Personalized novelty-aware recommendation in social recommender systems: A Framework

Abstract

This paper introduces a novel framework for improving social recommender systems by incorporating a personalized notion of item novelty grounded in user’s social interactions. Unlike conventional approaches that treat novelty as a static, item-specific feature, our method estimates the novelty of each item relative to a given user by analyzing behavioral patterns within the user’s social communities. Additionally, we model each user’s individual tendency toward novel content, allowing for personalized calibration of the novelty–relevance trade-off in recommendations. The proposed method operates independently of the underlying recommendation algorithm and can be seamlessly integrated as a post-processing step over candidate lists generated by various base models. Experimental evaluations on two real-world datasets—Epinions, and LastFM—demonstrate that our framework consistently enhances diversity, coverage, and novelty while preserving recommendation relevance. These findings underscore the value of socially contextualized and user-personalized novelty modeling in elevating the effectiveness and user satisfaction of recommender systems.