Neural Network Student Cognitive Diagnosis Model Based on Multi-Relational Graph

The Internet has enabled online learning to become an important means for students to acquire knowledge, and student cognitive diagnosis models have played a key role in this process. These models can evaluate students' learning behaviors and knowledge mastery, providing support for personalized learning. Unfortunately, traditional models have their limitations when dealing with complex relational data related mainly to the relationship between students and problems, while the characteristics as well as potential relationships among students are almost left neglected. In response, this paper proposes a Neural Network Student Cognitive Diagnosis Model Based on Multi-Relational Graph (MGNCD) that brings together the powers of Graph Attention Network (GAT) and Relational Graph Convolutional Network (RGCN) to model multi-dimensional heterogeneous nodes such as students, classes, problems, and skills. The introduction of various edge types and random edge connection strategies allowed the model to tackle data sparsity and heterogeneity. Experimental results indicate that the proposed model outperforms traditional methods in accuracy (ACC) and area under the curve (AUC), which provides an encouraging novel insight into cognitive diagnosis.