Personalized Course Recommendation on MOOC Platforms: A Survey

Abstract

Massive Open Online Courses (MOOCs) delivered through platforms such as Coursera, edX, Udacity, FutureLearn, XuetangX and SWAYAM have placed tens of thousands of courses within reach of millions of learners. That abundance has produced a side effect: acute information overload, difficult course selection and persistently high dropout. Recommendation systems have become the principal mechanism for guiding learners toward suitable courses, concepts and learning paths. This paper presents a structured review of the literature on course recommendation for MOOC platforms. A taxonomy is developed that organises prior work into collaborative filtering (memory- and model-based, including matrix factorization and Bayesian personalized ranking), content-based and knowledge-based methods, sequential and session-based models, knowledge-graph-based approaches, deep-learning architectures (neural collaborative filtering, recurrent and attention networks), reinforcement-learning formulations and hybrid designs, together with concept and learning-path recommendation. The benchmark datasets and platforms that underpin reproducible research, including MOOCCube, MOOCCubeX and XuetangX, are summarised. The evaluation metrics used across the field are then compared: ranking measures such as precision, recall, F1@K, NDCG, MAP, hit-rate and MRR alongside learning-centric indicators. The review synthesises the open challenges of cold-start, data sparsity, dropout-aware recommendation, scalability, explainability, fairness and diversity, pedagogical alignment with the learner knowledge state, and privacy. Finally, emerging directions are outlined. These span large-language-model and conversational recommenders, knowledge-tracing-aware systems, federated and privacy-preserving recommendation, and fairness. The survey is intended as a consolidated reference for researchers and practitioners building learner-centred recommendation systems in online education.