Knowledge Tracing (KT) is a fundamental task in intelligent education that tracks students' knowledge states through their historical interactions. Early deep learning-based KT mainly relies on sequential models, which often find it more challenging to explicitly model the complex many-to-many relationships between students and knowledge entities, such as exercises and knowledge components. Graph-based Knowledge Tracing (GbKT) further leverages graph neural networks to explicitly represent these structural relationships, creating new opportunities for learning-state monitoring and personalized intervention. In this survey, we conduct an in-depth exploration of recent progress in GbKT research. First, we introduce a two-dimensional taxonomy that aligns the primary modeling motivations with graph structural forms, and we characterize them with unified message-passing equations. Second, focusing on applications, such as the discovery of knowledge structures, modeling with educational psychology theories and interpretability, and personalized recommendation and path planning, we show how graph semantics endow the models with new interpretability and intervention capabilities. Third, we systematically evaluate 17 public datasets, offering researchers guidance on dataset selection. Finally, we outline promising future research directions for GbKT.
Xu et al. (Sun,) studied this question.