Automated Detection of Students’ Gaze Interactions in Collaborative Learning Videos: A Novel Approach

Abstract

Gaze behaviours have been considered important social signals to explore human learning. Over the past decade, previous research showed positive relationships between certain features of gaze behaviours and the quality of collaborative learning. However, most studies focus on detecting students’ gaze behaviours with eye-tracking tools which are costly, logistically challenging, and can be obtrusive in real-world physical collaboration spaces. This study presents a novel approach to detecting students’ gaze behaviours from videos of real-world collaborative learning activities. Pre-trained computer vision models were used to detect objects on the scenes, students’ faces, and their gaze directions. Then, a rule-based approach was applied to detect gaze behaviours that are associated with peer communication and resource management aspects of collaborative learning. In order to test the accuracy of the proposed approach, twenty collaborative learning sessions, each lasting from 33 min to 67 min, from five groups in a 10-week-long higher education course were analysed. The results showed that the proposed approach achieves 66.57% overall accuracy at automatically detecting students’ gaze interactions in collaborative learning videos. The implications of these findings for supporting students’ collaborative learning in real-world technology-enhanced learning environments are discussed.