Automatic Evaluation of Discussion Quality Using Topic Relevance and Participants' Performance

People often have discussions with others in working cooperatively. Discussions are conducted with given topics. Participants in discussions give their opinions to have an agreement on an issue. They listen to others' opinions and give their views. The duration of a discussion is limited, so a discussion can not be held for a long time. Since the participants have to get an agreement in a limited duration, the quality of discussion must keep high. However, it is difficult for the participants to assess the quality of the discussion. An automatic evaluation method for the quality of discussion is required. This paper proposes an automatic evaluation method of the discussion quality. We assume if a discussion has much information related to discussion topics, the quality of the discussion must be high. We also assume if all participants give their opinions related to the discussion topics, the quality of the discussion must be high. Based on the assumptions, the proposed method takes discussion texts and keywords related to discussion topics to evaluate topic relevance and participant's performance. The proposed method generates an equation of the discussion quality using multiple regression analysis. If a new discussion is given, the generated equation can automatically assess the discussion's Quality. In evaluation experiments, we used 20 discussion texts and obtained an equation for the quality of discussion. We found that the obtained equation could assess the quality of discussion with high accuracy (a multiple $R$ was 0.92).