Application of Exercise Recommendation Model Based on Deep Reinforcement Learning

Abstract

In order to consolidate the learning achievements of students at a certain stage, Teachers often provide students with corresponding exercises before or after the beginning of this stage. Therefore, making the exercises appropriately challenging is one of the main goals of adaptive online learning systems. However, because each student's learning status is different, a student's learning situation in the same period of time will also be different. Therefore, it is also very challenging for students to choose appropriate exercises. In this paper, we propose a new method for problem recommendation. Firstly, we use GCKT model to model the user's answer sequence. Obtain the students' mastery of each concept. The learned user and problem representation are integrated into an extended framework to predict the likelihood of user mastery of the problem. Then use this as the basis for recommending exercises. On this basis, the deep reinforcement learning technology is used and the knowledge tracking model is used as a student simulator. The difference in the performance of the student simulator on all the exercises before and after solving the exercises provided by the exercise recommendation model is used as a reward, Make the model learn what kind of exercises can improve students' ability to the greatest extent, and recommend such exercises to students. Finally, the experiment is carried out in the actual use environment. The results show that the model has better performance than the current common recommendation models.