カウンセリングデータにおけるトピックモデルを用いた文書分類

カテゴリ: 論文誌(論文単位)

グループ名: 【C】電子・情報・システム部門

発行日: 2016/02/01

タイトル(英語): A Text Classification for Counseling Data Using Topic Model

著者名: 単　壮（名古屋工業大学大学院工学研究科情報工学専攻），加藤　昇平（名古屋工業大学大学院工学研究科情報工学専攻）

著者名(英語): Zhuang Shan (Dept. of Computer Science and Engineering, Graduate School of Engineering, Nagoya Institute of Technology), Shohei Kato (Dept. of Computer Science and Engineering, Graduate School of Engineering, Nagoya Institute of Technology)

キーワード: 文章分類，潜在的ディリクレ配分法，カウンセリング　　Text classification，Latent dirichlet allocation，Counseling

要約(英語): In order to build a highly efficient psychological counseling system, we provide a text classification method to aid the system in processing the complicated psychological knowledge base. This paper presents a method of text classification, which include LDA, morpheme parse, and majority voting. We believe that by classifying the user's question, it will increase the precision of finding the relevant answer to the user's question. We firstly collected three categories of psychological problem texts which are love-related, interpersonal relationship, self-knowledge which are questioned most commonly as training data and texting data. And all of the texts have tags that show their categories. Then we use these data to train LDA to obtain the most usable topic distributions of each category and most usable word distributions of each topic. We use a majority voting method to classify category unknown input with using these topic distributions and word distributions. We also did comparison experiments which were based on TF・IDF and SVM so as to identify the validity of our approach. The experimental results demonstrated the feasibility and effectiveness of our approach. According to the classification experiment, precision rate of our approach exceeds 80.3%, while the two comparison methods got 62.6% and 68.0%. We think by using our approach, counseling system can provide more accurate and effective answer to user.

本誌: 電気学会論文誌C（電子・情報・システム部門誌） Vol.136 No.2 （2016）

本誌掲載ページ: 226-232 p

原稿種別: 論文／日本語

電子版へのリンク: https://www.jstage.jst.go.jp/article/ieejeiss/136/2/136_226/_article/-char/ja/