[2C1-GS-6] Language media processing: corpus / conversation

[2C1-GS-6-02] A Study on Extraction of Agreeing/Disagreeing Dialogue via Chat Disentanglement

〇Yasuhiro Sogawa¹ (1. Hitachi, Ltd.)

Keywords:Information Extraction, Dialogue Structure Analysis, Chat Disentanglement

Recent dramatic advances of speech recognition enable an automatic transcription of a conversation. Due to its efficiency, we can easily store a lot of conversation logs each of which length sometimes becomes large according to the elapsed time of the conversation. Therefore we have difficulties to extract necessary information from the long conversation log. To solve this problem, various text summarization and information extraction techniques are frequently discussed. In this study, we propose a method to extract a dialogue related to agreements and disagreements which is important in business conversations. While conventional researches tend to focus on words or sentences frequently appearing or being of importance in the log, our approach aims to extract the dialogue by estimating their dialogue structure via chat disentanglement technique. We evaluate our approach with Business Scene Dialogue Corpus and show its validity.