A Deep Knowledge Tracing Model Based on Double Attention Coding of Exercises and Concepts

A huge amount of learner interaction data is collected in the online learning system, and the learner's knowledge mastery status is hidden in the data, to find out that hidden information is the key point of smart education. Currently, knowledge tracing technology is widely used in learning systems, which can assess learners' mastery level of knowledge points by analyzing learners' interaction records with exercises. Among the knowledge tracing models based on deep learning, the SAINT model applies Transformer to knowledge tracing for the first time. However, the SAINT model does not consider “concept” as an input feature, and from the cognitive perspective, “concept” is the basis for learners to construct a knowledge system. Therefore, this paper improves the SAINT model by considering “concept” as an auxiliary feature and proves the feasibility and effectiveness of this model by comparing it with other deep tracking models on publicly available datasets, such as SAINT, DKT, DKVMN, SAKT.