Explainable Course Recommendations for Students Using Knowledge Graphs and Generative AI

Abstract There are more self-paced courses than ever before. For students, course recommendations can become a daunting task. Most recommendation systems rank courses by similarity or popularity, but who knows why? Students are left choosing according to faith and not with an understanding. This paper proposes a recommendation system consisting of a model including a student’s academic background, skills, and goals. The model eases the opaqueness of recommendation systems and provides explanations of course recommendations. The model consists of a multi-relational knowledge graph and a generative AI platform. The knowledge graph consists of students, courses, skills, and feedback. The generative AI platform retrieves relationships pertaining to the system’s recommendations and explains them to the student. The system reduces query times as catalogues grow using an efficient similarity-search index. The model was evaluated against five recommendation system benchmarks using NPTEL, Coursera, and Udemy. Over 1,200 student profiles and 200 courses were anonymized for testing. Our model significantly increased accuracy, coverage, and overall explainability compared to the benchmarks. We consider our testing to be a proof of concept and do not assume our results reveal performance as a whole. We believe the results show increased confidence in students, as these recommendations are no longer based on similarity but are grounded in actual educational philosophy.