Analyzing Collaborative Learning in Peer Education Using Deep Graph Neural Networks

Abstract

Peer education has gained importance in the contemporary education systems because of its capacity to enhance student interaction, engagement and performance at school. Nonetheless, the conventional ways of evaluation do not adequately represent the dynamic and detailed associations among learners. This paper overcomes this limitation and presents a Deep Graph Neural Network (DGNN)-based model to investigate the analysis of collaborative learning in peer education settings. The research technique will be to model the relationships of learning, peer pressure, and group dynamics by using the graph structure to model the students as nodes and their interactions as edges. The data is comprised of 2300 student nodes with interaction weighted edges and a weekly time window that represents temporal learning behaviour. The proposed model combines the graph representation learning, attention systems and temporal feature modelling to enhance predictive performance. As experiment results demonstrate, the baseline model will have an accuracy of 0.74 and F1-score of 0.72, whereas the introduction of Graph Neural Networks will lead to an increase in the accuracy to 0.87 and F1-score to 0.85. Additional improvements with temporal characteristics bring the results to 0.89 accuracy and 0.88 F1-score. Attention-based GAT model has the highest accuracy of 0.91 and F1-score of 0.90, whereas the full DGNN proposed model has the highest accuracy of 0.93 and F1-score of 0.95. The findings show that the combination of graph learning, temporal dynamics and attention mechanisms to collaborative learning analysis shown considerable positive impact. The research reaches the conclusion that DGNN is a powerful and efficient model to study the patterns of peer interaction, to find the influential learners and to predict education performance in the collaborative learning setting.