Presentation Information
[2Yin-B-08]Discussion Analysis with LLMs for Facilitation Support and Evaluation via a Consensus Game
〇Atsuki Yoshikawa1, Ryo Taguchi1 (1. Nagoya Institute of Technology)
Keywords:
LLM,Audio Data,Group Discussion,Consensus Game,Discussion Analysis
This study proposes an automatic method for evaluating the state of group discussions based on speech data, with the aim of supporting facilitation in educational settings. After transcribing speech audio recorded individually from each participant, a large language model (LLM) is used to estimate the discussion topics addressed in each utterance, infer group judgment trends during the discussion process, and determine the presence or absence of qualitative elements in utterances based on the Instructional Quality Assessment (IQA) framework. Experimental results confirm that these analyses can be performed using an LLM. Furthermore, this study examines how these analysis results influence the progression of discussions and interactions among participants.