RNR: A Generic Bayesian-based Framework for Enhancing Top-N Recommender Systems

Rong-Qin Xu, Ming-Yang Zhou, Hao Liao

Abstract: In personalized top-N recommender systems, a core task is to design effective methods to measure user-item preference scores and then to suggest, for each user, a small set of personalized items with high scores. However, little attention was paid to the recommendation of low-score user-item links. In this work, based on the Bayesian estimation theory, we propose a novel metric RNR (Recall-to-Noise Ratio) to characterize the ability to recommend both high-score and low-score user-item links. Then we propose a generic framework that leverages RNR to transfer the link scores of the state-of-the-art recommendation methods. Empirical experiments show that the proposed framework could optimally improve the recommendation accuracy.