Advice Generation Using Influence Estimation on the Utterances of Elementary School Teachers
Sakuei Onishi, Hiromitsu Shiina, Tomohiko Yasumori · 2024
For busy teachers, there is an urgent need to develop a mechanical method to support teachers' reflections on their lessons. In this study, we propose a method of advice generation that uses influence estimation on teachers' utterances with the aim of assisting in the reflective activities of teachers. Influence estimation leverages the Attention mechanism of a Large Language Model (LLM), and specific advice is generated based on the results. An evaluation of the experimental results suggests that it may be possible to generate useful advice by utilizing influence estimation on utterances.