Maximizing Stability of Recommendation Algorithms: A Collective Inference Approach
Gediminas Adomavičius, Jingjing Zhang · 2014
Abstract. This paper focuses on stability of recommendation algorithms, which measures the consistency of recommender system predictions. Stability is a desired property of recommender systems and has important implications on users ' trust and acceptance of recommendations. Prior research has reported that some popular recommendation algorithms suffer from high degree of instability. In this study we propose a novel meta-algorithm that can be used in conjunction with different traditional recommendation techniques to improve their stability. Our experimental results on real-world movie rating data demonstrate that the proposed approach can achieve substantially higher stability as compared to the original recommendation algorithms, while, perhaps as importantly, providing additional improvements in predictive accuracy as well. 1. Introduction Recommender systems represent technologies for assisting users in finding a set of items which users are likely to find interesting or relevant [1]. Recommender systems play an important role in many different settings, including e-commerce. Examples such as Amazon and Netflix, where the provided recommendations are critical to retain users, show that a substantial portion of the product sales may result from the recommendations. Much of the research in