Research Support Office Research Advancement Division. Tokyo University of Agriculture and Technology
TEL | +81-42-367-5944 |
---|---|
FAX | +81-42-367-5946 |
Home > Introduction of our tenure-track faculties > Komiya Kanako
Komiya Kanako
Affiliation | Institute of Engineering |
---|---|
Division | Division of Advanced Information Technology and Computer Science |
Research field | Natural language processing |
Keyword(S) | Word sense disambiguation, information extraction, domain adaptation, transfer learning |
Url | http://web.tuat.ac.jp/~komiya/main.html |
Research experience | ・Apr.2009–Mar.2010: Researcher at Precision and Intelligence Laboratory, Tokyo Institute of Technology, Japan |
---|---|
Educational background | ・Mar.2005: B. S. at Institute of Engineering, Tokyo University of Agriculture and Technology, Japan |
Awards | ・Minoru Sasaki, Kanako Komiya, Classification of Question-Answer Pairs of QA Site Using Difference of Word Delimiter, IDR User Forum 2019, (2019.11.29). (*Yahoo! Award) |
Selected papers and publications | ・Kanako Komiya, Minoru Sasaki, Hiroyuki Shinnou, Manabu Okumura, Domain Adaptation using Word Embeddings for Word Sense Disambiguation, Journal of Natural Language Processing, Vol.25, No.4 pp.463-480,(2018.9). |
Natural language processing is processing of languages such as English and Japanese using computers. This research area is included in research on artificial intelligence. We use machine learning, a technique where computers automatically find some rules from many examples. Our main tasks are word sense disambiguation, information extraction, sentiment analysis, recommendation systems, text classification, and so on.
In recent years, deep learning techniques have become the standard technique of artificial intelligence. In addition, we have researched on Japanese natural language processing for long years, with a special focus on properties inherent in Japanese. For example, word boundaries in Japanese are unspecific because Japanese does not have word delimiters between words and Japanese has homonyms and heteronyms because Japanese use ideograms. Moreover, we conduct research on cross-lingual text processing using English, taking into consideration these Japanese properties.
In addition, we focus on research on domain adaptation and transfer learning, which are the techniques to improve system performance when there are few data on a certain domain but there are many data on domains close to the domain. For example, we develop systems for blogs using newspapers and systems for old documents using contemporary documents.
This tenure track has some support systems, including start-uo grant and mentor system, I think these support systems greatly help researchers.
I will do my best and enjoy my research.