A Confusion-Enhanced Deep Learning Model for Knowledge Tracing
Ming Yin, Ruihe Huang · 2024
Knowledge tracing (KT) refers to the task of tracing the changing knowledge state of learners and predicting their future performance through their learning records. It is the core of computer assisted adaptive cognitive learning. Existing KT research uses deep learning methods to explore the relationship between concepts and problems, which improves prediction performance. However, few models apply educational psychology theories. Guiding students to learn through educational psychology theories is also important in the process of learning, which is an important basis for distinguishing different students' knowledge state. To solve this problem, we propose a gated network with confusion analysis mechanism (CDKT) to predict the confusion level of learners and trace the knowledge state of learners during learning based on educational psychology theory. The proposed model can model the knowledge state of students through three gated network to adapt to the individual development of learners and individual differences after training. It can also predict students' performance in new exercise.