Online Course Recommendation Based on Multi-Agent Systems and Knowledge Graphs

Abstract

In order to address shortcomings in personalization, prerequisite dependency understanding, and explainability in current online course recommendation, proposes a model based on a multi-agent architecture and a course knowledge graph. Traditional single-model approaches struggle to jointly handle learning-intent parsing, cross-source knowledge retrieval, and constraint aware personalized ranking, which limits recommendation accuracy and stability. To improve precision, we design a multi-agent architecture in which different agents are responsible for learner profiling and intent alignment, knowledge retrieval, and recommendation generation. Meanwhile, by incorporating a course knowledge graph, the model better integrates the structured semantics of the online-education domain, capturing relationships among courses, skills, and prerequisite dependencies, thereby improving the reliability and explainability of the recommendations. Experimental results show that the proposed model outperforms baseline methods in terms of accuracy, recall, and F1.