Generation of advice for teacher speech based on class speech analysis using the Attention mechanism and automatic evaluation by LLM

大西朔永，椎名広光，保森智彦

教育システム情報学会（JSiSE）2024年度特集論文研究会

In this study, we propose an advice generation method using the Attention mechanism to evaluate the impact of teacher utterances in order to support teachers. Specifically, a neural dialogue model between a teacher and a child is constructed, the influence of the teacher's utterances is evaluated by the Attention mechanism, and advice is generated by LLM using the evaluated influence. In addition, automatic evaluation by LLM is conducted to evaluate the effectiveness of the advice.