[3Xin2] Poster session 1

[3Xin2-55] The Effect of Ruminative Replies in Human-LLM Agent Interactions on Increasing Familiarity

〇Tatsuki Serizawa¹ (1.ARISE analytics, inc.)

Keywords:Large Language Model, Dialog Agent, Trust, Parroting, Human Agent Interaction

One key element in building trust in dialog agents using large language models (LLMs) is to foster a sense of familiarity with the user. This study focuses on the technique of 'parroting' (rumination), a skill in listening, as one method to create this sense of familiarity. We hypothesized that an agent's ruminative responses to user statements contribute to the formation of familiarity and conducted tests to verify this hypothesis. The results showed that agents engaging in ruminative responses were more effective in fostering a sense of familiarity with users compared to those that did not. This is thought to be due to the sense of empathy created by the ruminative responses. Additionally, it was suggested that the act of rumination could induce an effect similar to step-by-step prompt engineering in LLMs, potentially leading to more in-depth conversations.