A recommender model based on strong and weak social Ties: A Long-tail distribution perspective

Abstract

A collaborative filtering recommender model for E-commerce users (CF-S&WLT) is proposed that incorporates an improved PageRank algorithm, strong and weak social ties and long-tail distribution items. The main design feature of the CF-S&WLT model is the introduction of incentive coefficients for long-tail items based on triadic closure regulation. It was compared to three alternative models in terms of accuracy (precision and F-measure), diversity, and novelty of returned Top-N items. Results reveal that CF-S&WLT improved diversity and novelty by 20.1% and 3.8% respectively, but reduced precision by 3.7% based on a first dataset (DS1), and enhanced diversity (14.2%) and novelty (18.3%) while reducing precision cost by 4.6% using dataset DS2. Overall, the proposed CF-S&WLT model performs better than other models with respect to diversity and novelty while maintaining acceptable levels of accuracy.