[2F6-GS-5] Agents:

[2F6-GS-5-05] Application of LLM-based Discussion Agents to the Discussion Support System AIR-VAS

〇Arisa Naka¹, Mamoru Yoshizoe¹, Hiromitsu Hattori¹ (1. RitsumeikanUniversity)

Keywords:Discussion Support, Agent, Large language Model

In today's increasingly diverse society, a broad perspective and flexible thinking are required in a variety of situations. We are developing a discussion support system, AIR-VAS, with the aim of supporting awareness of diverse values in discussions and dialogues. AIR-VAS has been designed to support group discussions in which multiple groups engage in discussion at the same time, and has been implemented with a function to extract and present stimulus information from other groups to make the participants aware of the value of the discussion. In this paper, we introduce a discussion agent based on large language models (LLM) to AIR-VAS with the aim of easing the usage environment of the system. Based on the contents of previous discussions, we implemented an agent that performs utterances and information presentation as a member of discussion participants. Experimental results confirmed the agent's ability to generate natural speech sentences and to output characteristic speech content based on the speaker type assigned to the agent.