A Hybrid Course Recommendation Framework Integrating Association Rule Mining and Semantic Content Analysis

Abstract

The increasing complexity of higher education curricula presents significant challenges for students in selecting optimal course sequences that align with their academic strengths and career objectives. While traditional recommender systems show promise in educational contexts, they often suffer from limitations including poor interpretability, limited personalization, and cold start problems. This paper introduces a novel hybrid course recommendation approach that synergistically combines semantic content analysis, association rule mining, and explanation generation powered by large language models. The methodology is assessed using a real-world dataset obtained from the Faculty of Computer Science at October University for Modern Sciences and Arts (MSA), which includes 12,847 course enrolment records from 2,143 individuals. The data was carefully annotated to encompass student identifiers, course codes, final grades (designated as A-F, P, W), semester sequences, and prerequisite frameworks. The framework employs a multi-phase architecture that mines performance-based association rules from historical academic data, constructs semantic student profiles using transformer based course content summarization, and generates recommendations through a weighted hybrid scoring mechanism. Experimental evaluation on a comprehensive dataset of 12,847 course enrolment records demonstrates that our approach achieves a Precision@5 of 0.78 and MAP of 0.69, representing significant improvements over baseline methods. Furthermore, the integration of Mistral 7B Instruct for natural language justification generation enhances transparency and educational value. This research advances educational recommender systems by providing a comprehensive solution that balances statistical accuracy with pedagogical appropriateness, ultimately supporting improved academic decision making and student success.