AI-Enhanced Multimodal Analysis of Mathematical Learning Through Participative Epistemology

• In Event: Qualitative Research and AI

Abstract

This study investigates how integrating Multimodal Large Language Models (MLLMs) with human interpretation might enhance qualitative research into embodied mathematical learning interactions. We adopted a participative epistemology approach and rhizomatic framework, iteratively collaborating with an MLLM to analyze video of high school students learning rational exponents. Our findings reveal both potential and limitations of MLLMs in capturing fine-grained multimodal details of learning. The process exposed implicit assumptions in data analysis and highlighted challenges in timestamp accuracy, diarization, and hallucination. The MLLM showed promise in collaborative inquiry, but significant human oversight remains crucial. This work offers insights for future MLLM applications in educational research and underscores the need to refine techniques for improved accuracy in AI-assisted qualitative analysis.

Authors

• Jennifer Kleiman, University of Georgia

• Xiaoming Zhai, University of Georgia

• Amy Ellis, University of Georgia