Application of Personalized Learning Path Recommendation Based on DQN

This paper proposes a personalized learning path recommendation method based on Deep Q Network (DQN), which aims to address the knowledge integration characteristics and dynamic teaching needs of Science curriculum in Canadian middle schools. This course integrates physics, chemistry, biology and geoscience, and the traditional static teaching path is difficult to adapt to students 'cognitive differences and interdisciplinary knowledge convergence needs. By modeling learning path planning as Markov decision process, a state space integrating knowledge mastery, cognitive features and behavioral feedback is designed, and a reward function is constructed to quantify positive and negative learning effects. The algorithm innovatively introduces experience replay mechanism and target network technology to solve the problem that traditional recommendation methods cannot capture long-term knowledge dependence. Empirical results show that, on a real dataset containing 28,500 learning sequences, the cumulative reward (425.8) and discounted cumulative reward (380.2) of the DQN model are significantly better than those of the control algorithms, and the success rate of interdisciplinary paths is increased to 91.2%. This method provides extensible intelligent decision support for comprehensive science education by dynamically optimizing knowledge unit sequence and cognitive load distribution.