Enhanced Content-Based Filtering Algorithm Applied in Core Topic Recommendation System for First-Year Computer Science Students

Abstract

This study presented an enhanced content-based filtering algorithm for topic recommendation system tailored for the first-year computer science students at University of the City of Manila. The enhancement focused on addressing the cold-start problem, a common issue experienced by new users caused by the algorithm's dependence on user interactions for generating personalized recommendations. To overcome this, the Felder-Silverman Learning Style Model (FSLSM) was incorporated to create a more comprehensive user profile, enabling consistent personalized recommendations for new users. FSLSM considers four dimensions: (1) Processing Dimension (Active/Reflective), (2) Perception Dimension (Sensing/Intuitive), (3) Input Dimension (Visual/Verbal), and (4) Understanding Dimension (Sequential/Global). Additionally, the user's current semester was included to better align recommendations with their curriculum. The enhanced content-based filtering algorithm demonstrated significant improvements over the baseline, effectively addressing the cold-start problem. It achieved a mean precision of 91.7%, recall of 91.5%, and F1-score of 91.6%, reflecting balanced accuracy and effectiveness. The system also attained a mean average precision (MAP) of 91.7% and an average consistency of 91.7%, ensuring stable and reliable recommendations across users.