Interactive Exercise Recommendation via Evolutionary Algorithm with Reinforcement Learning

Interactive exercise recommendation aims to supply students with exercises that correspond to their evolving knowledge states, thereby boosting learning efficiency and supporting their all-round development. Conventional methods usually adopt a performance-based approach to exercise selection, which can result in instability and high computational costs. Besides, these methods frequently ignore the significance of the learning process and may suggest exercises that are redundant and overly similar. To tackle these issues, we introduce a new method called Interactive Exercise Recommendation via evolutionary algorithm (IER) within the context of deep reinforcement learning. Our method decomposes the problem into two distinct tasks: exercise selection and reward determination. In the exercise selection task, we introduce an evolutionary module to assist in exercise selection, which provide diverse exercise that can effectively enhance the model's robustness. In the reward determination task, we establish two complementary reward objectives in accordance with the rules of student learning. Our approach not only promotes sequential learning for students but also maintains their interest in the learning process effectively. Experimental results have shown that our IER outperforms various state-of-the-art recommendation solutions in different scenarios.