GraphScholar: Discovering Student Learning Patterns Through Multi-Graph Neural Networks and Clustering

This research addresses the lack of unsupervised graph-based approaches for discovering student learning patterns by introducing GraphScholar. It combines student-level feature engineering, student–student and student–module graphs, GraphSAGE, GCN, and GAT, and evaluates their representations using K-Means, Agglomerative, and Spectral clustering with multiple metrics. The study shows that graph representation, particularly the student–module GAT representation, can substantially improve student grouping and reveal meaningful behavioral profiles.