Enhancing Knowledge Tracing Efficacy with Expert-defined Graphs: A Case Study in Introductory Physics Classes
Zhenting Yan, Rui Zhang · 2024
Knowledge Tracing (KT) is essential in online education for tracking student progress and forecasting future performance. Despite the effectiveness of existing KT models, enhancing their educational interpretability and reliability remains crucial for both academic and practical applications. This study introduces an improved Graph-based Knowledge Tracing (GKT) model, enriched with domain expertise, instructional insights, and contextual features, to overcome current limitations. Our enhanced GKT model employs an expert-defined graph structure for more accurate domain knowledge representation. It integrates critical contextual features, like question difficulty and prompt usage, into the response matrix for a comprehensive context. Additionally, the model leverages second-order neighborhood features to more effectively capture complex interrelations between knowledge concepts. Validation using an Introductory Physics assignment dataset demonstrated that our updated GKT model surpasses its predecessor in both Area Under the Curve (AUC) and accuracy (ACC) metrics. These improvements are instrumental in refining knowledge graphs and developing personalized teaching strategies, thereby facilitating more effective and personalized educational experiences.