Knowledge Tracing Based Wheel Spinning Detection Model in Intelligent Tutoring Systems

Mingzi Chen, Yanan Chen, Shouxiang Ni, Peilin Li, Xuguang Zhang

Abstract: The recent spread of Intelligent Tutoring Systems (ITS) provides students with access to adaptive learning. However, existing ITS systems face significant challenges in student supervision, primarily in designing robust detection models to characterize the students' learning state and to prevent students from getting stuck in wheel spinning. To this end, this paper proposes a wheel spinning detection model based on Mediating Deep Knowledge Tracing (MDKT). First, the detection model applies “student question mastery” as a mediator variable to improve its interpretability. Specifically, we utilize a Classification and Regression Tree (CART) in the detection model, which enhances the model compatibility with a variety of data by incorporating explicit features (e.g., hint usage). In addition, we employ deep knowledge tracing to merge implicit time-dependent features (e.g., students' knowledge proficiency and forgetting) so as to improve the detection accuracy. Finally, an interpretable neural network based on logistic regression is applied to determine the probability of a student experiencing wheel spinning behavior. Experimental results validate the effectiveness and interpretability of our wheel spinning detection model.