Construction of a conversation system for robots considering the relationships among three parties

Abstract

The purpose of this paper is to construct a conversational system for robots considering the relationships among three parties. Although research has been conducted on robots that recognize relationships based on the frequency and time of conversations, there has been no research on recognizing relationships in real time based on the content of conversations or on changing robot behavior based on relationships. In this paper, we propose a system called RelBot (dialogue Relational roBotic system) that uses LLM to estimate the current relationship and ideal balance relationship in real time from the content of a conversation in a three-way conversation between one human and two robots, and generates statements to establish the ideal balance relationship. The ideal balance relationship in this paper refers to the relationship desired by humans, adjusted to satisfy the balance state of balance theory. To verify the effectiveness of RelBot, we conducted an evaluation experiment on the accuracy of estimating the current relationship and the human desired relationship, as well as the effectiveness of statements to establish the ideal balance relationship. The results of the experiment showed that the current relationship and the human desired relationship can be estimated with high accuracy. Furthermore, the effectiveness of statements to establish the ideal balance relationship was revealed.