Preliminary Study on Topic Change by Conversational Agents Based on Trend Information

Abstract

This paper proposes a method for conversational agents to change topic during conversation based on trend information. Detection of dialogue breakdown is one of important research themes for developing conversational agents. However, few studies have been done on how to do after detecting dialogue breakdown. The proposed method finds a topic that relates with current topic in terms of temporal trend by using context search engine. This paper also proposes a method to detect dialogue breakdown based on response time by a human. The results of preliminary experiments are reported to examine the possibility of the proposed methods.