Integrating Knowledge Graph Reasoning and Reinforcement Learning for Explainable MOOC Recommendations

In the context of the rapid development of digital education, MOOC platforms face the dual challenges of meeting the diverse needs of learners and addressing complex course systems. Traditional recommendation models have overly focused on prediction accuracy while neglecting explainability, failing to meet users’ demand for transparent decision-making. To address this, we propose an innovative Explainable Knowledge-Aware Reinforcement Learning method (EKARL) that deeply integrates knowledge graphs (KG) and reinforcement learning (RL) technologies. By leveraging revised reward mechanisms and adaptive path generation strategies, this framework successfully breaks free from reliance on predefined patterns. The adaptive path generation strategy allows the model to freely explore paths of any type and length within the knowledge graph without relying on predefined path patterns. This not only enhances the model’s generalization capabilities but also better adapts to the complex and dynamic needs of learners in educational scenarios. In our experiments on the public dataset MOOCCube, we constructed an educational knowledge graph, employed the TransH model for semantic entity representation, and introduced pruning strategies to enhance inference efficiency. The results show that the EKARL method achieved a NDCG score of 19.81%, representing a 39.51 percentage point improvement over the optimal baseline model, demonstrating the method’s significant advantage in enhancing recommendation accuracy. Additionally, multi-hop path reasoning analyzes multi-hop paths between users and entities such as courses, such as “user-course-user-course” and “user-course-concept-course” to provide learners with clear recommendation explanations, thereby enhancing learners’ trust in the recommendation results.