Large Language Models for Course Recommendation on Mooc Platforms: A Review

Learners on a massive open online course (MOOC) platform choose from thousands of courses, and recommender systems have long been used to narrow that choice. Since 2022 a new ingredient has arrived: large language models, which can read course descriptions, rank candidates from a prompt, hold a conversation about a learner's goals and write a reason for each suggestion. This review asks what those models change for course recommendation and how much of the claimed benefit has been shown on educational data. We sketch the earlier MOOC recommenders and their datasets only briefly, then take four roles of a language model in turn: text encoder, ranker, conversational recommender and explanation writer. Most of the evidence for these roles comes from films, books and retail products. Studies on course data exist, but they are few, recent and almost all offline. We then look at how such systems are evaluated, where offline scores mislead, and the risks that weigh most in education: invented courses or prerequisites, popularity and provider bias, unequal treatment of learner groups, privacy and cost. This is a review of published work. No new experiment is reported.