Estimating the Impact of Classroom Speech Using a Large Language Model

Teachers in elementary and junior high schools in Japan have little time for reflection and other training, and there is an urgent need to develop a system for analyzing lessons. If the impact of the teacher's and children's speech in class could be numerically determined, this would lead...

Full description

Saved in:
Bibliographic Details
Published in2024 16th IIAI International Congress on Advanced Applied Informatics (IIAI-AAI) pp. 409 - 414
Main Authors Onishi, Sakuei, Kojima, Shousei, Shiina, Hiromitsu, Yasumori, Tomohiko
Format Conference Proceeding
LanguageEnglish
Published IEEE 06.07.2024
Subjects
Analytical models
Classroom Utterance Analysis
Correlation
Estimation of Utterance Impact
Global Variational Transformer (GVT)
Indexes
Large Language Model (LLM)
Large language models
Numerical models
Reflection
Training
Transformers
Tuning
Visualization
Online AccessGet full text
DOI10.1109/IIAI-AAI63651.2024.00081

Cover

More Information
Summary:Teachers in elementary and junior high schools in Japan have little time for reflection and other training, and there is an urgent need to develop a system for analyzing lessons. If the impact of the teacher's and children's speech in class could be numerically determined, this would lead to class improvements. In this study, fine tuning of a large language model was performed using LoRA to imitate teachers and children, and the Attention Weight and the variation of the generation probability obtained from the generation of utterances by the model are used to estimate the impact received and given by teachers' and children's utterances.
DOI:10.1109/IIAI-AAI63651.2024.00081