[3H1-GS-3] Data mining: Applied data mining (1)

[3H1-GS-3-04] Similarity Based Independent Topic Tracking

〇Takahiro Nishigaki¹, Kenta Yamamoto², Takashi Onoda¹ (1. Aoyama Gakuin University, 2. Aoyama Gakuin University Graduate School of Science and Engineering)

Keywords:Independent Topic Analysis, Topic Extraction, Topic Tracking, Text Mining

In this research, it is focused on independent topic analysis, which is one of the topic extraction methods. Independent topic analysis is a method that can extract independent topics from each other in document data. There are many studies on independent topic analysis. And there is a study on tracking independent topics. In that study, it was defined five topic changes based on similarity of independent topics. The defined topic changes are the integration, separation, appearance, disappearance, and resurrection of independent topics. In this paper, we evaluate the similarity that defines the topic change. The similarity is compared by three methods using cosine similarity, Euclidean distance, and important words of independent topics. The user compares and evaluates two independent topics by three similarities methods. The experimental results were showed that cosine similarity is effective. In addition, the validity of the proposed method was shown by user evaluation experiments.