Promoting Recommendations: An Attack on Collaborative Filtering

The growth and popularity of Internet applications has reinforced the need for effective information filtering techniques. The collaborative filtering approach is now a popular choice and has been implemented in many on-line systems. While many researchers have proposed and compared the performance of various collaborative filtering algorithms, one important performance measure has been omitted from the research to date -that is the robustness of the algorithm. In essence, robustness measures the power of the algorithm to make good predictions in the presence of noisy data. In this paper, we argue that robustness is an important system characteristic, and that it must be considered from the point-of-view of potential attacks that could be made on a system by malicious users. We propose a definition for system robustness, and identify system characteristics that influence robustness. Several attack strategies are described in detail, and experimental results are presented for the scenarios outlined.