Massive Open Online Courses (MOOCs) have emerged as a technical challenge in the form of information overload to learners due to the explosive expansion of MOOCs on platforms like Coursera, edX, MIT OpenCourseWare, and Khan Academy. It is no longer that simple to find the correct course out of hundreds of thousands of options, and the traditional keyword-matching or rating-based recommenders always lose the ability to read between the lines in a free-form natural language query such as I want to learn AI but I am terrible at math and I am a complete beginner. In this paper, NLPRec, a full-stack intelligent course recommendation system that is developed on the basis of seven principles, is presented: (1) a powerful seven-stage NLP preprocessing pipeline including selective preservation of negation; (2) sublinear TF-IDF-vectorisation with bigram-based features using a multi-source course corpus; (3) cosine similarity retrieval with a log-dampened collective engagement boost; (4) a nine-step query comprehension engine, including; (a) abbreviation expansion; (b) spell correction with domain 10-curated query validation at K = 5 gives mean Precision @ 5 = 0.72, Recall @ 5 = 0.98 and F1 @ 5 = 0.82 - 71.4, 71.9 and 70.8 improvements over the baseline of keyword-matching. The system is released as open-source Streamlit web application that provides a reference implementation of the NLP-enhanced EdTech community that can be reproduced.
Dawkar et al. (2026) studied this question.