Study of Discussion Record Analysis Using Temporal Data Crystallization and Its Application to TV Scene Analysis

The purpose of this project is to develop methods to analyze discussion records. We analyze discussion records by two methods – Scene analysis and Logical analysis. Scene analysis focuses on topic change in the record and divides a record into several scenes using Temporal Data Crystallization (TDC). Then, by measuring the polarity of words and the volume of utterance, we estimate the atmosphere and emotion in the scene. Logical analysis focuses on the role of phrases in utterances, and attaches logical labels to phrases based on Toulmin diagram. By these labels, we extract a logical structure of arguments in the discussion, calculate its semantics based on Theory of Computational Argumentation and evaluate various argumentation skills of the discussion record. In addition to discussion analysis, we apply the scene analysis techniques to caption data of TV programs to recognize change of scenes.