With the increasing integration of computer technologies into education, accurately modeling students’ knowledge mastery has become a central problem in intelligent education systems. However, existing cognitive diagnosis models often suffer from sparsity in the knowledge–item association matrix (Q-matrix) and limited model capacity, which restrict their ability to capture complex student–item interaction patterns. Collaborative filtering–based approaches further exhibit insufficient capability in modeling fine-grained cognitive relationships, leading to reduced diagnostic accuracy. To address these limitations, this study proposes a cognitive diagnosis model enhanced by an augmented knowledge association matrix, termed CAG-NCD. The proposed model refines the Q-matrix to improve the expressiveness of item–knowledge correspondences and employs nonlinear interaction functions to capture relational features in students’ response processes. Specifically, convolutional neural networks are used to extract local semantic patterns from student–item interactions, while graph convolutional networks model the global structural dependencies among knowledge points. By jointly integrating semantic and structural information, the model effectively captures complex dependency relationships. Experimental results show that CAG-NCD achieves performance improvements of 3.7% on the FrcSub dataset and 4.5% on the Math dataset, significantly reducing prediction errors and enhancing the interpretability and accuracy of cognitive diagnosis across multiple datasets.
Wang et al. (Tue,) studied this question.