西北师范大学学报（自然科学版）2025，Vol.61Issue(4)：71-81,99,12.DOI:10.16783/j.cnki.nwnuz.2025.04.009
结合学生上下文特征的解耦认知诊断模型
Disentangled cognitive diagnosis model integrating student contextual features
摘要
Abstract
In response to the problem that existing cognitive diagnosis models cannot fully explore the intricate interactions among different student contextual features and fail to analyze the impact of different latent traits on student states,a disentangled cognitive diagnosis model integrating student contextual features is proposed.Firstly,a hypergraph neural network is employed to capture complex interactions among contextual features,generating both student and background embedding representations.Secondly,contrastive learning and minimal mutual information techniques are applied to disentangle students' latent traits,to learn both intrinsic and extrinsic feature representations.Finally,the model predicts students' performance on exercises based on students' cognitive states and exercise difficulty,while estimating cognitive state parameters through a backpropagation algorithm.Experimental results on the SLP and PISA datasets show that the proposed model achieves an accuracy of 83.48％on the SLP dataset and 71.83％on the PISA dataset.Compared to the Neural Cognitive Diagnosis model(NeuralCDM),the proposed model improves the area under the curve(AUC)by 0.68 and 3.95 percentage points on the SLP and PISA datasets,respectively.关键词
认知诊断/个性化学习/超图神经网络/特征解耦/对比学习Key words
cognitive diagnosis/personalized learning/hypergraph neural network/feature disentangling/contrastive learning分类
信息技术与安全科学引用本文复制引用
梁亚丽,张志昌,杜育明,张珂,罗华正..结合学生上下文特征的解耦认知诊断模型[J].西北师范大学学报（自然科学版）,2025,61(4):71-81,99,12.基金项目
国家自然科学基金资助项目(62163033) （62163033）
