Discussion Analysis Using Temporal Data Crystallization

This paper introduces a discussion analysis tool which extracts topic flow and important utterances from a discussion record based on word occurrences. We have proposed a discussion analysis method called Temporal Data Crystallization (TDC). This method divides the entire discussion record hierarchically at points where the topic changes, and analyzes some features of flow of topics for each period. In this paper, we showed the effect of hierarchical division by analyzing an example discussion record. Then, we introduced the extension of TDC by considering nonverbal information such as actions, facial expression, loudness of voice, and so on.