Proactive Group Learning Design Support Through Multimodal Evidence from VR Environments

Abstract

Virtual reality (VR) environments generate rich multimodal evidence that informs educational design. However, existing multimodal learning analytics either rely on black-box predictive models that lack interpretability or focus on descriptive posthoc analyses that provide insufficient support for proactive instructional design. This study aims to forge new pathways for leveraging multimodal evidence in proactive group learning design. Facial activation data from 20 undergraduate participants were collected during an immersive lecture session. Aggregated emotional intensity indicators and pairwise temporal similarity of confusion were jointly optimized through a genetic algorithm to form heterogeneous groups. Beyond grouping, pairwise temporal analysis reveals divergence and alignment patterns that inform role assignment and instructional adjustment. The illustrated case demonstrates how interpretable multimodal indicators can guide proactive, theory-informed group learning design.