How Vision-language Models Can Support Eye-tracking Analysis in Mathematics Education: a Methodological Study on Number Line Activities

Asghari, Parviz; Schindler, Maike; Schulte, Franziska; J. Lilienthal, Achim

Abstract

Eye tracking (ET) is a valuable tool for identifying students’ learning processes, also in number line activities. While prior research has shown that AI can successfully identify student number line processes, the approaches operate as a “black box” and require extensive data collection and manual labelling. Vision-Language Models (VLMs) offer a promising, generally applicable, alternative. They can analyze ET data without prior labelling by leveraging domain knowledge. The aim of this study is to investigate how effectively VLMs can identify student number line processes. Based on an ET study with 136 fifth graders, we found that our VLM framework VETA (VLMs for ET Analysis) achieved very high agreement with the researchers’ manual coding (Cohen’s κ ≥ 0.80 – 1.0), demonstrating reliability and potential for ET analyses.

Keywords

AIEye TrackingNumber line

Recommended citation

Asghari, P., Schindler, M., Schulte, F., & J. Lilienthal, A. (2026). How Vision-language Models Can Support Eye-tracking Analysis in Mathematics Education: a Methodological Study on Number Line Activities. In Krzywacki H., Hähkiöniemi M., Haataja E., Hannula M.S., Hästö P. (Eds.), Proceedings of the International Group for the Psychology of Mathematics Education, Volume 1 (pp. 83–90). Psychology of Mathematics Education (PME). ISSN 2790-3648.

BibTeX RIS

← All PME 49 Research Reports