结构感知与认知增强的知识追踪模型OA

Most existing knowledge tracing methods have limitations in feature modeling.They often rely only on fixed se-quences or static graph structures,which makes it difficult to characterize the dynamic relationships between students and problems over time.In addition,they frequently overlook the effects of temporal and behavioral factors on knowledge states.To address these issues,this paper proposed a structure-aware and cognition-enhanced knowledge tracing model(SACEKT),which comprehensively considered the dynamic relationships between students and problems,response time,and behavioral features,as well as learning and forgetting mechanisms,to model the dynamic evolution of students' knowledge states from both structural and cognitive perspectives.Specifically,SACEKT introduced a dynamic graph neural network to capture the temporal evolution information of student-problem nodes.Also,this model designed a multi-feature encoding module that dis-tinguished between short-and long-term time intervals and attempt behaviors.Furthermore,it proposed a cognition-aware ga-ted recurrent unit(CA-GRU)to explicitly model the influence of learning progress and forgetting effects on students' know-ledge states.Experiments on three public datasets show that SACEKT outperforms existing models in prediction accuracy and generalization capability.