Assessing Impacts of a Power User Attack on a Matrix Factorization Collaborative Recommender System

Collaborative Filtering (CF) Recommender Systems (RSs) help users deal with the information overload they face when browsing, searching, or shopping for products and services. Power users are those individuals that are able to exert substantial influence over the recommendations made to other users, and RS operators encourage the existence of power user communities and leverage them to help fellow users make informed purchase decisions, especially on new items. Attacks on RSs occur when malicious users attempt to bias recommendations by introducing fake reviews or ratings; these attacks remain a key problem area for system operators. Thus, the influence wielded by power users can be used for both positive (addressing the "new item" problem) or negative (attack) purposes. Our research is investigating the impact on RS predictions and top-N recommendation lists when attackers emulate power users to provide biased ratings for new items. Previously we showed that power user attacks are effective against user-based CF RSs and that item-based CF RSs are robust to this type of attack. This paper presents the next stage in our investigation: (1) an evaluation of heuristic approaches to power user selection, and (2) evaluation of power user attacks in the context of matrix-factorization (SVD) based recommenders. Results show that social measures of influence such as degree centrality are more effective for selection of power users, and that matrix-factorization approaches are susceptible to power user attacks.