Cognitive-driven adaptive learning: a dynamic path optimization framework based on knowledge graph and reinforcement learning

Addressing the challenge of static and non-dynamic learning paths in existing adaptive learning systems, this paper proposes a cognitive-driven adaptive learning framework. This framework organically integrates Knowledge Graph (KG) and Reinforcement Learning (RL) to construct and dynamically optimize personalized learning paths for each learner. Specifically, the framework first utilizes a Knowledge Graph to represent domain knowledge in a structured manner. It then accurately captures the learner's cognitive state through a state representation that merges the learner's global knowledge mastery with their current learning focus. Finally, guided by a multi-objective reward function that balances knowledge gain, exploration, and learning cost, a Deep Q-Network (DQN) algorithm makes real-time decisions on the optimal subsequent learning content. Simulation experiments conducted on a large-scale, modern public dataset validate the framework's effectiveness. Compared to several baseline algorithms, our framework achieves significant improvements in both learning efficiency and knowledge coverage, demonstrating excellent generalizability and robustness across learner groups with different initial knowledge levels.