Collaborative Filtering Recommender System for Online Learning Resources with Integrated Dynamic Time Weighting and Trust Value Calculation

Abstract

Traditional educational models struggle to meet the demands of students seeking personalized online learning resources (OLRs). Collaborative filtering (CF) algorithms are widely employed for personalized OLR recommendations, yet they encounter issues such as poor scalability, cold start, and sparse data issues. In response, an enhanced CF algorithm is proposed, incorporating a fusion of time weighting and a credibility selection strategy. Initially, interactions and ratings among learners are analyzed. Subsequently, the algorithm integrates learner similarity and trust, calculating the credibility value weight between learners. Dynamic time weighting is then introduced separately into CF algorithms based on OLRs and learners, respectively. Ultimately, the algorithm predicts learner ratings for unknown OLRs. Experimental comparisons demonstrate that the performance metrics of the hybrid algorithm presented in this paper show significant improvement over traditional and other improved algorithms. It exhibits enhanced rating prediction accuracy, facilitating precise recommendations of personalized OLRs to learners.