Diversified Concept Attention Method for Knowledge Tracing

Through the learner's historical learning to study the knowledge state over time, knowledge tracing can predict the learner's future learning performance. It is an important issue in personalized tutoring. Although there are many related works on knowledge tracing, the existing methods still have some problems. For example, the tracing exercises are limited to a single historical exercise, the concept to be tested is limited to a single knowledge, and the semantic information on multiple concepts is rarely explored. To address this issue, a Diversified Concept Attention model for Knowledge Tracing (DCAKT) is proposed in this paper. Our method makes up for the shortcoming of the single concept of tracing, which applies natural language processing technology to get multiple concepts from the attention layer of emotional consciousness over time. Then we use an attention mechanism that evaluates the correlation between the exercises to be tested and historical concepts. This article uses the real datasets of the ASSISTments intelligent tutoring platform and a college engineering course on statics to assess the performance of DCAKT. In addition, our method can independently learn meaningful exercise sequences containing correct concepts.