Abstract—Although collaborative filtering with privacy schemes protect individual user privacy while still providing accurate recommendations, they might be subject to shilling attacks like traditional schemes without privacy. There are various studies focusing on either proposing privacy-preserving collaborative filtering schemes or developing robust recommendation algorithms against shilling attacks. However, such studies fail to address preventing shilling attacks or providing privacy, respectively. We investigate a privacy-preserving memory-based collaborative filtering scheme with respect to shilling attacks. We study how to design random and bandwagon shilling attacks against such scheme and scrutinize the effects of them on the system in terms of robustness using some real data-based experiments. We show that it is still possible to create attacks to manipulate a database containing masked data. Our empirical results demonstrate that random and bandwagon attacks designed to manipulate the privacy-preserving collaborative filtering scheme affect the system’s robustness. Thus, more attention should be given to designing shilling attacks against recommendation schemes with privacy and correspondingly developing robust algorithms and detection strategies.

Index Terms—collaborative filtering, privacy, shilling, recommendation. 

Cite:Ihsan Gunes, Alper Bilge, Cihan Kaleli, and Huseyin Polat , "Shilling Attacks against Privacy-Preserving Collaborative Filtering ," Journal of Advanced  Management Science, Vol. 1, No. 1, pp. 54-60, March 2013. doi: 10.12720/joams.1.1.54-60