Research on Cognitive Diagnosis Model Based on Self Attention Mechanism and Residual Blocks

To surpass the deficiencies of conventional cognitive diagnostic models, particularly in their ability to uncover deep links between various knowledge areas and overcome the challenges presented by complex nonlinear associations, and to enhance the precision in assessing students' appreciation of knowledge, novel model, named the Self-Attention Residual Cognitive Diagnostic (SARCD) model, is introduced. It effectively utilizes the self-attention mechanism and integrates residual connection units. Adopting the self-attention approach, the SARCD model skillfully identifies the intricate and long-range dependencies existing among knowledge vectors, and then refines the feature integration procedure by employing a residual framework. The incorporation of residual units adeptly addresses the difficulties arising from gradient disappearance and explosion issues, which can arise during the course of training deep neural networks. The evaluation outcomes using the readily available Assist dataset, it is demonstrated that the SARCD model offers notable superiority over current benchmark models across various assessment indicators. The SARCD framework exhibits superior functional efficiencies, specifically by achieving improvements of 1.3% in accuracy (ACC), 1.6% in the root mean square error (RMSE) value, and a notable 3.1% increase in the area under the curve (AUC). The comprehensive experimental data robustly confirms that the SARCD model exhibits remarkable competence in mastering long dependencies among knowledge points and complex nonlinear relationships, bolstering the diagnostic performance.