Generating questions asked by facilitator agents using preceding context in web-based discussion

Discussion facilitation is important to promote the constructive discussion. Our goal is to develop autonomous facilitator agents for facilitating Web-based discussion. In previous studies, analysis results of human facilitator utterances in a real-world public debate showed the importance of facilitator questioning. In this paper, we propose a method for generating facilitator questions from extracted opinions of discussion participants in the preceding context. First, the opinions in the preceding context are extracted using clue expressions. A facilitator's question is generated with pattern-matching rules using the case structure of a predicate in the extracted opinion. This method is based on the assumption that an appropriated type of question can be selected using a surficial case structure. We evaluated our method through a subjective experiment. The results show that our method has the potential to be utilized for developing autonomous facilitator agents.