What Does Shared Understanding in Students’ Face-to-Face Collaborative Learning Gaze Behaviours “Look Like”?

Abstract

Several studies have shown a positive relationship between measures of gaze behaviours and the quality of student group collaboration over the past decade. Gaze behaviours, however, are frequently employed to investigate i) students' online interactions and ii) calculated as cumulative measures of collaboration, rarely providing insights into the actual process of collaborative learning in real-world settings. To address these two limitations, we explored the sequences of students' gaze behaviours as a process and its relationship to collaborative learning in a face-to-face environment. Twenty-five collaborative learning session videos were included from five groups in a 10-week post-graduate module. Four types of gaze behaviours (i.e., gazing at peers, their laptops, tutors, and undefined objects) were used to label student gaze behaviours and the resulting sequences were analyzed using the Optimal Matching (OM) algorithm and Ward's Clustering. Two distinct types of gaze patterns with different levels of shared understanding and collaboration satisfaction were identified, i) peer-interaction focused (PIF), which prioritise social interaction dimensions of collaboration and ii) resource-interaction focused (RIF) which prioritise resource management and task execution. The implications of the findings for automated detection of students' gaze behaviours with computer vision and adaptive support are discussed.

Harnessing Transparent Learning Analytics for Individualized Support through Auto-detection of Engagement in Face-to-Face Collaborative Learning

“…This indicates that students who expressed their ideas and opinions more frequently were likely to report they are more active in collaborative learning. This alignment is important to illustrate the validity of the proposed approach to detect student engagement in collaborative learning using automated measures of speech detection and gaze behaviours [51].…”

“…It is argued that conducting multilevel analysis, both individual-level and group-level, is helpful in the adoption of a learning analytics system in collaborative learning [19]. Many studies only used group-level measurements, such as the quality of the collaboration productions [50], and human-observed collaboration quality [41], as their target constructs in the analysis. Given the fact that group-level analysis only provides information about the learning performance as a collaborative group, it is a challenge to provide individual learning suggestions for students in practice.…”

“…The pattern of these gaze behaviours has been illustrated to have the potential to distinguish the students who exhibited dierent learning processes in collaborative learning. [50] Also, these gaze behaviours can be automatically detected by computer vision algorithms with acceptable accuracy [48]. In order to work with a more accurate ground truth to answer the research questions posed, in this study, the gaze behaviours were annotated manually by two researchers.…”

“…Then, the group interactions were detected based on all group members' speech behaviours and gaze behaviours. Seven types of group interactions have been identied from students' gaze behaviours and speaking behaviours using a rule-based method presented in previous research [51] which illustrated how these types of group activities can be used to distinguish the collaborative learning process based on dierent learning outcomes. When we identied the group interactions, we used a slide-window approach to transfer students' gaze behaviours at a given moment into a continuous behaviour of "paying attention to".…”

Exploration of the characteristics of teachers' multimodal behaviours in problem‐oriented teaching activities with different response levels

“…In addition, since the learning process in POT is a dynamic, emergent process that changes over time ( Cukurova et al, 2020), it can be very challenging to understand, support and facilitate use of traditional tools and methods. Advances in understanding of teacher and student behaviours in the classroom, for example sequential pattern mining of student behaviours (Kucuk & Sisman, 2017; Sisman et al, 2022; Zhou et al, 2022); process evaluation of interactive behaviours (Kent & Cukurova, 2020), analysis of multimodal behaviours using machine learning methods (Cukurova et al, 2020; Yang et al, 2020), have improved the diversity of analytical methods used to reveal the correlated and dynamic nature of behaviours. Epistemic network analysis (ENA) is a quantitative ethnographic method that can be used to build network models to describe a phenomenon (Swiecki et al, 2020).…”