Abstract—Exercises, as an important learning resource, play an important role in testing learners’ learning achievements and judging learners’ mastery of knowledge points, and are an indispensable part of personalized learning. Exercise recommendation technology is an important means to improve learners’ learning efficiency, and has become an important research topic in the field of learning. We propose a personalized exercise recommendation model based on Deep Knowledge Tracing (DKT), named DKT-SA. The DKT-SA model utilizes self-attention mechanism to predict the scores of unfinished exercises, enhancing its ability to predict learners’ knowledge mastery status, especially when dealing with cold start problems. We use three real-world datasets: ASSIST12, EdNet, and CPSS for empirical analysis. The experimental results show that the DKT-SA model is superior to the traditional DKT model in the four key indicators of Accuracy (ACC), Area Under the Curve (AUC), Precision and Recall. The results of this paper verify the effectiveness of DKT-SA model in personalized exercise recommendation tasks, and provide a new solution for the field of smart education.
 
Keywords—personalized exercise recommendation, deep learning, deep knowledge tracing, self-attention mechanism


Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).