Enhancing cognitive diagnosis with group-aware graph augmentation and self-supervised alignment

In the field of intelligent education, cognitive diagnosis aims to assess students’ cognitive states by analyzing their historical response data, serving as a foundation for supporting downstream tasks such as personalized learning recommendations. However, in real learning scenarios, the distribution of student response data is often highly imbalanced. Students with fewer responses suffer from data sparsity, leading to inaccurate diagnoses, which poses a major challenge for existing models. To this end, we propose a Group-Aware Cognitive Diagnosis (GACD) framework. The core idea is to group students with similar answering patterns together. Within each group, we artificially creates additional “peer interactions” for students with sparse data, effectively giving the model more information to learn from. Notably, as correct and incorrect responses reflect distinct behavioral patterns, we first decompose the interaction graph into two separate subgraphs based on response types and learn representations from each independently. To ensure accurate modeling of students’ cognitive states, we introduce a self-supervised strategy to align node representations between the original and augmented graphs. Finally, the learned representations are integrated into existing cognitive diagnosis models. Extensive experiments on three real-world datasets demonstrate that the GACD framework significantly improves prediction performance. Not applicable.