A Multi–Model Machine Learning Framework for Personalised Course Recommendation with Dynamic Feedback

Purpose: To develop a personalised course recommendation system that automatically suggests relevant courses based on users’ specific queries and interests, thereby improving learning outcomes and user satisfaction.

Design/Methodology/Approach: The system leverages a Learning-to-Rank (LTR) framework using XGBoost to model the relationship between user queries and course content. The raw dataset comprised a total of 2.3 million records, consisting of 1,000,000 records from Coursera, 900,000 from Udemy, and 400,000 from edX. These records span 12 major subject categories (e.g., Computer Science, Business, Data Science, Health Sciences) and three difficulty levels (Beginner, Intermediate, Advanced). At the Learning-to-Rank stage, a pairwise ranking strategy was implemented using XGBoost. Training instances were constructed as course pairs (i,j) where course i was preferred over j based on past user behaviour.

Findings: The performance is strong, with a Normalised Discounted Cumulative Gain (NDCG) score of 0.83 and high Mean Reciprocal Rank (MRR) values. The system achieves low query processing latency, ensuring real-time responsiveness. Post-retraining user feedback shows a 90% improvement in user satisfaction, validating the adaptive nature of the approach.

Research Limitation: The dataset is platform-dependent, and the framework’s generalizability to smaller or less structured educational repositories requires further validation.

Practical Implication: The framework can be applied across diverse educational platforms to provide tailored learning pathways, enhancing engagement and improving course discovery efficiency for learners.

Social Implication: By personalising education and reducing information overload, the system supports lifelong learning, democratizes access to relevant skills, and enhances the overall quality of digital education.

Originality/Value: This work introduces an adaptive, feedback-driven recommendation framework that combines LTR, dynamic feature refinement, and real-time responsiveness, offering a novel and scalable solution for personalised education.