Feature Expression Extraction from Discussion Facilitators' Utterances in Web-Based Forum System towards Autonomous Facilitator Agents

Discussion facilitators play an important role for better consensus building in public debates because diverse stakeholders addressing public issues have diverse view points. We investigated how facilitators facilitate public debate with a statistical analysis approach using a public debate corpus collected by a Web-based forum system. This debate corpus consists of over one thousand utterances posted by debate participants and facilitators written in Japanese. To investigate linguistic expressions suitable for facilitating public debate, we extracted feature expressions of facilitators' utterances with focusing on several contextual features. The results show that facilitators' questioning plays an important role for avoiding stagnation of discussion and for increasing the number of participants' utterances. We are planning to apply the extracted feature expressions to generating utterances of autonomous facilitator agents in future work.