DKVMN-RFG: A Knowledge Tracing Model Based on Relation-aware Graph Attention

Knowledge tracing technology is an effective tool to address this issue of mastery level. Dynamic Key-Value Memory Network (DKVMN) performs well, but it is insufficient in modeling the dynamics of learning and forgetting as well as the associations among knowledge points. To this end, we propose the DKVMN-RFG model, which integrates forgetting behavior modeling and relation-aware graph attention technology to enhance representational capabilities. This model employs GAT and mechanisms to process knowledge graphs to capture semantic associations. DKVMN-RFG integrates explicit and implicit relationships and demonstrates strong robustness in specific scenarios. Experiments on the ASSISTments 2009 and STATICS datasets show that DKVMN-RFG outperforms other models in the AUC metric, verifying its superior performance.