[2K4-GS-9] Human interface

[2K4-GS-9-03] Method for Building an Autonomous Dialogue Agent with an Ability for Interaction Adapted to the Characteristics of the User Community

〇Atsushi Kobayashi¹, Tomonori Kubota¹, Kohei Ogawa¹, Satoshi Sato¹ (1. Nagoya University)

Keywords:dialogue system, human-agent interaction, virtual agent, multi-party dialogue

An autonomous dialogue agent is expected to engage in live chat like Virtual YouTubers or to give classes and lectures in application of one-to-many dialogue where the agent interacts with multiple interlocutors forming a community. In one-to-one dialogue, the effectiveness has been shown that an agent interacts with its interlocutor adapting to their characteristics. However, in one-to-many dialogue with community members, no method has been established to realize an agent with the dialogue ability adapted to the characteristics of the community. In this study, we proposed a two-step method to build such an agent: (1) collecting dialogue examples from chat in which community members operate an agent to interact with the other members, (2) developing the agent by using these examples. We conducted experiments to verify this method with our developed system enabling dialogue examples collection and agent implementation. The results suggested that it is possible to collect dialogue examples that reflect the characteristics of a community and to realize an agent capable of dialogue adapted to the characteristics.