Dynamic Multi-skill Knowledge Tracing for Intelligent Educational System

Abstract

Knowledge tracing (KT) is the task of tracing students’ evolving knowledge proficiency in learning interactions. In KT research, the modeling of exercise-student relations always plays a key role. How to construct the exercise and student representation is still a pending problem. To solve this problem, we propose a novel Dynamic Multi-skill Knowledge Tracing (DMKT) method in this paper. First, the Res-embedding layer is exploited to make the exercise representation more complete. Then, a new approach is proposed for simulating students’ learning process. Furthermore, a Learning Absorption Indicator (LAI) is designed to effectively model the student's knowledge mastery. To verify the performance of our method, we implement DMKT with several baselines on three real-world datasets. Experimental results demonstrate the superiority and effectiveness of the proposed method.