Interpreting Learner Success: Enhancing Knowledge Tracing with Attention-Based IRT Models in Modern Education

In the past few decades, Computer-Assisted Education Systems (CAES) have experienced rapid development. These platforms can recommend adaptive learning materials, conduct knowledge tests, offer learning path suggestions, and create personalized study plans for learners. Knowledge tracing (KT) is a fundamental task in the field of artificial intelligence education. It involves the estimation of students' knowledge levels based on the likelihood of their correct responses to questions, allowing the prediction of their future performance. In recent years, it has gained significant attention due to its importance in subsequent tasks such as organizing learning materials. However, previous knowledge tracing models have struggled to simultaneously enhance interpretability and predictive accuracy. In this paper, we propose a knowledge tracing model based on attention mechanisms. This model establishes connections between the current problem and each past question the learner has answered, embedding a simple and interpretable Item Response Theory (IRT) model to capture individual differences between questions. Our extensive experiments, conducted on five benchmark datasets (Statics2011, ASSISTments2009, ASSISTments2015, Algebra06 and Slepemapy), demonstrate significant improvements in our model compared to previous knowledge tracing methods. Further analysis illustrates how our method improves the interpretive aspects of knowledge tracing, signifying its enhanced applicability in practical educational contexts.