Research on Deep Knowledge Tracing Model Integrating Knowledge Structure Graph

In the teaching process of online education, to enhance learners' learning outcomes, it is essential to provide personalized and intelligent teaching services. Therefore, how to scientifically assess learners' knowledge mastery levels based on their historical learning records and formulate personalized learning plans has become a critical issue in online education. Knowledge Tracing (KT) aims to estimate learners' knowledge proficiency and predict their future learning performance based on their historical response records. Current knowledge tracing approaches integrate learner characteristics and domain-specific features, achieving certain improvements. However, several challenges remain: 1) The structural relationships between knowledge points remain inadequately enriched when addressing domain feature integration; 2) Existing embedding representation methods for multiple relationships between knowledge points require further refinement. To address these issues, we propose a deep knowledge tracing model incorporating knowledge structure graphs (Tr-dkt). This model leverages learners' historical records and subject-specific knowledge characteristics to uncover latent relationships between knowledge points. It achieves effective integration of domain features and knowledge tracing through joint representation of knowledge point relationships and learner- exercise interaction information. Experimental studies investigate how the quantity, types, and representation dimensions of knowledge point relationships influence the prediction accuracy of the knowledge tracing model, revealing corresponding impact trends.