Knowledge Tracing with Exercise-Enhanced Key-Value Memory Networks

Knowledge tracing is a critical component of an intelligent tutoring system. It can track students’ knowledge states and skill mastery in order to provide a more helpful learning environment for them (for example, personalized exercise recommendations). People have been trying to apply the deep learning framework in recent years to tackle the challenge of tracing knowledge. Although these deep learning models produce impressive outcomes, they do have significant drawbacks. In existing prediction models, each exercise is often represented by a basic sequence number code, and the semantic information contained in the exercise text description has not been fully mined. We introduce a new profound education method, Knowledge Tracing with Exercise-Enhanced Key-Value Memory Networks (EKVMN) in this study, that fully utilizes the text information of questions as well as the memory function of the existing deep knowledge tracing model to predict students’ performance. Experiments on real-world knowledge tracing datasets reveal that our proposed model outperforms the baselines in terms of prediction performance. It also demonstrates the significance of context semantic information in the knowledge tracing task.