An Interpretable Exercise Recommendation System Combining Semantic Enhancement and Cognitive Forgetting Modeling

In online education platforms, achieving accurate and interpretable personalized exercise recommendation based on students’ cognitive states is critical for improving learning efficiency and learning experience. To address the problems of insufficient semantic relationship modeling among knowledge concepts and the neglect of individual differences in student forgetting behavior in the existing knowledge graph recommendation model KG4Ex, this paper proposes the SF-KG4Ex (Semantic Forgetting KG4Ex) model. The proposed model employs a pretrained language model to enhance semantic relationships among knowledge concepts, introduces a Half-Life Regression framework to model students’ forgetting processes in a personalized manner, and jointly trains the related models to accurately characterize students’ cognitive states. Experimental results on three public datasets demonstrate that the proposed model achieves superior performance in terms of both accuracy and novelty. Compared with KG4Ex, the proposed model improves average accuracy by 6.9 percentage points and novelty by 2.7 percentage points, verifying its effectiveness and superiority.