Kazuyuki Matsumoto

Associate Professor

Tokushima University

Resarch Interests: natural language processing information science affective computing sentiment analysis

In his first and second year of graduate school (2004-2007), he worked on research with the main theme of emotion estimation from spoken sentences. Spoken sentences are natural language sentences spoken by humans. As an emotion estimation method, he proposed a rule-based method that patterns emotion expression dictionaries created manually and emotion occurrence events. Based on this research, he wrote a master's thesis and a doctoral dissertation, and obtained a doctorate (engineering). Since 2008, hehave been involved in research on campus guidance robots and development of intelligent English composition support systems as an assistant professor at the University of Tokushima Graduate School's one-year research project (Advanced Technology Education Project). During this period, he built a Japanese-English dialogue sentiment corpus, implemented an intelligent English composition support system, and experienced presentations on research results both in Japan and overseas. Since October 2009, he had been employed by Tokushima University Graduate School as an assistant professor (5-year term) and has been involved in teaching assistance for graduation and master's studies at the undergraduate level. And research on youth language using natural language processing technology. Specifically, when detecting emotional expressions from sentences, the position information (where in the sentence is located) is added to the feature information, so that attributes such as conventional part of speech of the word are featured. Enables more effective detection than the detection method using conditional random fields. In addition, by focusing on youth words that are frequently used on the Internet, called net slang, research on registering youth words that were not often registered in the dictionary in the emotional expression dictionary, I conducted research to estimate emotions. From FY2011 to FY2013, I proposed a model that estimates the emotional state by focusing on the emotions that fluctuate in the dialogue in the young researcher (B) that the applicant represented. During this period, the “Youth Language Emotion Corpus” was constructed to estimate emotions from utterances containing youth language. In this corpus, utterances collected from about 20,000 blogs and SNSs are registered, and tagging for youth words and emoticons and sentence emotion tags are added. It took about a year to build, and based on this corpus, I published two journal articles and two at international conferences. Since 2014, he has conducted machine learning experiments to semi-automate youth language emotion corpora, published one in an IEEE international journal by co-authoring, and conducted research on search methods for Japanese lyrics containing errors. Researched in collaboration with other researchers and published one in an international journal. In addition, by analyzing the interaction between characters in the play script, he studied a method to measure the intimacy between characters, and published one in a Japanese paper (in 2018, intelligence and information). Did. Furthermore, in the study of a method for converting youth words into standard words, word dispersion expressions were obtained by devising the learning source corpus, enabling conversion with higher accuracy than the baseline method. It was announced in a magazine (Japanese). In connection with this research, he researched a method to determine whether or not a word is a youth by combining it with character features such as word familiarity information and stroke count, and presented one case at an international conference in 2016, and received the Best Paper Award. Awarded. In the same year, the research results on the method of automatically estimating the poster's egogram (one of the personality diagnosis methods) by using the distributed expression of the sentences posted on Twitter and the features that characterize the formal degree of the sentences, in natural language Presented at NLP-KE, an international conference on processing and knowledge engineering, and won the Best Paper Award (this research result was published as an original paper in an international journal in 2017). From 2020, he has been employed by Tokushima University Graduate School as an associate professor.