Multimodal Dialogue Breakdown Detection Considering User’s Personality Traits

Abstract

Automated dialogue systems have become familiar in recent years, but dialogue breakdowns are still occurring. Therefore, researchers are carrying out dialogue breakdown detection. Previous research reported that individual differences were included in responses to dialogue breakdown, but few studies detected dialogue breakdown considering individual differences in user responses. It is crucial to carry out the detection considering individual differences because such differences may lower the detection accuracy. In this study, we propose a method of dialogue breakdown detection by focusing on individual characteristics which are thought to be related to emotional changes and considering personality traits. Specifically, users are clustered based on personality traits, and a detector is constructed for each cluster. As a result of the experiment, it became clear that the breakdown detection accuracy and the expression method of the breakdown reaction were different for each cluster.