Adaptive understanding of proposal-requesting expressions for conversational information retrieval system

This paper considers a conversational system in which information is provided in accordance with the conditions presented by the user, and proposes a method that can adequately deal even with unknown expressions. In most conventional systems, the relation between the expression and the intention of the utterance by the user is built into the system beforehand. Thus, it is difficult to deal adequately with unknown expressions which have not been learned. We propose a framework which adaptively manages on-line the relation between the expression and the intention by interaction with the user. The proposed method produces a framework in which the connection between the expression and the intention is dynamically modified according to the explicitness or implicitness of the affirmative or negative attitude shown by the user to the proposal made by the system. It is verified by an evaluation experiment that the system can adequately learn the relation between the expression and the intention of the user by the proposed method, and can deal adequately with unknown expressions.