PHGKT: A Knowledge Tracing Model Based on Hypergraph Neural Network and Transformer

With the development of adaptive learning, Knowledge Tracing (KT) has increasingly provided significant assistance to educational platforms in tracking user's learning progress. Traditional Deep KT methods relying on sequence models such as RNNs and bipartite graph-based KT models based on question and skill often fail to capture the high-order relationships and multi-dimensional feature information among question, skill, and user, limiting their performance. To overcome these limitations, we propose PHGKT, a deep knowledge tracing model based on hypergraph neural networks and Transformer. PHGKT spatially constructs a hypergraph structure connecting question, skill, and user, employs a bidirectional attention mechanism for information aggregation, and utilizes Transformer to fuse achieve the fusion of low-dimensional embeddings of the hypergraph structure and the question features. Additionally, PHGKT introduces a historical interaction retrospective module from a temporal perspective to model the influence of user's historical sequences on their current knowledge state. Upon evaluation across public datasets, it has been observed that PHGKT surpasses the current top-performing baseline models, achieving an average enhancement of around 1.73 % in terms of AUC, validating its effectiveness in modeling complex high-order interactions and multi-dimensional feature information.