Dynamic Key-Value Memory Networks based on Concept Structure for Knowledge Tracing

Abstract

Knowledge tracing (KT) is a popular research topic in adaptive personalized assisted learning. In recent years, a Deep Knowledge Tracing (DKT) model based on recurrent neural networks has emerged based on the development of big data-driven and deep learning. However, it is impossible to specify which specific concepts students are proficient in the DKT model, so a deep knowledge tracing model based on a dynamic key-value memory network (DKVMN) emerges, which uses a static key matrix to store knowledge concepts and a dynamic value matrix to store the mastery of corresponding concepts. Although the DKVMN model explicitly singles out concepts for individual processing, it does not consider the association relationship between concepts. Even though it can mine the potential association, we think it is far from enough, so we propose a DKVMN model based on concept structure (DKVMN-CS), which introduces the concept association relationship a priori knowledge through concept structure graph, acting on both the static matrix of stored concepts and the weight calculation of the value matrix. Experiments show that our proposed DKVMN-CS model has a significant improvement in performance metrics compared to mainstream deep knowledge tracking models such as DKVMN.