MAKT: A Knowledge Tracing Model Based on Meta Path and Attention Mechanism

With the deep integration of artificial intelligence technology and education, the traditional educational pattern has changed hugely. And the adaptive learning based on automatically tracing the knowledge status of students at various stages has attracted much attention. As a key technology, knowledge tracing has become an important research. Although deep learning has been used in knowledge tracing and promoted certain performance improvement, it still has drawbacks. First, current researches consider less the explicit representation of meta path between users, exercise items and knowledge points, ignoring some of the higher-order information. Secondly, the effect of higher-order information of knowledge points on prediction is ignored. Therefore, we proposes a meta-path based four-way co-attention mechanism model MAKT to inversely infer the unobservable knowledge cognitive proficiency of learners. Based on meta path, the MAKT model integrates instance information and higher-order information between nodes to effectively enhance the representation of user, exercise item and knowledge points. The effectiveness of the model was demonstrated in tests on a real data set.