Improving Knowledge Tracing through Embedding based on Metapath

Abstract

The goal of knowledge tracing (KT) is to track students’ knowledge status and predict their future performance based on their learning logs. Although many researches have been devoted to exploiting the input information, they do not strictly distinguish between questions and the involved skills when taking the learning logs as input, and hence leading to performance degradation due to the fact that the inherent relations between skills and questions are not fully utilized. To solve this issue, we propose an embedding pre-training method based on metapath by explicitly considering the relations between skills and questions in the domain. Specifically, we construct a heterogeneous graph composed of skills and questions, and obtain the meaningful embeddings of nodes using the metapath2vec method, hence the explicit relation information can be embedded in the dense representation of skills and questions while still maintaining their own characteristics. Adopting these pre-trained embeddings to existing models, experiments on three public real-world datasets demonstrate that our method achieves the new state-of-the-art performance, with at least 1% absolute AUC improvement.