A Survey of Attack-Resistant Collaborative Filtering Algorithms

With the increasing popularity of recommender systems in commercial services, the quality of recommendations has increasingly become an important to study, much like the quality of search results from search engines. While some users faithfully express their true opinion, many provide noisy or incorrect ratings which can be detrimental to the quality of the generated recommendations. The presence of noise can violate modeling assumptions and may thus result in unstable estimates or predictions. Even worse, malicious users can deliberately insert attack profiles in an attempt to bias the recommender system to their benefit. This is a particularly important issue, and it is necessary for systems to provide guarantees on the robustness of recommendations to ensure continued user trust. While previous research has attempted to study the robustness of various existing Collaborative Filtering (CF) approaches, the explicit design of robust recommender systems remains a challenging problem. Approaches such as Intelligent Neighborhood Selection, Association Rules and Robust Matrix Factorization are generally known to produce unsatisfactory results. In this paper, we review previous approaches to robust collaborative filtering; we also describe promising recent approaches that exploit a Singular Value Decomposition (SVD) and are both accurate as well as highly stable to shilling.