same texts in more than one language, but also has sentence-level language-to-language correspondence, as well as allowing easy computer processing.
150.	Yasuhito Tanaka and Kenji Kita : JCKE Multilingual Corpus of Major Asian Languages, Proceedings of the Fifth International Congress on Terminology and Knowledge Engineering, 660-670, Innsbruck, Aug. 1999. (Summary) This discusses the development of a multilingual corpus of major Asian languages, including Japanese, Chinese, Korean, as well as English. The multilingual corpus here not only contains the same texts in more than one language, but also has sentence-level language-to-language correspondence, as well as allowing easy computer processing.
151.	Xiaoying Tai, Yasuhiko Kato and Kenji Kita : Automatically Compiling Multilingual Translations from the World Wide Web, Proceedings of the International Conference on Machine Translation and Computer Language Information Processing, 516-521, Beijing, Jun. 1999. (Summary) The World Wide Web provides almost unlimited accesses to the textual documents and it also contains parallel pages in many languages. In this paper, we focus on the problem of automatically compiling multilingual translations from the Web.
152.	Hiroki Oda and Kenji Kita : A Character-Based Japanese Word Segmenter Using PPM^-Based Langauge Model, Proceedings of the 18th International Conference on Computer Processing of Oriental Languages, 527-532, Tokushima, Mar. 1999. (Summary) This paper presents a new word segmentation method, which is suitable for those languages that have no delimiters between words, such as Japanese and Chinese. To find the best word segmentation for the input sentence, we use a character-based n-gram model and a PPM-based language model as the statistical langugage model.
153.	Yasuhiko Kato and Kenji Kita : Modern Japanese Processing Problems --- Problems of "Kata-Kana" Appeared in Loan Words ---, Proceedings of the 18th International Conference on Computer Processing of Oriental Languages, 305-308, Tokushima, Mar. 1999. (Summary) This paper shows problems of processing modern Japanese text written in modern kana orthography, which are caused by Kana-dzukai (kana characeter usage) difference, especially kata-kana characters appeared in loan words from Western langugaes.
154.	Ysuhito Tanaka and Kenji Kita : JCKE Multilingual Corpus of Major Asian Languages, Proceedings of the First ASIALEX Regional Symposium, 296-305, Guangzhou, Jan. 1999. (Summary) This paper discusses the development of a multilingual corpus of major Asian languages, including Japanese, Chinese, Korean, as well as English. The multilingual corpus here not only contains the same texts in more than one language, but also has sentence-level language-to-language correspondence, as well as allowing easy computer processing.
155.	Minoru Sasaki and Kenji Kita : Automatic Text Categorization based on Hierarchical Rules, Proceedings of IIZUKA'98, 935-938, Fukuoka, Oct. 1998. (Summary) In this paper, we propose an efficient method for automatic text categorization based on the rule learning algorithm.
156.	Minoru Sasaki and Kenji Kita : Rule-Based Text Categorization Using Hierarchical Categories, Proceedings of the 1998 IEEE International Conference on Systems, Man and Cybernetics, 2827-2830, San Diego, Oct. 1998. (Summary) In this paper, we point out the problem of automatic document categorization by using the rule learning algorithm RIPPER. We extend the algorithm to construct a set of hierarchical rules automatically besides a set of rules RIPPER constructs.
157.	Kenji Kita, Minoru Sasaki and Xiaoying Tai : Rule-Based Hierarchical Document Categorization for the World Wide Web, Web Technologies and Applications: Proceedings of Asia Pacific Web Conference (APWeb98), 269-273, Beijing, Sep. 1998. (Summary) Document categorization , which is defined as the classification of text documents into one of several fixed classes or categories, has become important with the explosive growth of the World Wide Web. In this paper, we propose an efficient method for hierarchical document categorization based on the rule learning algorithm. We conducted experiments using actual Web page documents and obtained encouraging results.
158.	Yasuhito Tanaka and Kenji Kita : Machine-Readable Dictionary Headwords, Proceedings of the 12th Pacific Asia Conference on Language, Information and Computation, 310-315, Singapore, Feb. 1998. (Summary) Selecting headwords is an important problem in developing a dictionary. This paper discusses how we collect, sort and allocate information to the dictionary headwords.
159.	Kenji Kita, Zhifang Sui and Shiwen Yu : 基於概率模型的語言的集類方法: 根据多語種語料庫進行語言系統樹的再構造, JSCL-97, 109-114, Beijing, Aug. 1997.
160.	Kenji Kita, Yoshikazu Fukui, Masaaki Nagata and Tsuyoshi Morimoto : Automatic Acquisition of Probabilistic Dialogue Models, International Conference on Spoken Language Processing, 196-199, Oct. 1996.
161.	Kenji Kita, Yoshikazu Fukui, Masaaki Nagata and Tsuyoshi Morimoto : Dialogue Knowledge Acquisition from Annotated Corpora, IEEE International Conference on Systems, Man and Cybernetics, 556-561, Beijing, Oct. 1996. (Summary) In this paper, we automatically deduce dialogue structures from a corpus with probabilistic methods. Each utterance in the corpus is annotated with a speaker label and an utterance type called IFT (Illocutionary Force Type). We use an Ergodic HMM (Hidden Markov Models) and the ALERGIA algorith, an algorithm for learning probabilistic automata by means of state merging, to model the speaker-IFT sequence.
162.	Kenji Kita and Minoru Sasaki : Improvement of a Probabilistic CFG Using a Cluter-Based Language Modeling Technique, 4th International Conference on Soft Computing, 929-932, Oct. 1996. (Summary) This paper proposes an improved probabilistic CFG using a cluster-based language modeling technique. The basic idea of this model involves clustering a training corpus into a number of subcorpora, and then estimating separate probability distribution from each subcorpora.
163.	Kenji Kita and Minoru Sasaki : Automatic Acquisition of Probabilistic Dialogue Models, 4th International Conference on Soft Computing, 925-928, Oct. 1996. (Summary) In this paper, we automatically deduce dialogue structures from a corpus with probabilistic methods. Each utterance in the corpus is annotated with a speaker label and an utterance type called IFT. We use an Ergodic HMM and the ALERGIA algorithm, an algorithm for learning probabilistic automata by means of state merging, to model speaker-IFT sequences.
164.	Kenji Kita and Tatsuya Iwasa : Probabilistic Language Modeling Based on Mixture Probabilistic Context-Free Grammar, ROCLING VIII, 127-136, Taipei, Aug. 1995. (Summary) This paper proposes an improved probabilistic CFG, called mixture probabilistic CFG, based on an idea of cluster-based language modeling. The basic idea of this model involves clustering a training corpus into a number of subcorpora, and then training probabilistic CFGs from these subcorpora. The resulting probabilistic CFGs become context- or topic-dependent, and thus accurate language modeling would be possible.
165.	Tatsuya Iwasa and Kenji Kita : Error Correction of Speech Recognition Outputs Using Generalized LR Parsing and Confusion Matrix, ROCLING VIII, 101-110, Taipei, Jul. 1995. (Summary) This paper describes a method for correcting misrecognition results derived from a speech recognizer. Our method is based on generalized LR parsing and uses a confusion matrix in order to select the best sentence out of multiple candidates.
166.	Kenji Kita, Kazuhiko Ashibe, Yoneo Yano and Hiroaki Ogata : VOICEDIC: A Practical Application of Speech Recognition Technology, Sixth International Conference on Human-Computer Interaction, 535-540, Yokohama, Japan, Jul. 1995. (Summary) This paper presents a practical application of speech recognition technology, a voice-accessible or voice-activated dictionary VOICEDIC. The paper shows an outline of VOICEDIC as well as its advantages.
167.	Kenji Kita, Yoneo Yano and Tsuyoshi Morimoto : Continuous Speech Recognition Based on LR-Parser-Driven One-Pass Search Algorithm and Hidden Markov Modeling of Speech: Towards Real-Time Intelligent Human-Computer Interaction by Voice, SPICIS '94, B347-B352, Singapore, Nov. 1994. (Summary) This paper presents a novel continuous speech recognition algorithm that integrates three major technologies: (1) hidden Markov modeling for speech, (2) a generalized LR parser for handling context-free grammar (CFG) constraints, and (3) the one-pass search algorithm for achieving an efficient search. We implemented an experimental Japanese speech recognition system based on the proposed algorithm and conducted a comparison between our algorithm and other methods.
168.	Kenji Kita, Yoneo Yano and Tsuyoshi Morimoto : One-Pass Continuous Speech Recognition Directed by Generalized LR Parsing, 1994 International Conference on Spoken Language Processing, 13-16, Yokohama, Sep. 1994. (Summary) This paper formulates a novel continuous speech recognition algorithm that integrates a generalized LR parser for handling context-free grammar (CFG) constraints, and the one-pass search algorithm for achieving an efficient search. Our algorithm enables us to efficiently find an optimal hypothesis for a given speech signal according to a specified CFG in a frame-synchronous process.
169.	Kenji Kita, Takashi Omoto, Yoneo Yano and Yasuhiko Kato : Application of Corpora in Second Language Learning, --- The Problem of Collocational Knowledge Acquisition ---, Second Annual Work Shop on Very Large Corpora, 43-56, Kyoto, Aug. 1994. (Summary) Second language learning offers one interesting potential application of corpus-based natural language processing. This paper is primarily concerned with the acquisition of collocational knowledge from corpora for use in language learning. First we discuss the importance of collocational knowledge in second langugae learning, and then take up two measures, mutual information and cost criteria, for automatically identifying or extracting collocations from corpora.
170.	Kenji Kita, Kazuhiko Ashibe and Yoneo Yano : An Unknown Word Model for Unrestricted Japanese Texts Based on Generalized Mutual Information Statistics of Characters, Neural Nets and Soft Computing, 217-218, Iizuka. Japan, Aug. 1994. (Summary) In order to handle unrestricted naturally occuring texts, we have to cope with the unknown word problem. In this paper, we present an unknown word model based on generalized mutual information statistics.
171.	Kenji Kita, Kazuhiko Ashibe and Yoneo Yano : Natural Language Speech Recognition Using Finite-State Approximation for Context-Free Grammars, Neural Nets and Soft Computing, 537-538, Iizuka. Japan, Aug. 1994. (Summary) This paper first presents a continuous speech recognition system using an LR state transition table, which is a finite-state approximation for a context-free grammar. Then, we propose a forwrad-FSA backward-CFG search algorithm for accurate continuous speech recognition.
172.	Kenji Kita, Kazuhiko Ashibe and Yoneo Yano : CFG-Based One Stage Dynamic Programming Algorithm for Connected Word Recognition, Neural Nets and Soft Computing, 535-536, Iizuka. Japan, Aug. 1994. (Summary) This paper presents a CFG(context-free grammar)-based one-stage dynamic programming algorithm for the problem of connected word recognition. This algorithm enables us to efficiently find an optimal hypothesis for a given speech signal according to a specified CFG in a frame-synchronous process.
173.	Kenji Kita, Yasuhiko Kato, Takashi Omoto and Yoneo Yano : Automatically Extracting Collocations from Corpora for Language Learning, International Conference on Teaching and Language Corpora, 53-64, Lancaster, U.K., Apr. 1994. (Summary) This paper takes up a collocational knowledge acquisition problem, and discuss the following points: (1) Why is collocational knowledge so important in language learning? (2) What measures can we use to automatically extract collocations from corpora? and (3) What characteristics does each measure have?
174.	Kenji Kita, Toshihiro Hayashi and Yoneo Yano : Corpus-based Language Learning: Towards Prctical Language Learning Systems, ICCE '93, 355-357, Taipei, Dec. 1993. (Summary) This paper proposes a quite new paradigm for language learning: corpus-based language learning. Corpus-based language learning is a promising approach for the development of truly useful and practical language learning systems, because it compensates the domain-limitedness and the lack of sophisticated knowledge of the traditional AI-based approach, at least to a considerable extent. As the first step to corpus-based language learning, we focus on knowledge acquisition from corpora for language learning systems.
175.	Gen-ichiro Kikui, Mark Seligman, Toshiyuki Takezawa, Tsuyoshi Morimoto, Masami Suzuki, Kenji Kita, Masaaki Nagata, Toshihisa Tashiro, Herbert Tropf, Shigeki Sagayama, Jun-ichi Takami, Kazumi Ohkura and Akira Kurematsu : A Spoken Language Translation System: ASURA, 13th International Joint Conference on Artificial Intelligence, 1705, Chambery, France, Aug. 1993. (Summary) This paper introduces a speech-to-speech machine translation system ASURA, which translates Japanese speech into English and German speech.
176.	Shigeki Sagayama, Jun-ichi Takami, Akito Nagai, Harald Singer, Kouichi Yamaguchi, Kazumi Ohkura, Kenji Kita and Akira Kurematsu : ATREUS: A Speech Recognition Front-end for a Speech Translation System, Eurospcch'93, 1287-1290, Berlin, Germany, 1993. (Summary) This paper describes the continuous speech recognition subsystem ATREUS which is used as the speech input stage in the experimental speech translation system. The speech recognition algorithm is SSS-LR/VFS which consists of context-dependent phone models, a generalized LR parser, and vector field smoothing for speaker/environment adaptation.
177.	Kenji Kita, Tsuyoshi Morimoto, Kazumi Ohkura and Shigeki Sagayama : Continuously Spoken Sentence Recognition by HMM-LR, International Conference on Spoken Language Processing, 305-308, Oct. 1992. (Summary) This paper describes recent efforts to improve the HMM-LR speech recognition system for continuously spoken sentences. These improvements include: (1) HMM training with continuous utterance as well as word utterance, (2) development of a sentential grammar, and (3) introduction of stochastic language models. The stochastic language models investigated were a stochastic CFG and a bigram model of production rules. The recognition experiments demonstrated that the rule-bigram model is much superior to stochastic CFG.
178.	Yamaguchi Kouichi, Shigeki Sagayama, Kenji Kita and Frank H. Soong : Continuous Mixture HMM-LR Using the A^* Algorithm for Continuous Speech Recognition, International Conference on Spoken Language Processing, 301-304, Oct. 1992. (Summary) This paper describes a new continuous mixture HMM version of the HMM-LR continuous speech recognition system. We also describe two new search methods; an A* algorithm based and a hybrid best-first search. These two new algorithms are experimentally compared with conventional beam search techniques. The A* algorithm based search has been found effective in reducing the number of phoneme verification by half with little degradation of recognition performance.
179.	Akito Nagai, Kenji Kita, Toshiyuki Hanazawa, Tadashi Suzuki, Tomohiro Iwasaki, Takeshi Kawabata, Kunio Nakajima, Kiyohiro Shikano, Tsuyoshi Morimoto, Shigeki Sagayama and Akira Kurematsu : Hardware Implementation of Realtime 1000-Word HMM-LR Continuous Speech Recognition, International Conference on Spoken Language Processing, 237-240, Oct. 1992. (Summary) This paper describes the design and development of the architecture of dedicated hardware which recognizes continuous speech sentence using a 1000-word vocabulary. It is based on the HMM-LR mechanism, which is an integration system of speech recognition and language analysis. The paper gives performance data with results within a pipeline delay of approximately 2 or 3 seconds, independently of the length of the sentence.
180.	Akito Nagai, Shigeki Sagayama and Kenji Kita : Poneme-Context-Dependent LR Parsing Algorithms for HMM-Based Continuous Speech Recognition, Second European Conference on Speech Communication and Technology, 1397-1400, Sep. 1991. (Summary) In continuous speech recognition, the phoneme pattern variations are highly correlated with the phoneme environment. Since, the phoneme context is the most significant among the phoneme environment factors, it is advantageous in improvement of speech recognition accuracy to use phoneme-context-dependent phone models. This paper discusses two approaches for combining an efficient LR parser and phoneme-context-dependent HMM phone models and compares them through continuous speech recognition experiments.
181.	Toshiyuki Takezawa, Kenji Kita, Junko Hosaka and Tsuyoshi Morimoto : Linguistic Constraints for Continuous Speech Recognition in Goal-Directed Dialogue, IEEE International Conference on Acoustics, Speech and Signal Processing, 801-804, Toronto, Canada, May 1991. (Summary) This paper discusses how to apply langugae models to continuous speech recognition, and what kinds of linguistic constraints are really important for improving speech recgnition accuracy. We conducted an experiment to see how inter-phrase grammar contributes to the improvement of sentence speech recognition rates. We also describe what kinds of constraints should, or can be, used to suppress each remaining error by referring to the actual example in the experiment.
182.	Kenji Kita and Wayne H. Ward : Incorporating LR Parsing into Sphinx, IEEE International Conference on Acoustics, Speech and Signal Processing, 269-272, Toronto, May 1991. (Summary) This paper describes the integration of an LR natural language parser with the Sphinx speech recognition system developed at CMU. We have incorporated LR pasing into Sphinx to provide more global grammatical constraints.
183.	Kenji Kita, Terumasa Ehara and Tsuyoshi Morimoto : Processing Unknown Words in Continuous Speech Recognition, International Workshop on Parsing Technologies, 136-142, Cancun, Mexico, Feb. 1991. (Summary) This paper proposes a continuous speech recognition method which accepts any utterance that might include unknown words. In this method, words not in the lexicon are transcribed as phoneme sequences, while words in the lexicon are recognized correctly.
184.	Kenji Kita, Toshiyuki Takezawa, Junko Hosaka, Terumasa Ehara and Tsuyoshi Morimoto : Continuous Speech Recognition Using Two-Level LR Parsing, International Conference on Spoken Language Processing, 905-908, Kobe, Japan, Nov. 1990. (Summary) This paper describes a continuous speech recognition system using two-level predictive LR parsing. In Japanese, a sentence is composed of several phrases, thus two kinds of grammars, namely an intra-phrase grammar and an inter-phrase grammar are sufficient for recognizing sentences. Two-level predictive LR parsing makes it possible to use not only intra-phrase grammatical constraints but also inter-phrase grammatical constraints during speech recognition.
185.	Kenji Kita, Takeshi Kawabata and Toshiyuki Hanazawa : HMM Continuous Speech Recognition Using Stochastic Language Models, IEEE International Conference on Acoustics, Speech and Signal Processing, 581-584, Albuquerque, Apr. 1990. (Summary) In this paper, the following three stochastic language models are investigated for an HMM continuous speech recognition system. (1) Trigram model of Japanese syllables, (2) Stochastic shift/reduce model in LR parsing, and (3) Trigram model of context-free rewriting rules.
186.	Toshiyuki Hanazawa, Kenji Kita, Satoshi Nakamura, Takeshi Kawabata and Kiyohiro Shikano : ATR HMM-LR Continuous Speech Recognition System, IEEE International Conference on Acoustics, Speech and Signal Processing, 53-56, Albuquerque, Apr. 1990. (Summary) This paper describes an improvement of the HMM-LR continuous speech recognizer using multiple codebooks, HMM state duration control and fuzzy vector quantization.
187.	Kenji Kita, Takeshi Kawabata and Hiroaki Saito : Parsing Continuous Speech by HMM-LR Method, International Workshop on Parsing Technologies, 126-131, Pittsburgh, Aug. 1989. (Summary) This paper describes a speech parsing method called HMM-LR. In HMM-LR, an LR parsing table is used to predict phones in speech input, and the system drives an HMM-based speech recognizer directly without any intervening structures such as a phone lattice.
188.	Kenji Kita, Takeshi Kawabata and Hiroaki Saito : HMM Continuous Speech Recognition Using Predictive LR Parsing, IEEE International Conference on Acoustics, Speech and Signal Processing, 703-706, Glasgow, England, May 1989. (Summary) This paper proposes a new continuous speech recognition method using an efficient parsing mechanism, an LR parser, driving HMM modules directly without any intervening structures such as a phoneme lattice. Accurate and efficient speech parsing is achieved by combining HMM and LR parsing.
189.	Kiyohiro Shikano, Takeshi Kawabata, Alex Waibel, Kaichiro Hatazaki, Hidefumi Sawai, Satoshi Nakamura, Toshiyuki Hanazawa, Kenji Kita and Akira Kurematsu : Speech Recognition Research at ATR, The Second Symposium on Advanced Man-Machine Interface through Spoken Language, Hawaii, Nov. 1988. (Summary) Speech recognition research activities in ATR Interpreting Telephony Research Laboratories are briefly described. Hidden Markov phoneme models have been improved and successfully applied to Japanese phrase utterance recognition combining with the LR predictive parser.

Proceeding of Domestic Conference:

1.	田中隆太, 新井広幸, Kohei Torii, Kenji Kita, 小林健一郎 and 阪本貴司 : Implementation of Artificial Intelligence for identifying implant types, 第54回公益社団法人日本口腔インプラント学会学術大会, Nov. 2024. (Keyword) dental care / medical image / artificial intelligence
2.	Jinsei Miyoshi, Kenji Kita, Goel Ajay and Tetsuji Takayama : <シンポジウム>リキッドバイオプシーによる食道扁平上皮癌の診断~AI を用いたmicroRNA 診断モデルの構築~., 第31回日本がん予防学会総会(日本がん予防学会学術大会2024徳島), Sep. 2024.
3.	Haruto Uda, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Investigation on Accuracy Improvement of Emotion Classification Based on Text Data, 第37回人工知能学会全国大会講演論文集, 1-4, Jun. 2023. (Summary) Recently, SNSs have facilitated the collection of a wide variety of text data. However, SNS text data has problems such as short sentences with abbreviations and colloquial expressions, which make labeling difficult, and the difficulty of collecting a large amount of data in a short period of time. To solve this problem, data expansion is an effective method for efficiently preparing large-scale, high-quality labeled text data for machine learning. In this research, we aim to improve the learning accuracy of sentiment classification by extending the data to Japanese texts. EDA (Easy Data Augmentation) was used as the data expansion method. In particular, the use of various models for text manipulation in the EDA increased the range of data expansion. The expanded text generated by data expansion was evaluated based on the semantic similarity and the degree of textual change. The optimal data for training was selected by determining a threshold value. The WRIME corpus was used as the dataset to ensure the reliability of the labels. In this presentation, we report the results of learning accuracy in sentiment classification using data expansion. (Keyword) data augmentation / easy data augmentation / sentiment analysis
4.	Kazuyuki Matsumoto, Ryota Kishima, Minoru Yoshida and Kenji Kita : コロナ禍前後におけるTwitterユーザの性格別感情変化の分析, 言語処理学会第29回年次大会発表論文集, 2365-2369, Mar. 2023. (Summary) Three years have passed since covid-19 pandemic began at the end of 2019, and the situation is changing daily as restrictions on behavior are gradually being eased. Even now, the dramatic changes in lifestyles continue to have a significant impact on people in terms of physical and mental stress. Our study analyzes the changes that occurred in people's statements, thoughts, and actions by conducting emotion analysis of the content of statements made on Twitter before and after corona virus pandemic, extracting characteristic expressions, and conducting a comparative analysis. In addition, we considered the changes in emotions by user personality type. (Keyword) social media / emotion estimation / COVID-19 virus pandemic / MBTI
5.	Hirotaka Hayamichi, Kazuyuki Matsumoto, Minoru Yoshida, Ryota Nishimura and Kenji Kita : Development of health maintenance support system using spoken dialogue robots, HAI シンポジウム2023 講演論文集, Mar. 2023. (Summary) Lifestyle-related diseases account for 30% of all medical costs and 60% of all deaths. It is important to maintain good lifestyle habits to protect one's health. In recent years, an increasing number of services and devices allow voice input. Speech recognition makes it easier to use computers. In our study, we develop a system that helps people improve their lifestyle habits by using a web application that analyzes the content of conversations with an interactive robot and suggests lifestyle habits and other information. Using the developed system, we actually collected lifestyle habits, analyzed them, and evaluated the system. (Keyword) health maintenance support system / spoken dialogue robots / lifestyle habits
6.	TATSUYA Miki, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Highlight Extraction in Stories Considering Character's Emotional Movements, HCGシンポジウム2022発表予稿集, C2-7, Dec. 2022. (Summary) In this study, we propose a method for extracting and summarizing important sentences in novels by quantifying the magnitude and changes of characters' emotions. We selected five novels from the Aozora Bunko as research subjects and performed automatic summarization of each novel using the proposed method. We created correct answer data that was manually extracted sentences that express characters' emotions and evaluated the proposed method by precision and average precision of the top 10 sentences (Keyword) natural language processing / Extractive Summarization / emotion
7.	Daiki Mori, Kazuyuki Matsumoto, 篠山学, Minoru Yoshida and Kenji Kita : Development of dialogue model reflecting individuality based on interview dialogue, 令和4年度電気・電子・情報関係学会四国支部連合大会講演論文集, 139, Sep. 2022. (Summary) 人間が AI とより親しみのあるコミュニケーションをするための要素の一つとして「個性」が挙げられる．AI に必要とされる個性は場面や状況に応じて異なるので，対話する相手に応じて個性を切り替えられるような仕組みが求められる．本研究では，日本語Transformer encoder-decoder対話モデルを使用し，インタビュー対話において異なる個性を持つ対話データごとにファインチューニングすることで対話モデルの個性に違いや特徴が出るかどうかを評価することを目的とする． (Keyword) dialogue system / dialogue model / interview dialogue
8.	TATSUYA Miki, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 感情辞書を用いた物語のハイライト抽出, 第19回テキストアナリティクス・シンポジウム, Sep. 2022.
9.	Reishi Amitani, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Affective analysis and visualization from posted text, replies, and images for analysis of buzz factors, 第21回情報科学技術フォーラム講演論文集(ハイブリッド開催), Vol.2, 63-68, Sep. 2022. (Summary) 本研究ではTwitter上で起こるバズ現象について，ツイート及び添付された画像，ツイートに対して寄せられたリプライの感性を分析することにより探る．Twitter に投稿されるテキストの文字数には制限があるが，テキストのみで表現することが難しい情報には画像を添付することで視覚的に伝えやすくなる．また，バズ現象を分析するという観点からも，画像付きのツイートはテキストのみのツイートよりも分析に使える情報量が増えるため，より詳細な分析が可能になると考えられる．ツイートの事後情報であるリプライテキストも分析に用いることができる．リプライは，ツイート投稿後に，そのツイートに対する反応として投稿されるものであるため，具体的なバズ現象の要因分析につながると考えた．本研究では，画像付きツイートについて，ツイートテキスト，画像，リプライテキストからそれぞれ抽出した感性情報を分析することで，バズ現象の要因分析を検討する. (Keyword) buzz factor / information diffusion / Twitter / affective analysis
10.	DAICHI Kawahara, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : コメントデータと歌詞に基づく楽曲動画検索システム, Proceedings of the 36th Annual Conference of the Japanese Society for Artificial Intelligence, 4O1-GS-4-05, Jun. 2022. (Summary) Currently, video sharing services such as YouTube are used by many people. In these services, keyword search is the mainstream. In the case of keyword search, it is difficult to find a video that matches the user's interest unless appropriate words are used. In this study, We propose a method to retrieve music videos with similar impressions by analyzing the comment data given by viewers to music videos on YouTube and the lyrics of the music. In the proposed method, comments and lyrics are converted into vectors by Word2Vec, and the music videos with similar impressions are retrieved based on clustering by Fuzzy c-means. According to the mean reciprocal rank (MRR) scores, it was clear that the output of the music videos had the same impression within the top 3 songs. (Keyword) 楽曲動画 / word2vec / ファジィクラスタリング
11.	TOTTORI Kento, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Color Image Scale Reconstruction Using Multivariate Analysis and Machine Learning, JSKE, 35-38, Mar. 2022. (Keyword) color image scale / machine learning / multivariate Analysis / dimensionality reduction
12.	Naomichi Tabuchi, Kazuyuki Matsumoto, Minoru Yoshida, Ryota Nishimura and Kenji Kita : SNS の投稿内容に基づく生活習慣病発症リスクの分析, 言語処理学会第28回年次大会発表論文集, 1946-1950, Mar. 2022. (Summary) 国内における死亡原因は，がん・循環器疾患・糖尿病などで全体の約 6 割を占めている．これらの病気はいずれも食事・運動・睡眠などの生活習慣と深い関りがあり，生活習慣病と呼ばれている．生活習慣病には自覚症状がほとんどないため，病気の悪化に気づきにくいという特徴がある．本研究では，より簡易的に生活習慣病を予防するシステムの構築を目指し，ソーシャル・ネットワーキング・サービス(SNS)を用いて，ユーザの投稿内容から食事・運動・体調・精神状態に関するツイートの分析を行った．調査の結果，本研究の成果がユーザの投稿内容から生活習慣病発症の危険因子を検出するシステムの構築につながることが分かった． (Keyword) Social Networking Services / lifestyle-related disease / SNS / BERT / Latent Dirichlet Allocation
13.	Tatsuya Miki, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 登場人物の感情表現文の抽出による物語のハイライト生成, 電子情報通信学会2022年総合大会情報・システムソサイエティ特別企画ジュニア&学生ポスターセッション予稿集, Mar. 2022. (Summary) 電子書籍の普及により数多くの文学作品をインターネット上で読むことが可能になった．しかし，読者がそれらの中から読む作品を選ぶことは容易ではない．本稿では物語について登場人物の感情が表現されている文を抽出することによる指示的要約の手法を提案する． (Keyword) 感情表現 / ハイライト生成 / 物語 / automatic abstracting
14.	Motonobu Yoshida, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : BERTを用いたSNS上における攻撃的文章訂正システム, 情報処理学会第84回全国大会講演論文集, Mar. 2022. (Summary) 現代社会において SNS の発展は凄まじく，ほとんどの人が使用しているという状況に置かれている．それに伴い，誹謗中傷や炎上などの被害にあう件数も増加している．2020 年には，某テレビ番組における出演者の行動や言動に対して，Twitter 上で誹謗中傷が多数投稿されたことから，自殺に追い込んでしまうといった事件が起きている．本研究では，SNS ユーザが発言を投稿する前に，投稿文の攻撃性を判定し，攻撃性を緩和した文章に変換して提示するシステムの構築を目指す． (Keyword) natural language processing / BERT / Social Networking Site / harmful expressions / toxic expressions
15.	Reishi Amitani, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 投稿文および画像といいね数との関係に基づく流行要因分析, HCG シンポジウム2021, Dec. 2021. (Summary) 本研究ではソーシャルメディアの動向を調査し，Twitterでのバズ現象を支える要因を探るための分析手法を提案することを目的とする．Twitterに投稿されたテキスト内容だけではバズ現象の原因を特定できない場合があるため，画像が添付されたツイートに限定し，テキストと画像のそれぞれの特徴に関連性があるかどうか，また，これらの特徴の関連性が，人気の指標である「いいね」や拡散の規模を表す「RT」の数とどのように関連するかを調べた．画像とテキストから抽出した特徴量を入力とし，いいね数とRT数を出力した後，2つの入力(画像とテキスト)から同じ次元数の特徴ベクトルを中間層から抽出するためにマルチタスク学習型のニューラルネットワークを学習する．これらの特徴ベクトル間の距離といいね数との関係を分析した結果，BERTによる分散表現とInceptionResNetV2の特徴量がいいね数とRT数を予測する要因となり得ることがわかった． (Keyword) Multi-task learning / Buzz classification / Social media / Trend analysis
16.	Kohei Torii, Eiichi Honda, 小林健一郎, 友江剛, Jiro Tsuruki and Kenji Kita : AI技術を用いたパノラマ歯科X線画像診断支援システムの開発, 歯科放射線61巻増刊号(第2回秋季学術大会), Vol.61, 31, Oct. 2021. (Keyword) artificial intelligence / パノラマX線画像 / 診断支援 / 歯科診断支援
17.	Tsubasa Miyashita, Minoru Yoshida, Kazuyuki Matsumoto, Ryota Nishimura and Kenji Kita : Lifestyle Extraction from Corpus of Disease Fighting Blogs, Proceedings of 2021 Shikoku-Section Joint Convention Record of the Institutes of Electrical, Electronic, and Information Related Engineers, 187, Sep. 2021. (Summary) 闘病ブログは，患者の生活に関する行動情報(病状・食事・運動)について書かれているものがあり，疾患への理解や納得を得られるものも見られ，同病者の励みとなっている．闘病ブログコーパスとは，闘病ブログを収集し，記事のタグ付けを行ったものである．本研究では年々増加傾向にある2 型糖尿病患者による闘病ブログコーパスから，闘病活動に関わる生活習慣の分析を行う．この分析結果より，病状悪化の危険因子や病状改善の手がかりとなる要素を，非医療従事者が自然言語処理および機械学習手法を用いて発見し，闘病者や糖尿病予備軍の人々に助言できるシステムの構築を目指す． (Keyword) disease fighting blogs / lifestyle analysis / lifestyle-related disease / type 2 diabetes / text mining
18.	Atsuya Miyake, Eiichi Honda, Kohei Torii, 小林健一郎, 友江剛 and Kenji Kita : 深層ニューラルネットワークを用いた歯科インプラントの同定, 令和3年度電気・電子・情報関係学会四国支部連合大会, 14-4, Sep. 2021. (Keyword) implant / deep learning / artificial intelligence
19.	Reishi Amitani, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Buzz tweet classification based on texts and images, JSKE 第23回日本感性工学会講演論文集, 1-5, Sep. 2021. (Summary) As the information diffusion on social networking sites has become more common, content buzzes have become a frequent occurrence on the Internet via social networking sites. In marketing, it is important to know what is trending, and it is necessary to have a means of quickly detecting buzz on SNS. In this study, we define buzz tweets based on the number of likes and RTs on a tweet, and analyze the content of the tweets to obtain the characteristics unique to buzz tweets, so that we can predict buzz tweets in advance. In this study, we propose a classification method to predict whether a tweet is a buzz tweet or not, based on the statement of the tweet and the attached image. The results of the classification experiments show that the classification accuracy is higher when the statement of the tweet and the image are used as features than when only the content of the tweet and the image are used as features, indicating the effectiveness of the proposed method. (Keyword) buzz tweet / information diffusion / multitask learning / social networking site
20.	Takumi Matsunaga, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Extraction of Emotional Information from Distributed Representation Using Emotion Classification, 第20回情報科学技術フォーラム(FIT2021)講演論文集, Vol.2, 41-46, Aug. 2021. (Summary) 近年，自然言語処理の分野では単語の一般的ではない用法の検出など，単語分散表現から意味推定を行う技術が研究されている．それらの研究では分散表現全体から単語の情報を抽出しており，感性以外の情報が多く含まれている．本研究では感性情報のみを抽出することでその精度上昇を目指す．その手法として単語分散表現から感情分類を行うニューラルネットワークを作成し，ネットワークが学習した重みから感性情報が強く表れている次元の情報を抽出する．評価方法としては，コサイン類似度を用いて抽出次元同士が類似した単語の比較を行う. (Keyword) emotional information / distributed representation / neural networks
21.	Xiaohan Cai, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Facts analysis of food tweets, 情報処理学会第83回全国大会, Mar. 2021.
22.	Ryota Kishima, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Effectiveness of Emotional Information in Construction of MBTI Personality Estimation Model, 日本感性工学会春季大会(JSAEE2021)講演論文集, 1-6, Mar. 2021. (Summary) In recent year, anyone has become possible to interact with many people all over the world easily because of that SocialNetworking Services (SNS) such as Twitter and Instagram have become widespread. Many researchers are analyzing user's personalityand emotion using features such as text information and image information of SNS. Our study analyzes the effectiveness of usingemotional information as a feature when constructing a personality estimation model for Twitter users. For analysis, we use MBTI(Myers-Briggs Type Indicator), one of the widely used personality theories for personality indicators. (Keyword) Myers-Briggs Type Indicator / Twitter / emotion
23.	芳川明奈, Jinsei Miyoshi, akinari kasai, 松本れいか, 宮城愛, Hiroshi Miyamoto, Yasushi Sato, Naoki Muguruma, Kenji Kita and Tetsuji Takayama : AI-based ESCC miRNA diagnosis model: retro-prospective multinational cohort study., 第74回日本食道学会学術集会, Dec. 2020.
24.	hara kazuki, Kohei Torii, Kazuyuki Matsumoto, Minoru Yoshida, Ryota Nishimura, Jiro Tsuruki, 小林健一郎 and Kenji Kita : Tooth detection and segmentation from panoramic radiographs using Mask R-CNN, 令和2年度電気・電子・情報関係学会四国支部連合大会, No.15-9, Sep. 2020. (Summary) In Japan, the number of diagnostic items in dental diagnosis is considerably larger than in other countries, and it is necessary to reduce the burden of diagnosis on dentists because it is time-consuming for dentists to write dental formulae from scratch. Since interpretation of pantomime images is essential for diagnosis, understanding of dental health status, and planning of treatment, processing of pantomime images is an important issue for automation. In this study, tooth detection and segmentation from pantomime images are performed automatically. In this paper, we describe the method. (Keyword) medical image / deep learning
25.	fukuda tsubasa, Ryota Nishimura, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : A Study on the Analysis of Comics Using Deep Learning, 令和2年度電気・電子・情報関係学会四国支部連合大会講演論文集, Vol.15-7, 1, Sep. 2020. (Summary) The way we enjoy manga, a Japanese subculture, has been changing in recent years. In recent years, the way we enjoy manga, a Japanese subculture, has been changing. Looking at the sales trends in the market, we can see that e-books, which are read on electronic devices instead of paper books, are becoming popular. However, while smartphones and other small-screen devices are highly portable and easy to use, they are too small to display comic books, which are usually in paperback or A5 format, page by page. However, it is too small to display comics, which are usually in paperback or A5 size, page by page. If it is possible to extract and rearrange frames, it may provide an optimal reading environment for readers. In this study, we investigate the detection of coma regions to solve this problem.
26.	Akira Fujisawa, Kazuyuki Matsumoto, Kazuki Othta, Minoru Yoshida and Kenji Kita : ASCII Art Category Classification Based on Transfer Learning and Data Augmentation from Different Types of Image Corpus, JSKE 第22回日本感性工学会大会講演論文集, 1-7, Sep. 2020. (Summary) In this paper, we propose an ASCII art category classification method based on transfer learning and data augmentation.ASCII art is one of the nonverbal expressions that visually express emotions and intentions.There are similar expressions such as emoticons and pictograms. However, most of them are either represented by a single characteror embedded in the statement as an inline expression. ASCII art is expressed in various styles, including dot art illustration and lineart illustration. Basically, ASCII art can represent almost any object, therefore the category of ASCII art is very diverse. Many existing image classification algorithms use color information, however, since most ASCII art is written in character sets, there is no color information available for categorization. Create an ASCII art category classifier using the grayscale edge image and the ASCII art image transformed from the image as a training image set. We also use VGG16, ResNet-50, Inception v3 and Xception's pretrained networks to fine-tune our categorization. As a result of the experiment of fine tuning by VGG16 and data augmentation, an accuracy of 80% or more was obtained in the "human" category. (Keyword) ASCII art / Transfer learning / Fine tuning / Data augmentation / Domain adaptation
27.	Taiga Kirihara, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : テレビ番組視聴者のツイートからのキーワード抽出および分類, 2020年度人工知能学会全国大会(第34回), Jun. 2020. (Summary) 近年，若者を中心にテレビ離れが進んでいる．その主な理由としては，長時間番組であったり，視聴者の視聴時間が限られているためである．また，近年，ソーシャルメディアの発達に伴い，テレビ番組の視聴において，リアルタイムに意見や感想を述べる機会が増加した．実際にTwitter上では，いわゆる実況ツイートが盛んに行われている．そこで，本研究では，テレビ番組実況ツイートを用いてテレビ番組のイベントシーンに関するキーフレーズを抽出する手法を提案し，効率的なテレビ番組視聴を目指したシーン検索システムを提案する．実験の結果，番組時間に対して約5%∼10%の誤差で番組内容を推定した．また，抽出されたキーフレーズをt-SNEアルゴリズムによってイベントシーンカテゴリごとに視覚化した． (Keyword) 自然言語処理 / 情報検索 / Twitter
28.	Goto Tomoaki, Ryota Nishimura and Kenji Kita : Mapping of acoustic vector sequence and document vector sequence, 情報処理学会全国大会講演論文集, No.4Q-05, 2-181-2-182, Mar. 2020. (Summary) In this study, we have constructed a model that allows for the mapping of acoustic features to document vectors. The proposed method uses RNN-LSTM for training time series data to enable cross-media mapping. We conducted evaluation experiments using the model and confirmed that the model mapped well in closed tests. In the future, we will conduct an open test to investigate the generalization performance.
29.	Takumi Kojima, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : ツイート内容による夜間活動休止時間の推定に関する研究, 情報処理学会第82回全国大会講演論文集, 469-470, Mar. 2020.
30.	Shogo Kono, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 文書分散表現を用いた音楽アーティストの特徴分類に関する研究, 情報処理学会第82回全国大会講演論文集, 477-478, Mar. 2020.
31.	Kazuyuki Matsumoto, Manabu Sasayama, Rei Terazono, Minoru Yoshida and Kenji Kita : Analysis of Utterance Intention and Kansei on Interview Dialogue Corpus, 日本感性工学会春季大会, 1-6, Mar. 2020. (Summary) In this study, we analyze interview dialogue corpora taken from an interview-type TV program, aiming at realizing aninterview dialogue system that brings out the best of the users of the system. After defining a tag set that classifies intention of utterance,these tags were added to six types of interview dialogue corpora, then we conducted utterance intention estimation using utterancesentence vectors obtained by using a pre-trained model. In the experiments, we analyze the expressions that are strongly related to theutterance intention, other important expressions, and the tendency of utterances of both the interviewers and the interviewees. (Keyword) Interview Dialogue System / Dialogue Act Prediction / Kansei Analysis / Corpus Analysis
32.	AIERKEN MAIERHABA, Kohei Torii, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 深層ニューラルネットワークに基づくテキストと画像間のクロスメディア検索, 電子情報通信学会2020年総合大会学生ポスターセッション, No.ISS-A-076, Mar. 2020. (Keyword) deep learning / クロスメディア検索
33.	Kohei Torii, Kazuyuki Matsumoto, Minoru Yoshida, Jiro Tsuruki, 小林健一郎 and Kenji Kita : 深層学習を用いたパントモ画像からのインプラント体検出, 電子情報通信学会2020年総合大会学生ポスターセッション, No.ISS-A-060, Mar. 2020. (Keyword) deep learning / Panoramic radiographs / implant / Implant detection
34.	Jinsei Miyoshi, Kenji Kita and Tetsuji Takayama : 人工知能(AI)による機械学習解析を用いた食道扁平上皮癌の血中microRNA新規AI診断モデルの構築., JDDW2019(第27回日本消化器関連学会週間), Nov. 2019.
35.	Yuko Sakuta, Junichi Tsunobe, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Earth Mover's Distance を用いた画像の印象推定, 第21回日本感性工学会大会, No.13A2-02, Sep. 2019.
36.	Junichi Tsunobe, Yuko Sakuta, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 深層学習に基づく配色イメージの感性マッピング, 第18回情報科学技術フォーラム (FIT2019), No.J-004, Sep. 2019.
37.	Naoya Fujino, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Word Mover's Distanceを用いたコーパス拡張による感情推定精度向上の検討, 言語処理学会第25回年次大会(NLP2019), 129-131, Mar. 2019. (Summary) 近年，深層学習をはじめとする機械学習はテキストマイニングや画像認識などの様々な研究分野で利用されている．今後の技術発展において必要不可欠であり，莫大な費用を投じている企業も少なくない．また，感情推定技術は人間とコンピュータ同士の円滑なコミュニケーションの実現に重要な役割を果たす．そういった中，よく問題点として挙げられるのが訓練データ量の不足や質の低下である．データ量の不足，質の低下は機械学習そのものの質の低下に直結する．本研究では，代表的な SNS の一つである Twitter に投稿されるテキストである「ツイート文」に着目し，そこに投稿されるツイート文を感情推定の対象とする．また，コーパス拡張が感情推定モデルの精度にどのような影響を及ぼすかを分析し考察する． (Keyword) Twitter / Word Mover's Distance / emotion estimation / data augumentation
38.	Yuuta Hada, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : リプライを用いたバズツイートの分類, 言語処理学会第25回年次大会(NLP2019), 237-240, Mar. 2019. (Summary) 近年 SNS の発達により，多数のユーザ間で口コミの拡散や共有が盛んにおこなわれるようになった．これによりインターネット上でコンテンツの流行が度々起こるようになった．ユーザ同士のコミュニケーションによっておこる流行のことを「バズる」と呼ぶ．この「バズる」ことを利用したマーケティングが重要視されており，いち早くバズに気づき取り入れることが今後のマーケティングで重要になってくると考えられる．既存研究において，時系列での拡散のされ方などから流行を検知するものがあるが，それらの手法では拡散された要因までは知ることができないという問題点がある．本研究では，Twitter においてバズったツイートに寄せられたリプライをベクトル化し，ツイートに対するリプライ全体を特徴ベクトルとすることで，「バズっている」特徴の定義を目指すとともにバズツイート内におけるタイプ別の分類分けの可能性について検討する． (Keyword) Twitter / reply / social media / buzz
39.	Hirokimi Fukuda, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Comment Mover's Distanceを用いた類似動画検索, IDRユーザフォーラム 2018, Nov. 2018. (Keyword) 動画コメント / Comment Mover's Distance / 類似動画検索
40.	Kazuki Ohta, Kazuyuki Matsumoto, Akira Fujisawa, Minoru Yoshida and Kenji Kita : Classification of ASCII arts by Deep Convolutional Neural Networks, 平成30 年度電気関係学会四国支部連合大会講演論文集, 155, Sep. 2018. (Summary) 近年のSNS でのテキストコミュニケーションにおいて，顔文字，絵文字，スタンプといった非言語表現は感情伝達において重要であり，ショートテキストで円滑なコミュニケーションを行う際には欠かせないものとなっている．非言語表現のうち，アスキーアート(ASCII Art; 以下AA)は使用されている文字が特に意味を持たないことが多く，その配置や形状により視覚的に感情や意図を伝達，表現できる．しかし，それ故に使用されている文字を手がかりとした分類が難しい．本研究では，AA を画像化し深層学習を用いて分類する手法を提案する． (Keyword) ASCII art / classification / deep convolutional neural networks
41.	Kazuyuki Matsumoto, 藤澤日明, Minoru Yoshida and Kenji Kita : Comparison of Classification Methods for Ascii Art, 第32回人工知能学会全国大会講演論文集, 1-4, Jun. 2018. (Summary) In recent years, a lot of non-verbal expressions have been used on social media. Ascii art (AA) is an expressionby visual technique using characters. In this paper, we set up an experiment to classify AA pictures by usingcharacter features and image features. We try to clear which feature is more effective for the method to classify AApictures. We proposed four methods; 1) character frequency based method, 2) character importance value basedmethod and 3) image feature based method, 4) characters image feature based method. We trained the neuralnetworks by using these four features. As the experimental result, the best classification accuracy was obtainedwith the feed forward neural networks using characters image feature. (Keyword) ASCII Art / image feature / character feature / neural network
42.	Dilixiati Jirela, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Emotion Recognition from Emoticons based on Deep Neural Netwotks, 2018年電子情報通信学会総合大会ISS特別企画学生ポスターセッション予稿集, Mar. 2018. (Summary) Text communication has been developed in recent years, all kinds of daily contacts are through the SNS regardless of age, gender and active. As an advantage of text-based communication via the Internet. There is a point that you can quickly respond to anyone wherever you are. In this paper, by letting the deep convolution neural network learn the image features of emoticons. We propose a method to classify by emotion expressed by emoticons. By treating emoticons as images, it is possible to capture visual features and think that natural identification can be made rather than handling as character strings. (Keyword) emoticon / emotion recognition / deep neural networks
43.	Kazuyuki Matsumoto, Fuji Ren, Minoru Yoshida and Kenji Kita : 絵文字を手がかりとした俗語感情コーパスの拡張, 言語処理学会第24回年次大会発表論文集, 797-800, Mar. 2018. (Summary) 絵文字が含まれていない文からでも，絵文字を推定することができれば，感情推定が可能になると考える．本論文では，俗語と絵文字を両方含む多量のコーパスを文字レベルで学習させ，絵文字推定器を作成し，俗語を含んだ発話文コーパスへの感情タグ付け手法を提案する． (Keyword) emoji / slang / emotion / corpus
44.	Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : プロフィール情報を用いたテキストの視点分類, IPSJ SIG Technical Report, Vol.2017-NL-234, No.11, 1-5, Dec. 2017. (Summary) 同一トピックに関する Twitter への投稿を，視点によって分類する手法を提案する．提案手法は，ユーザーのプロフィール情報を援用し，各投稿内容の分散表現を学習することで，類似する視点からの投稿をまとめる．
45.	Naoya Fujino, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Relation of Users Gender and Emotion Expression Tendency, Proceedings of the 31st Annual Conference of the Japanese Society for Artificial Intelligence, 2017, 1-4, May 2017. (Summary) In this study, we analyze the relation between emotion expressions of the Twitter users and their genders torealize emotion estimation adaptive to the users gender. We train a model to generate distributed representationsof sentences from a corpus divided according to gender. We train the features extracted with this model by using aneural network to construct an emotion estimator. The model constructed with our proposed method can estimateemotion with higher accuracy compared to the model without considering a gender. We also focus on slang andemoji as being related to emotion and analyzed their use tendencies to examine the difference between males andfemales. (Keyword) Twitter / emotion recognition / attribute
46.	Hirokimi Fukuda, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Index Generation from BGM by Using the Distinctive Comments Posted on the Songs, Proceedings of the 31st Annual Conference of the Japanese Society for Artificial Intelligence, 2017, 1-2, May 2017. (Summary) The purpose of this study is to generate indexes to the BGMs based on the distinctive comments annotated tothe BGM videos posted on Nico Nico Videos. Our proposed method detects the end/start positions of the BGMsconsidering the increase/decrease of the number of the comment data, the comments distinctive to the songs andthe exclamatory expressions annotated to the videos. The evaluation result showed that the proposed methodcould generate correct indexes within 10 seconds error for the BGM videos with length of less than 40 minutes andwith annotation of over 30,000 comments. (Keyword) index generation / Nico videos / distinctive comment
47.	Akira Fujisawa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Classification method for large ASCII arts using image features, Proceedings of the 31st Annual Conference of the Japanese Society for Artificial Intelligence, 2017, 1-2, May 2017. (Summary) ASCII arts are visual expression that composed by strings. In the field of natural language processing, ASCII arts are treatedas a sequence of strings. In this paper, we consider to use shapes of ASCII art as features. Even if scale of ASCII arts and kindof characters that used to compose ASCII arts are different, proposed method enable to compare visual similarity. To extractshapes feature of ASCII art, we extract image features from ASCII art by converting ASCII arts to images. In the experiment,we compared visual similarity between ASCII arts and emoticons. As the experimental result, it was found that the imagefeature was effective to compare visual similarity even if the ASCII art and the emoticon used different kind of character (Keyword) ASCII art / image feature / classification
48.	Kazuyuki Matsumoto, Fuji Ren, Minoru Yoshida and Kenji Kita : 感情表出傾向の分析による対話破綻検出, 言語処理学会第23回年次大会発表論文集, 158-161, Mar. 2017. (Summary) 本研究では，ユーザの感情の変化に合わせてシステム自身の感情を表出させることができないために起こる対話破綻について，発話文から推測可能な感情傾向の比較により検出する手法を提案する．具体的には，発話文の分散表現ベクトルをもとに感情が表出されている強度を推定するモデルを用い，直前の 2 発話における推定結果との類似度の計算を行うことで，そのパターンを特徴量として対話破綻のラベルを推定する．評価実験では，提案手法を用いた対話破綻ラベルの推定をおこない，文の分散表現ベクトルの類似度パターンによる手法との比較をおこなう． (Keyword) 対話破綻検出 / 感情推定 / 分散表現
49.	Matsumoto Ryusei, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Visualization of the occurrence trend of infectious diseases using Twitter, 人工知能学会インタラクティブ情報アクセスと可視化マイニング研究会(第15回), 48-53, Mar. 2017. (Summary) The purpose of this research is to extract comments on infectious diseases from Twitterand to create an infection map.As a method,we check the facts of the extracted utterance,extractthe past tweets of the infected person from the result,and estimate the place where the personlives.We used SVMs both for the fact check and place estimation. (Keyword) Twitter / visualization / trend of infectious diseases
50.	Miyake Takeshi, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Flame prediction based on harmful expression judgement using distributed representation, 人工知能学会インタラクティブ情報アクセスと可視化マイニング研究会(第15回), 1-6, Mar. 2017. (Summary) In recent years, flaming on social media has been a problem. To avoid flaming, it is useful for the system to automatically check the sentences whether they include the expressions that are likely to trigger flaming or not before posting messages. In this research, we target two harmful expressions. There are insulting expressions and the expressions that are likely to cause a quarrel. Firstly, we constructed a harmful expression dictionary. Because a large cost requires to collect the expressions manually, we constructed the dictionary semi-automatically by using word distributed representations. The proposed method used distributed representations of the harmful expressions and general expressions as features, and constructed a classier of harmful/general based on those features. An evaluation experiment found that the proposed method could extract harmful expressions with accuracy of approx. 70%. On the other hand, it was found that the proposed method could also extract unknown expressions, however, it wrongly extracted non-harmful expressions at a rate of approx. 40%. (Keyword) harmful expression / distributed representation / Support Vector Machine / Flame prediction
51.	Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : 表形式からの分散表現獲得, 情報処理学会自然言語処理研究会研究報告, Vol.2016-NL-229, No.19, 1-6, Dec. 2016. (Summary) 表形式における単語埋め込み手法について考える．通常の文章を対象とした学習手法に比べ，表形式においては，単語の周辺文脈として，縦方向・横方向の二種類の文脈が考えられる．本稿では，これら同時に考慮した，二次元的なモデルを提案し，一次元的なモデルとの比較を行う．
52.	Fujioka Yuuki, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : 8bit 音源に対する自動採譜システムに関する研究, 平成28年度電気関係学会四国支部連合大会講演論文集, 17-5, Sep. 2016. (Summary) 本研究では，自動採譜分野において 8Bit 音源の特徴に着目して研究をする．8bit 音源とは，三角波と矩形波が二つの三和音とノイズからなる，一般的な音楽を形成するにあたる最低限の要素で構成された，シンプルな音源である．つまり，8bit 音源の採譜ができればベースやメロディーの動きが明らかになりその楽曲の骨組みが分かるようになる．従来は，自動採譜の分野の研究において，一般的なクラシックやポップスなど，大量の楽音が混在した楽曲データを対象としていた．1)それに対し，本実験では 8-bit 音源を対象にし，その最低限の和音のみで構成されたものを対象にすることによる影響について考察した． (Keyword) 8bit音源 / 自動採譜
53.	Nagai Hiroki, Fujisawa Akira, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Anime Character Classification Using Face Image based on Deep Learning, 平成28年度電気関係学会四国支部連合大会講演論文集, 178, Sep. 2016. (Keyword) Anime Character / Deep Learning / Face image / classification
54.	Nagafusa Kei, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Facial Expression Identification by Deep Convolutional Neural Networks, 第15回情報科学技術フォーラム講演論文集, Vol.3, 293-294, Sep. 2016. (Keyword) Deep Learning / Facial Expression / Deep Convolutional Neural Networks
55.	Fujisawa Akira, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : An Illustration Image Classification By The Style Focusing on Infrequent Colors, 第15回情報科学技術フォーラム講演論文集, Vol.3, 87-88, Sep. 2016. (Keyword) Illustration image / IF-hist / style
56.	松田紘伸, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Twitterを用いた病気の事実性解析及び知識ベース構築, 人工知能学会全国大会2016講演論文集, Jun. 2016.
57.	藤岡亮太, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Twitterから抽出したユーザの発言に基づく健康状態の推定, 人工知能学会全国大会2016発表予稿集, Jun. 2016.
58.	Satoshi Tanaka, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Personality Analysis Method of Twitter User for Information Recommendation, The Proceedings of the 30th Annual Conference of the Japanese Society for Artificial Intelligence, 2016, 1-4, Jun. 2016. (Keyword) personality analysis / Egogram / Twitter / information recommendation
59.	Akira Fujisawa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Detecting emoticons in ASCII arts using image features, The Proceedings of the 30th Annual Conference of the Japanese Society for Artificial Intelligence, 2016, 1-2, Jun. 2016. (Keyword) ASCII arts / emoticon / image feature / Histogram Oriented Gradients / Scale-Invariant Feature Transform
60.	Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : 表形式のトピックモデルとその数値単位推定への応用, 情報処理学会自然言語処理研究会研究報告, Vol.2016-NL-226, No.16, 1-6, May 2016. (Summary) 表形式中の数値表現について単位が省略されている場合に，その単位を推定するための手法を提案する．Wikipedia 表形式中の，数値のみを含むセルを対象に，一行目のセル等の「周辺文脈」を利用し，適切な単位を推定する．また，表形式の外側の文章を利用するために，表形式と文章を同時にモデル化するためのトピックモデルを提案し，トピックの推定結果を単位推定に利用することで精度の向上を図る．
61.	Fumihiro Iwasa, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Twitterユーザの属性別感情推定の検討, 言語処理学会第22回年次大会講演論文集, Mar. 2016.
62.	Yuuki Murakami, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : 深層畳み込みニューラルネットワークを用いた顔認識の精度評価, 2016年電子情報通信学会総合大会学生ポスターセッション, Mar. 2016.
63.	Masaya Matsuoka, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : トピック変動の分析による俗語の特徴抽出, IPSJ SIG Technical Report, Vol.2016-NL-225, No.4, 1-5, Jan. 2016. (Summary) 本研究では俗語の使われ方が時間経過によって変化することに着目した．俗語の含まれたツイート文の集合を潜在的ディリクレ配分法により分析し時間ごとのトピックを得る．このトピックにおけるトピック重要語の集合について隣接する時間(月単位)での類似度を計算し，トピックの変化を抽出し，俗語の特徴となるものを検討する． (Keyword) slang / トピック変動 / Latent Dirichlet Allocation
64.	Yuki Murakami, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Shape-based image retrieval system for Two-dimensional Drawings, Journal of Shikoku-Section Joint Convention of the Institutes of Electrical and Related Engineers, 256, Sep. 2015.
65.	Takashi Matsuo, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : The optimal parameter estimation in the logo detection, Journal of Shikoku-Section Joint Convention of the Institutes of Electrical and Related Engineers, 256, Sep. 2015.
66.	Keisuke Hashimoto, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : A study on game music search by audio fingerprints, Journal of Shikoku-Section Joint Convention of the Institutes of Electrical and Related Engineers, 256, Sep. 2015.
67.	Ryuji Miyamoto, Minoru Yoshida, Kazuyuki Matsumoto and Kenji Kita : Sound Source Identification in Polyphonic Music Using a Mixed-sound Templates including Vocal Sound, Journal of Shikoku-Section Joint Convention of the Institutes of Electrical and Related Engineers, 268, Sep. 2015.
68.	Shimizu Hayato, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Cheating Detection by Using Correlation of Comment Distribution in Nicovideo, 第14回情報科学技術フォーラム講演論文集, Vol.2, 229-230, Sep. 2015. (Keyword) Nicovideo / Video Sharing Site / Cheat Video / Time series correlation analysis
69.	Okamoto Takuma, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : An Influence Estimation Method in Twitter Considering to Range of Influence, 第14回情報科学技術フォーラム講演論文集, Vol.2, 227-228, Sep. 2015. (Keyword) Twitter / Influence / Twitter API / Follower / Retweet
70.	Maekawa Kazuki, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Denoising techniques for flaming analysis of BBS, 第14回情報科学技術フォーラム講演論文集, Vol.2, 219-220, Sep. 2015. (Keyword) BBS / Denoising / Flame
71.	Yamada Tatsushi, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Category Classification of New Words Robust to Segmentation Errors, 第14回情報科学技術フォーラム講演論文集, Vol.2, 243-244, Sep. 2015. (Keyword) category classification / new word / segmentation error
72.	Hayato Shimizu, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Illegal Movie Detection by Using Comment Distribution Based on Temporal Correlation in NicoNico Douga, IEICE Technical Report, Vol.115, No.177, 97-101, Jul. 2015. (Summary) Recently, the video sharing site such as NicoNico Douga or YouTube are used for various use by individual and corporation. The authors of the video often make a profit by uploading video. However, because the video rating method is based on the frequency of looking and listening or the number of posted comments, the rating of the video can be intentionally increased by the author. We proposed the illegal video detection method by focusing on the correlation of temporal comment distribution rather than the frequency of looking and listening. The success detection rate was compared the proposed method with the baseline binary classification method by using machine learning based on the word feature. (Keyword) NicoVideo / comment distribution / temporal correlation
73.	Saitou Tatsuya, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Automatic Font Generation Based on On-yu, 情報処理学会全国大会講演論文集, 4-767-4-768, Mar. 2015.
74.	Maekawa Kazuki, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 書き込みのエントロピーに着目した速報スレッドにおける盛り上がり要因の分析, 言語処理学会第21回年次大会発表論文集, 620-623, Mar. 2015. (Keyword) BBS / news thread / entropy
75.	Keisuke Yoshida, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 顔文字のパーツの種類と表情の強弱に着目した顔文字表情推定, 言語処理学会第21回年次大会発表論文集, 449-452, Mar. 2015.
76.	Kazuyuki Matsumoto, Seiji Tsuchiya, Minoru Yoshida and Kenji Kita : 若者言葉の意味と感性を考慮した標準語への変換手法, 言語処理学会第21回年次大会発表論文集, 529-532, Mar. 2015. (Keyword) 若者言葉 / 俗語 / 印象 / 変換
77.	Kohsuke Miyauchi, Kazuyuki Matsumoto, 篠山学, Minoru Yoshida and Kenji Kita : 歌詞検索に必要な内容語数の分析およびクエリ拡張手法の提案, Journal of Shikoku-Section Joint Convention of the Institutes of Electrical and Related Engineers, Sep. 2014. (Summary) 本研究は間違いの含まれる入力クエリの検索可能最低クエリ数を調べることにより検索システムの効率化や柔軟で強固な検索システムの作成の参考になるデータをとることを目的とした．誤り歌詞片を用いた評価実験の結果，必要な最小内容語数は平均4単語となった． (Keyword) 歌詞検索 / クエリ拡張
78.	Higashi Atsunari, Shun Watanabe, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : LDPC符号とハミング符号を用いた補助情報を伴う符号化システムの性能評価, 情報・システムソサイエティ・学生ポスターセッション予稿集, ISS-P-255, Mar. 2014. (Summary) 近年, インターネットの普及により通信する情報量が増加しているため, 情報を効率良く送信されるための情報圧縮技術が求められている. その中で, 1975年にWyner[1], Ahlswede and Körner[2]らが補助情報を伴う符号化システムの許容伝送率領域の証明した. しかし, 彼らの論文において誤り訂正符号などを実際に使用しての実験は行なわれていない. そこで本稿は, 補助情報を伴う符号化システムにおいて主符号化器にLDPC符号, 副符号化器にハミング符号を使用した場合の性能評価を行なう. (Keyword) LDPC符号 / ハミング符号
79.	Kanematsu Hideto, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 知覚ハッシュによる類似画像検索, 情報・システムソサイエティ・学生ポスターセッション予稿集, ISS-P-256, Mar. 2014. (Summary) 近年，インターネットの普及により個人が触れることのできる画像データの量が膨大に増えている．それに伴い，画像から類似している画像を検索する類似画像検索の需要も増している．本研究では知覚ハッシュを使い，従来手法のグレースケールの代わりに3チャンネルのカラー情報を用いることで，人間の感性に近い類似画像検索を目標としている． (Keyword) Perceptual Hash / Block Mean Value Hash / Image Retrieval
80.	Okamoto Takuma, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : ナイーブベイズ法を用いたTwitterによる性格推定, 言語処理学会第20年次大会予稿集, Mar. 2014. (Summary) 本研究ではTwitterを利用し，Web上のツールを用いてエゴグラム分析を行ったユーザアカウントを対象に投稿内容を収集し，投稿内容に出現する単語を素性としてナイーブベイズ法による機械学習をおこない，投稿内容から性格の自動分類することを目的とする． (Keyword) ナイーブベイズ法 / Twitter / 性格推定
81.	Yamada Tatsushi, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : アーティスト評判分析に有効な特徴表現の抽出, 言語処理学会第20回年次大会予稿集, Mar. 2014. (Summary) 本研究では，n-gram出現頻度に基づき，アーティストに関するつぶやきデータから，評判分析に有効な特徴表現を抽出する．さらに，特徴表現に後続する単語を素性として，Support Vector Machine による機械学習によるカテゴリ分類をおこなう．評価実験の結果，アーティスト/カテゴリごとに差は出たものの，約80%の正解率で分類できることがわかった． (Keyword) text mining / マイクロブログ / Twitter / 評判分析
82.	Kazuyuki Matsumoto, Sasayama Manabu, Miyauchi Kohsuke, Minoru Yoshida and Kenji Kita : 歌詞における聞き慣れない表現と誤りとの関連性の分析, 言語処理学会第20回年次大会予稿集, Mar. 2014. (Summary) 本稿では，歌詞において特徴的な表現について，歌詞コーパスとWeb コーパスにおける出現頻度を比較することで分析する．また，聞き慣れない表現と誤りが起こるフレーズとの関連を，実際に楽曲の歌詞の書き起こしをおこなうことで得た歌詞片を用いることにより調査し，歌詞に特徴的な表現との比較もおこなう． (Keyword) 歌詞 / 単語親密度
83.	Norihide Kitaoka, 市川賢, Satoru Tsuge, 武田一哉 and Kenji Kita : 種々のテキスト検索モデルの頑健性向上による音声ドキュメント検索の高精度化, 音声ドキュメント処理ワークショップ, Mar. 2014. (Summary) 音声ドキュメント検索には様々なテキスト検索モデルが適用できる．それらについて，音節検索の併用や検索語彙の拡張などの手法を適用して高精度化を図り，最新の技術として比較検討を行った．
84.	Fujisawa Akira, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Identification method of author of illustration with using the face of the anime character, HUMAN COMMUNICATION GROUP SYMPOSIUM 2013, Vol.HCG2013-I-2-1, 523-530, Dec. 2013. (Summary) 本研究では，キャラクターを描いたイラスト画像に対し，顔部分に着目し，特に，目の部分からエッジ特徴量(Line Percent: LP)を抽出し，その書き込み具合をもとに，イラストレーター特有の特徴を見出して作者識別を行う．人による識別と，LPに基づく自動識別実験をおこない比較した結果，人による識別と同等とまではいかないが，ある程度の識別率を得ることができた．このことから，イラスト画像における部分的な書き込み具合を特徴量として用いることで，ある程度の精度での作者識別が可能なことがわかった． (Keyword) image processing / Illustration Image / Edge / Canny Edge Detection
85.	Akita Kyosuke, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Estimation of Intimacy between Characters in a Play Based on Response Rate, HUMAN COMMUNICATION GROUP SYMPOSIUM 2013, Vol.HCG2013-I-1-16, 511-515, Dec. 2013. (Summary) In this paper, we focused on the conversation sentences in a script and tried to obtain the relationships between the characters in the play. Generally, we understand the personalities of the characters and the relationships between them according to what they speak and how they behave in the story. Generally, we understand the personalities of the characters and the relationships between them according to what they speak and how they behave in the story. Therefore, we thought to use the emotional expressions in the dialogue sentences in the script to obtain the personalities of the characters and their relationships. There are many more elements except for emotional expressions in the conversations and we thought that such elements would be useful to predict interpersonal relationships. In this paper, we proposed a method to extract intimacy degree that is one of the most important elements in interpersonal relationships. We obtained the intimacy degree of the characters by calculating a response rate based on the speech frequency and the number of responses for the dialogue sentences between those characters. We constructed a corpus that collected dialogue sentences from the scripts and analyzed the effectiveness of the intimacy degree calculated by the proposed method by using the corpus. (Keyword) Scenario Dialogue Sentence / Response Rate / Intimacy Degree
86.	蜜石湧斗, Sasayama Manabu, Kazuyuki Matsumoto and Kenji Kita : Android Application for Lyric Search System Using Semantic Information, 平成25年電気関係学会四国支部連合大会講演論文集, Sep. 2013. (Summary) 我々は，入力する歌詞クエリの誤りに頑健な歌詞検索システムを開発することを目的とする．これまでに意味的なクエリの拡張を行い，意味的な類似度の高い順に歌詞を表示するシステムを実装した．本稿では多くのユーザに評価してもらうためユーザが利用しやすい Androidアプリを作成する．まず意味的情報を考慮したクエリの拡張について説明し，作成したアプリの構成について説明する． (Keyword) 歌詞検索 / Androidアプリ / 意味情報 / クエリ拡張
87.	Kyosuke Akita, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : 俗語の出現頻度と文末表現の関連性に関する分析, 平成25年度電気関係学会四国支部連合大会講演論文集, Sep. 2013. (Summary) 近年，スマートフォンなどの携帯型端末の爆発的な普及に伴い，Twitterやfacebookを始めとするSNSやブログサービスなどの利用者が増加している傾向にある．それにより，現在ネット上にはSNSなどから得られる膨大なテキストデータが存在している．これらのテキストデータでは，俗語と呼ばれているくだけた表現が多く用いられている．そのため，俗語が使用されているテキストデータを分析して情報抽出を行う場合，形態素解析誤りが影響し，信用できる結果が得にくい．しかし，現代の言葉を理解する研究は評判分析など様々な場所で応用できるものとして非常に重要であると考えられる．そのため，現代の言葉の特徴である俗語を分析し理解する研究は非常に重要であると考えられる．本論文では，ブログにおける文章の文末表現の混合率および変移率に着目し，俗語が表れやすい文章がどのような特徴を有するのかを分析する． (Keyword) 俗語 / 文末表現 / 混合率 / 変移率
88.	Gohda Sho, Shun Watanabe, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : Steganography using Coding with Cost Function, IEICE Technical Report, Vol.113, No.153, 5-9, Jul. 2013. (Summary) In recent years, the techniques that make communication safety are getting important as network communication advances. Cryptography cannot always meet this demand. Steganography can take measures to meet this situation. LSB substitution is the well-known technique of steganography. But it excludes the possibility of error bit generation, so that many techniquesbased on LSB substitution have been proposed. This paper propose the steganography using coding with cost function that convert embedded information to error pattern and embed it by exclusive or. Then this paper also shows that it can approach to theoretical bound asymptoticallyby numeric calculation. (Keyword) Information Hiding / Steganography / Coding with Cost Function / Embedded using Error Pattern
89.	Kyosuke Akita, Kazuyuki Matsumoto and Kenji Kita : 文字種と画数を用いた未知若者語の抽出, 言語処理学会第19回年次大会(NLP2013)予稿集, Mar. 2013. (Summary) 本研究では若者語は日常的に使用されやすい語を用いて生成されるという仮定に基づいて，語の表層的な特徴より文中の若者語を抽出することを検討する．文中からの抽出には条件付き確率場(以下CRF)を用いて確率モデルで文字列が若者語である確率の高い言葉を推定することにより行う． (Keyword) 単語抽出 / 条件付確立場 / 文字種 / 画数 / 俗語
90.	Ohnishi Tsubasa, Kazuyuki Matsumoto, Kenji Kita and Fuji Ren : Relation Analysis on Emotion and EEG Fluctuation in Dialogue, 平成24年度電気関係学会四国支部連合大会講演論文集, 95, Sep. 2012. (Summary) 脳波等の患者から発信される情報が重要なのはいうまでもないが，患者からの情報のみに頼るのではなく，介護を行う側の意識を変化させることが，より重要だと考える．本研究の最終目的は言葉が被介護者の感情にどのように影響するかを言語情報から予測することである．このような予測が可能になれば，患者の心を傷つけるような言葉を未然に防ぐことができる．本研究では，共感を分析しやすい対象として，演劇シナリオを用いることにした．演劇シナリオは，一般的な対話と比較すると，演出を重視しているためか，感情的な表現が多用されている特徴がある．介護場面などでは，演劇口調で話すことはほとんどないと考えられるが，本研究ではこうした感情表現の多用により，脳波に明確な変化が起きることを期待する． (Keyword) 対話分析 / 脳波変化
91.	Ryoh Ogo, Kazuyuki Matsumoto, Sasayama Manabu and Kenji Kita : Lyrics Search System Focused on Mishearing and Memorizing Mistake, 平成24年度電気関係学会四国支部連合大会講演論文集, 357, Sep. 2012. (Summary) 現在普及している楽曲検索において，アーティスト・曲名検索，歌詞入力検索ならば前者が主流である．実際には歌詞入力検索も同等の性能であると考えられるが，聞き間違いや記憶間違いを原因とした誤入力が検索精度を低下させる．本研究では，従来の言語研究のデータから使用者の入力ミス等の影響を最小限に抑え歌詞入力検索の精度の低下を防ごうと考えている．本研究では，歌詞の聴取実験により収集した誤り歌詞データを用いた分析をおこない，歌詞の誤りにどのように対処していくかを議論する． (Keyword) 歌詞検索
92.	Kazuyuki Matsumoto, Onishi Tsubasa, Kenji Kita and Fuji Ren : Construction and Analysis of Scenario Dialogue Emotion Corpus, 第11回情報科学技術フォーラム(FIT2012)講演論文集, Vol.2, 242-243, Sep. 2012. (Summary) 従来，言語・音声・画像といった情報からの感情推定を目的とした研究が多くの研究者によって行われてきたが，感情を表面的な情報から得るには限界がある．この理由として，感情は内在的な反応であるため，表情や行動を伴わないことも多いことがあげられる．しかし，発話者同士が親密な関係であればある程，感情の変化に伴う相手の言葉遣いの微細な変化に気付くことがある．このような言葉遣いの細やかな変化を検出できれば，対話における感情推定精度を向上できる可能性がある．徳久らは，発話文の文末表現に着目したが，言葉遣いの細やかな変化は，文末表現のみに表れるものではないと考えられる．たとえば，相手が普段使用しないような単語や慣用句を使用する場合でも，何かしら相手の心境の変化を察することもある．本研究では，話者の発話時における感情と，発話役割や文体との関係を調べる準備段階として，対話を記述したコーパスに感情，発話役割，文体タグの付与を行なった．このコーパスを分析することで，相手発話のどのような要素が影響して感情が変化するかを考察する． (Keyword) シナリオコーパス / 対話処理 / 感情推定
93.	Saito Narumi, Kazuyuki Matsumoto, Motoyuki Suzuki and Kenji Kita : Index Compression for Audio Fingerprinting Systems, 第11回情報科学技術フォーラム(FIT2012)講演論文集, Vol.2, 181-182, Sep. 2012. (Summary) 楽曲の曲名やアーティスト名，アルバム名などの楽曲情報を得るときには，楽曲検索システムが利用されている．マイクから楽曲の一部を読み込み，楽曲情報を得る技術にも使われている．多くの楽曲検索システムでは，オーディオ指紋(audio fingerprint)が使われている．オーディオ指紋は，楽曲の特定に用いられる特徴である．オーディオ指紋を楽曲情報データベースと共に用いることにより，楽曲の一部から楽曲情報を得ることができる．また，オーディオ指紋は楽曲を識別できることから，楽曲検索のみではなく，ネットワーク上に流通している著作権を侵害した楽曲の検出も可能である．効率的なオーディオ指紋検索には，接尾辞配列を利用した手法がある．これは，楽曲データベースの増大に比例して空間コストが増加する欠点を持つ．本稿では，データベースのインデックスを圧縮することにより，空間コストの削減を行う． (Keyword) audio fingerprint / information retrieval / 音楽検索
94.	Kazuyuki Matsumoto, 吉岡伸浩, Kenji Kita and Fuji Ren : うつキーフレーズと感情変動に基づくブログからのうつ検出手法, 言語処理学会第18回年次大会講演論文集, 1126-1129, Mar. 2012. (Summary) 我々は，ブログの書き手に，うつ傾向があることを自覚させることにより，うつ病の早期発見・早期治療に役立てることができると考えた．まず，うつ傾向の有無を，ブログ内容から推定することができるか否かをテキスト分類手法を用いた実験により明らかにし，どのような特徴がうつ判定に有効であるかを調査する．また，人間の心の変化をとらえるための特徴として，時間経過に伴う記述内容の変化から読み取れる感情状態の変化がある．多くのブログが，毎日の出来事や心情を時系列順に書き記したものであるため，そこから時間的変化を得ることが出来る．本稿では，うつ特有の言葉遣いを収集したうつキーフレーズ辞書に基づくうつ検出手法と，感情状態の変化に基づいたうつ検出手法を提案する． (Keyword) natural language processing / うつ検出 / ブログ
95.	吉岡直輝, Kazuyuki Matsumoto, Kenji Kita and Fuji Ren : 印象尺度に基づいた文章言い換え手法, 言語処理学会第18回年次大会講演論文集, 425-426, Mar. 2012. (Summary) 本稿では文中の単語を，同じ意味の，別の印象を持つ単語に置き換えることで，印象の違った文に変換する手法を提案する．友人や家族とコミュニケーションをとる際に用いるような単語の印象を，Casual(柔かい表現)，目上の人とコミュニケーションをとる際に用いるような単語の印象をFormal(堅い表現)の尺度と定義し，新聞記事とブログを基に，単語のCasual/Formalの印象付与を自動で行う．新聞記事は短い文で的確にニュースのポイントを読者に伝えるため，会話では使用されることが少ない漢字を多く用いる．このことから，新聞記事で多く用いられる単語をFormalな単語として分類する．ブログは個人的な体験や日記を掲載し，臨場感を出すためにくだけた会話表現を多く用いる傾向がある．このことから，本実験では，ブログで多く用いられる単語はCasualな単語として分類する． (Keyword) natural language processing / 言い換え / 印象
96.	平尾卓也, Kazuyuki Matsumoto, Kenji Kita and Fuji Ren : コーパスに基づく雑談を目的とした発話役割同定, 言語処理学会第18回年次大会講演論文集, 109-110, Mar. 2012. (Summary) 我々は，コーパスに基づく雑談を目的とした発話役割の同定の研究を行っている．西原らは発話役割の一部は助詞・助動詞の組み合わせで表されるとし，助詞・助動詞の意味を用いたルールによって発話役割の同定を行った．そこで本稿では品詞情報を用いた雑談の発話役割に有効な重みを模索し，コーパスに基づく雑談を目的とした発話役割の同定手法を提案する． (Keyword) natural language processing / 発話役割 / 雑談 / corpus / 形態素2-gram
97.	Kusumoto Yusuke, Kazuyuki Matsumoto and Kenji Kita : 脳波を用いた類似画像検索手法, Journal of Shikoku-Section Joint Convention of the Institutes of Electrical and Related Engineers, 4, Sep. 2011. (Summary) 本研究では，感性的画像検索を実現するため，脳波から抽出した相互相関係数を，感性特徴量として用いる．感性特徴量と画像特徴量を組み合わせて画像を検索することで，ユーザの感性に適合した検索結果を提示することを目標とする． (Keyword) electroencephalogram / filtering / information retrieval / 感性的画像検索 / information retrieval system
98.	Motoyuki Suzuki, Seiji Tsuchiya and Kenji Kita : Emotion recognition from speech based on both lingistic and prosodic features, 日本音響学会研究発表会講演論文集, 285-288, Sep. 2011.
99.	Li Yang, Kazuyuki Matsumoto, Kenji Kita and Fuji Ren : Automatic Construction of Japanese-Chinese Word Dictionary on Super-Function Extraction for Japanese-Chinese Machine Translation, 第10回情報科学技術フォーラム講演論文集, Vol.2, 221-222, Sep. 2011. (Summary) 本研究ではスーパー関数を用いた日英機械翻訳の利点と問題点を考察し，スーパー関数に基づく機械翻訳を用いて日中機械翻訳システムの試作を行う．また，新たに提案したSFの変数の拡張についても述べる．提案手法では変数となる品詞の種類が多くなるため，訳語を適切に選択するために，参照する単語辞書の自動作成手法を提案する．最後に，提案手法による自動翻訳実験を行い，スーパー関数に基づく日中機械翻訳の精度を向上するには，対象とする例文に応じた対訳単語辞書が必要であることを明らかにする． (Keyword) machine translation / スーパー関数 / 日中機械翻訳 / 訳語選択 / 辞書構築
100.	Kazuyuki Matsumoto, Kenji Kita and Fuji Ren : Estimating Emotion of Wakamono Kotoba Based on Similarity of Example Sentences, 第10回情報科学技術フォーラム講演論文集, Vol.2, 281-284, Sep. 2011. (Summary) 若者言葉は，既存の語の言い換えたものであったり，省略表現も多いが，複合的な意味を持つ語や，既存の概念には属さない語も多い．こうした語のために，新語義を定義する必要があるが，その定義方法について未解決の問題が多い．我々の研究の目標は，若者言葉を含む文からの感情推定である．若者言葉を特徴として感情推定を行う際に，若者言葉を，その意味や用法，さらには感情などで分類することができれば感情推定の精度向上につながると考えられる．そこで，本稿では，若者言葉の用例に基づき，感情表現に関する情報を抽出することで，若者言葉の感情を推定する手法を提案する． (Keyword) 感情認識 / sentience computer / 若者言葉
101.	Masaki Mifune, Motoyuki Suzuki, Fuji Ren and Kenji Kita : クラスタリングに基づくGMM学習法による話者のモデル構築, IEICE Technical Report, Vol.SP2011, Jul. 2011.
102.	Masashi Ohnishi, Masami Shishibori and Kenji Kita : 時系列アクティブ探索による類似音楽検索システムの改良, 第3回データ工学と情報マネジメントに関するフォーラム(DEIM2011), Mar. 2011.
103.	Tatsuki Marunishi, Masami Shishibori and Kenji Kita : SIFT特徴量を用いた映像データに対する人物検索システムの開発, 第3回データ工学と情報マネジメントに関するフォーラム(DEIM2011), Mar. 2011.
104.	Yuichi Nishino, Masami Shishibori and Kenji Kita : Earth Mover's Distanceを用いたSIFT特徴量に基づく類似画像検索手法, 第3回データ工学と情報マネジメントに関するフォーラム(DEIM2011), Mar. 2011.
105.	Sachi Tsuji, Masami Shishibori and Kenji Kita : ハミングを入力とする類似音楽検索システムにおける自動採譜手法の検討, 第3回データ工学と情報マネジメントに関するフォーラム(DEIM2011), Mar. 2011.
106.	岸本諭, 楠本祐介, 多田圭佑, Masami Shishibori and Kenji Kita : 感性特徴と画像特徴の相関関係に関する研究, Journal of Shikoku-Section Joint Convention of the Institutes of Electrical and Related Engineers, 151, Sep. 2010.
107.	山川義介 and Kenji Kita : 類似画像検索システム『SUDACHI』を用いた商品レコメンデーション, 第10回日本感性工学会大会, Sep. 2008.
108.	山川義介 and Kenji Kita : ECサイトにおける画像の類似性判定∼『SUDACHI』の開発と商品推薦エンジンへの応用∼, 日本行動計量学会第36回大会, 5-6, Aug. 2008.
109.	北條奈緒美, Masami Shishibori and Kenji Kita : WWW検索エンジンを用いた質問文内の用語特定手法, 平成19年度電気関係学会四国支部連合大会, 246, Sep. 2007.
110.	中出訓規, Masami Shishibori, Satoru Tsuge and Kenji Kita : 関連語を用いたWWW検索結果のフィルタリング手法, 平成19年度電気関係学会四国支部連合大会, 247, Sep. 2007.
111.	Satoru Tsuge, 清田啓二, Masami Shishibori, Kenji Kita, Fuji Ren, Minoru Fukumi and Shingo Kuroiwa : Analysis of Variation on Intra-Speaker's Speech Recognition Performances, 日本音響学会春季研究発表会, 165-166, Mar. 2007.
112.	Satoru Tsuge, 清田啓二, Masami Shishibori, Kenji Kita, Fuji Ren, Minoru Fukumi and Shingo Kuroiwa : Analysis of Variation on Intra-Speaker's Speech Recognition Performances, 日本音響学会春季研究発表会, 165-166, Mar. 2007.
113.	Satoru Tsuge, 小澤光広, Masami Shishibori, Minoru Fukumi, Fuji Ren, Kenji Kita and Shingo Kuroiwa : A Japanese Specific Speakers' Corpus Long Period, 日本音響学会2006年秋季研究発表会, 277-278, Sep. 2006.
114.	Satoru Tsuge, Masami Shishibori, Fuji Ren, Kenji Kita and Shingo Kuroiwa : Investigation on variability of performance on speaker depend speech recognition, 日本音響学会春季研究発表会, 157-158, Mar. 2006.
115.	宮城暖, 小泉大地, Masami Shishibori, Satoru Tsuge and Kenji Kita : WWW検索システムにおける分野別URLデータベースを用いたナイーブ・ベイズ選定手法, 言語処理学会第12回年次大会, 652-655, Mar. 2006.
116.	竹安真紀夫, Masami Shishibori, Satoru Tsuge and Kenji Kita : WWW画像検索システムにおける再検索型検索質問拡張に基づくフィードバック検索, 言語処理学会第12回年次大会, 785-788, Mar. 2006.
117.	Satoru Tsuge, Masami Shishibori, Fuji Ren, Kenji Kita and Shingo Kuroiwa : A study of the suppression method of speech recognition performance against the utterance variation on the long and short period in a speaker, 日本音響学会春季研究発表会, 221-222, Sep. 2005.
118.	大西泰代, Masami Shishibori, Satoru Tsuge and Kenji Kita : 距離尺度にEarth Mover's Distanceを用いたハミングによる類似音楽検索手法, FIT2005, 271-272, Sep. 2005.
119.	小泉大地, Masami Shishibori, 中川嘉之, Satoru Tsuge and Kenji Kita : WWW画像検索システムにおける有害画像フィルタリング手法, FIT2005, 45-46, Sep. 2005.
120.	Masami Shishibori, Satoru Tsuge and Kenji Kita : WWW画像検索システムにおける検索質問拡張に基づくフィードバック検索, FIT2005, 43-44, Sep. 2005.
121.	Satoru Tsuge, Masami Shishibori, Kenji Kita, Fuji Ren and Shingo Kuroiwa : A study of the utterance variation on the long and short period in a speaker, 日本音響学会春季研究発表会, 129-130, Mar. 2005.
122.	小泉大地, 大西正裕, Satoru Tsuge, Masami Shishibori and Kenji Kita : 言語メタデータに基づく画像コンテンツの検索, 言語処理学会第11回年次大会, Mar. 2005.
123.	Satoru Tsuge, Kazuma Hara, Kenji Kita and Shingo Kuroiwa : Blind equalization method with multi-refrence cepstrum for DSR front-end, 日本音響学会春季研究発表会, 169-170, Mar. 2004.
124.	Kenji Kita, Masami Shishibori and 岩田稔久 : 写真特徴に基づく顔画像検索, Japanese Journal of Science and Technology for Identification, Vol.8, No.Supplement, 153, Nov. 2003.
125.	Kazuma Hara, Satoru Tsuge, Masami Shishibori, Kenji Kita and Shingo Kuroiwa : A Study on Real-time Cepstral Variance Normalization, 日本音響学会秋季研究発表会, 151-152, Sep. 2003.
126.	Satoru Tsuge, Kazuma Hara, Kenji Kita and Shingo Kuroiwa : Real-time Channel Compensation Method for ETSI standard DSR Front-end, 日本音響学会秋季研究発表会, 111-112, Sep. 2003.
127.	光宗朋宏, Satoru Tsuge, Shingo Kuroiwa, Masami Shishibori and Kenji Kita : 話者情報を用いた映像検索システム, 情報科学技術フォーラム(FIT)2003, 261-262, Sep. 2003.
128.	Satoru Tsuge, Kazuma Hara, Shingo Kuroiwa and Kenji Kita : A evaluation of ETSI STQ DSR Advanced Front-End on Japanese speech corpus, 日本音響学会春季研究発表会, 57-58, Mar. 2003.
129.	井内厚志, Satoru Tsuge, Shingo Kuroiwa and Kenji Kita : 発話様式の変動に頑健な音声認識に関する研究, 平成14年度電気関連学会四国支部連合大会, 303, Oct. 2002.
130.	原一眞, Satoru Tsuge, Shingo Kuroiwa and Kenji Kita : 判別分析手法を用いた音声認識精度向上に関する研究, 平成14年度電気関連学会四国支部連合大会, 302, Oct. 2002.
131.	Satoru Tsuge, Shingo Kuroiwa and Kenji Kita : ETSI標準分散音声認識フロントエンドのための回線特性正規化手法, 日本音響学会秋季研究発表会, 117-118, Sep. 2002.
132.	Satoru Tsuge, Shingo Kuroiwa and Kenji Kita : ETSI標準分散音声認識フロントエイドのための回線特性正規化手法, 日本音響学会2002年秋季研究発表会, 117-118, Sep. 2002.
133.	Satoru Tsuge, Shingo Kuroiwa, Fuji Ren and Kenji Kita : ETSI標準分散音声認識フロントエンドを用いた音声認識実験, 日本音響学会春季研究発表会, 171-172, Mar. 2002.
134.	Satoru Tsuge, Shingo Kuroiwa, Fuji Ren and Kenji Kita : 音素依存線形判別分析の検討, 日本音響学会秋季研究発表会, 177-178, Oct. 2001.
135.	Kuldip Paliwal, Satoru Tsuge, Harald Singer, Toshiaki Fukada and Kenji Kita : A Comparison of Signal Processing Front Ends for Non-Read Speech Recognition, 日本音響学会秋季研究発表会, 5-6, Sep. 1997.
136.	Kenji Kita, 尾本貴志, 福井義和 and Yoneo Yano : 一般化相互情報量を用いた日本語形態素解析の予備検討, 平成5年度電気関係学会四国支部連合大会, 303, Oct. 1993.
137.	Kenji Kita, 芦辺和彦 and Yoneo Yano : 逆文法のLR状態遷移表を用いた構文制-One-Pass DP連続音声認識アルゴリズム, 平成5年度電気関係学会四国支部連合大会, 302, Oct. 1993.

Et cetera, Workshop:

1.	Ryota Ohga, Kazuyuki Matsumoto, Minoru Yoshida and Kenji Kita : ニコニコデータセットを用いた視聴者の感情を考慮した動画推薦アルゴリズムの提案, IDRユーザフォーラム 2022, Dec. 2022. (Keyword) natural language processing / artificial intelligence / classification / deep learning
2.	Daiki Mori, Kazuyuki Matsumoto, 篠山学, Minoru Yoshida and Kenji Kita : 個性を反映した対話モデルにおける発話文の感情及び性格推定, IEICE Technical Report, 34-38, Oct. 2022. (Summary) One of the factors that enable humans to interact with AI in a more friendly manner is personality, and since the personalityrequired of AI varies depending on the situation and context, a mechanism is needed that allows AI to switch between different personalities depending on the person with whom he or she is interacting. In this study, we use a Japanese language Transformer Encoder-decoder dialogue model (japanese-dialog-transformers) and aim to construct a dialogue model that reflects the personality of the target person by fine-tuning the interview dialogue data. By analyzing the output of the proposed dialogue models using the MBTI estimation model and the sentiment analysis model, we evaluate the consistency of the personality and discuss the differences and similarities among the dialogue models. (Keyword) interview dialogue / personality / personality estimation / dialogue model / sentiment analysis
3.	Kenji Kita, Kazuyuki Matsumoto, Minoru Yoshida, Satoru Tsuge, Norihide Kitaoka and 武田一哉 : ビット演算に基づく高速な音声ドキュメント検索語検出, 第8回音声ドキュメント処理ワークショップ, Mar. 2014.
4.	Kenji Kita and 肖清梅 : オーディオ指紋検索に適した高速なハミング空間検索, 情報処理学会音楽情報科学研究会, May 2011.
5.	柘植覚, 三好真人, 肖清梅 and Kenji Kita : フィルタバンク特徴量とEarth Mover's Distanceを用いた音楽検索, 情報処理学会音楽情報科学研究会, Feb. 2011.
6.	山本一晴, Masami Shishibori, Satoru Tsuge and Kenji Kita : 出現URLの類似性に着目したWWW空間からの関連語自動収集手法, 情報処理学会研究報告, No.2005-NL-170, 127-134, Nov. 2005.
7.	竹安真紀夫, Masami Shishibori, Satoru Tsuge and Kenji Kita : WWW画像検索システムを用いた関連語の自動収集手法, 情報処理学会研究報告, No.2005-NL-170, 45-51, Nov. 2005.
8.	小泉大地, 大西正裕, Satoru Tsuge, Masami Shishibori and Kenji Kita : 画像領域に対する言語メタデータ自動付与手法, 画像の認識・理解シンポジウム(MIRU2005), 1436-1441, Jul. 2005.
9.	西川伸紀, Masami Shishibori, Satoru Tsuge and Kenji Kita : 検索質問と字幕との文字画像特徴量間の距離に基づく映像シーン検索手法, 画像の認識・理解シンポジウム(MIRU2005), 707-714, Jul. 2005.
10.	Satoru Tsuge, Shingo Kuroiwa, Kazuma Hara and Kenji Kita : Real-time Frequency Characteristic Normalization for ETSI DSR Front-end, 音声言語情報処理研究会, Vol.49, No.3, 13-18, Dec. 2003.
11.	Takashi Otani, Satoru Tsuge, Masami Shishibori and Kenji Kita : A Method for Analysis of Table Contents of HTML Format and Its Application, 自然言語処理学会研究報告, 137-144, Mar. 2003.
12.	Isao Kohda, Satoru Tsuge, Masami Shishibori and Kenji Kita : Dimensionality Reduction and Query Expansion of Information Retrieval using Non-negative Matrix Factorization, 自然言語処理学会研究報告, 17-22, Mar. 2003.
13.	Shingo Kuroiwa, Satoru Tsuge, 田仁宏典, Tai Xiaoying, Masami Shishibori and Kenji Kita : Simple PCAを用いたベクトル空間情報検索モデルの次元削減, 電子情報通信学会技術報告, Vol.NLC2001-17, 61-66, Jul. 2001.
14.	Satoru Tsuge, Masami Shishibori and Kenji Kita : Non-negative Matrix Factorization を用いた情報検索, 情報処理学会研究報告, Vol.NL142, No.1, 1-6, Mar. 2001.
15.	Satoru Tsuge, Masami Shishibori and Kenji Kita : サポートベクターマシンによる適合性フィードバックを用いた情報検索, 情報処理学会研究報告, Vol.NL141, No.14, 83-88, Jan. 2001.
16.	辻井潤一, 金沢誠, 石崎雅人, 伝康晴, Kenji Kita and 徳永健伸 : 【座談会】言語と計算をめぐって, UP, Vol.28, No.12, 1-16, Dec. 1999.

Report:

1.	Kenji Kita : 意味空間への情報マッピングに基づくクロスメディア検索に関する研究, 2005∼2008年度科学研究費補助金研究成果報告書(基盤研究(B)), May 2009.
2.	Shingo Kuroiwa, Kenji Kita, Fuji Ren and Satoru Tsuge : 分散型話者照合方式に関する研究, 平成14年度∼平成16年度科学研究費補助金(基盤研究(B)(2))研究成果報告書, Vol.14350204, 1-227, May 2005.

Patent:

1.	Kenji Kita, Satoru Tsuge, 武田一哉 and 渡邉恵理子 : ハミング空間検索装置，ハミング空間検索方法，ハミング空間検索プログラムおよび記録媒体, 5599363 (Aug. 2012).
2.	Nagai Akito and Kenji Kita : SPEECH PROCESSING USING AN EXPANDED LEFT TO RIGHT PARSER, 6,058,365 (May 2000).
3.	Kenji Kita and 増川佐知子 : 画像処理装置, P2020-129514 (Jul. 2020), P6860254 (Mar. 2021).
4.	Kenji Kita and 柘植覚 : ハミング空間検索装置，ハミング空間検索方法，ハミング空間検索プログラムおよび記録媒体, P2014-168401 (Aug. 2014), P2014-238884A (Dec. 2014), JP6031475B (Nov. 2016).
5.	Masami Shishibori, Kenji Kita and Satoru Tsuge : 類似音楽検索装置，類似音楽検索方法，類似音楽検索プログラム及びコンピュータで読み取り可能な記録媒体, 2007-175607 (Jul. 2007), 2007-175607 (Jul. 2007).
6.	Masami Shishibori and Kenji Kita : クライアントデータベース構築方法，データ検索方法，データ検索システム，データ検索フィルタリングシステム，クライアントデータベース構築プログラム，データ検索プログラム，データ検索フィルタリングプログラム及びプログラムを格納したコンピュータで読み取り可能な記録媒体並びに記録した機器, 2006-071844 (Mar. 2006), 2006-071844 (Mar. 2006).
7.	Kenji Kita and 幸山秀雄 : 情報検索方法，情報検索装置，情報検索プログラム及びコンピュータで読み取り可能な記録媒体並びに記録した機器, 2004-352080 (Dec. 2004), 2004-352080 (Dec. 2004).
8.	持橋大地, Kenji Kita and 菊井玄一郎 : データ検索方法及びコンピュータプログラム, 2004-210243 (Jul. 2004), 2004-210243 (Jul. 2004).
9.	Kenji Kita, Masami Shishibori and Shunichiro Oe : Method for retrieving data, apparatus for retrieving data, program for retrieving data, and medium readable by machine, US20050086210A1 (Mar. 2004), US20050086210A1 (Mar. 2004).
10.	Kenji Kita, Masami Shishibori and Shunichiro Oe : データ検索方法，データ検索装置，データ検索プログラムおよびコンピュータで読み取り可能な記録媒体, 2003-174078 (Jun. 2003), 2003-174078 (Jun. 2003).
11.	Kenji Kita and Masami Shishibori : データマッチング方法，データマッチング装置，データマッチングプログラムおよびコンピュータで読み取り可能な記録媒体, 2002-204306 (Jul. 2002), 2002-204306 (Jul. 2002).
12.	永井明人, Kenji Kita and 嵯峨山茂樹 : 連続音声認識装置, (Nov. 1990), (Jun. 1992), 1874952 (Sep. 1994).
13.	坂野俊哉 and Kenji Kita : 音声認識候補絞り込み方式, 003991 (Jan. 1990), 208098 (Sep. 1991), 2120567 (Mar. 1996).
14.	Kenji Kita, 川端豪 and 森元逞 : 音声認識装置におけるビーム制御方式, 121945 (1989), 1915563 (Jul. 1995).
15.	Kenji Kita, 川端豪 and 斎藤博昭 : 連続音声認識装置, 266472 (1988), 1917600 (Mar. 1995).

Grants-in-Aid for Scientific Research (KAKEN Grants Database @ NII.ac.jp)

数値および専門性と話題性に着目したSNSからの科学的知識抽出に関する研究 (Project/Area Number: 21K12141 )
異なるコミュニティにおいて誤解されやすい表現の感性的言語変換に関する研究 (Project/Area Number: 20K12027 )
Analysis of lifestyle and health texts on the WWW consisting of text and numeric data. (Project/Area Number: 18K11549 )
Integrated Media Analysis for Medical/Health Information Matching on WWW (Project/Area Number: 15K00425 )
Development of the support system for Japanese calligraphy instruction by using image processing technologies (Project/Area Number: 24501136 )
To Create Robot Mind and Apply it to Nursing Robots (Project/Area Number: 22240021 )
Development of the Telop Retrieval System for Educational Video Data (Project/Area Number: 21500940 )
Semantic and Affective Multimedia Retrieval using EEG-based Biological Information (Project/Area Number: 21300036 )
拡張サフィックス・アレイを用いた語学学習用作文支援システムに関する研究 (Project/Area Number: 20650143 )
統計的手法を用いた知識情報モデルに基づく知的信号処理手法に関する研究 (Project/Area Number: 17656128 )
Development of the Full Text Search System for Presentation Movies Using the Video Processing Technology (Project/Area Number: 17500644 )
Analysis of Intra-Speaker Variation and Development of Distributed Speaker Recognition System (Project/Area Number: 17300065 )
A Study on Cross-media Information Retrieval based on Semantic Information Mapping (Project/Area Number: 17300036 )
Research for a method of analysis for emotion expressions with relation of semantic collocation and an application of retrieval for emotion information (Project/Area Number: 15300072 )
Machine-Aided Writing System based on Super-Function and Intellectual Corpus (Project/Area Number: 14380166 )
A Study of Distributed Speaker Recognition Methods (Project/Area Number: 14350204 )
自然言語処理技術を用いた理科教授学習システムに関する研究 (Project/Area Number: 14022237 )
Rule Insertion and Extraction in Evolutionary Neural Networks for Image Retrieval (Project/Area Number: 13680448 )
Study of Class-based Language Model and its Application to Japanese Morphological Analysis (Project/Area Number: 10680383 )
A fast n-gram full text search independent of document size and it application to a huge text base (Project/Area Number: 10480082 )
Study of Effective Speech Recognition based on the Bi-directional Search Algorithm (Project/Area Number: 07680401 )
確率・統計的手法による対話のモデル化とコーパスからの自動生成に関する研究 (Project/Area Number: 07221209 )
統計的言語情報を用いた音声認識の精度向上に関する研究 (Project/Area Number: 05780292 )
Development of an Environmental ICAI System for English Conversation Learning using Speech Recognition and Synthesis Techniques (Project/Area Number: 05558015 )
Search by Researcher Number (10243734)

Professor Emeritus : Kita, Kenji

Research

Personal Web Page

Field of Study

Subject of Study

Book / Paper

Book:

Academic Paper (Judged Full Paper):

Academic Paper (Unrefereed Paper):

Academic Letter:

Review, Commentary:

Proceeding of International Conference:

Proceeding of Domestic Conference:

Et cetera, Workshop:

Report:

Patent:

Grants-in-Aid for Scientific Research (KAKEN Grants Database @ NII.ac.jp)