A Hybrid Decision Approach to Detect Profile Injection Attacks in Collaborative Recommender Systems

Collaborative filtering is a vitally central technology of personalized recommendation, yet its recommended result is so sensitive to users' preferences that the recommender system has significant vulnerabilities. To overcome the addressed issue, this paper proposes a hybrid decision approach to effectively and efficiently detect profile injection attacks in collaborative recommender systems. Through modifying the algorithms of RDMA (Rating Deviation from Mean Agreement) and WDMA (Weighted Deviation form Mean Agreement), the hybrid decision approach is integrated from these modified algorithms and the UnRAP (Unsupervised Retrieval of Attack Profiles) algorithm. The extensive experiments based on three common attack models show that the proposed detection algorithm is the best comparing with the modified RDMA and WDMA or origin ones, by which the detecting accuracy significantly increases almost 35%, 25%, and 8% than the RMDA, WMDA, and UnRAP algorithms, respectively. Furthermore, for the mixed attack model, we compare it with the UnRAP algorithm and improve the 10% accuracy.

[1]  Martin Ester,et al.  TrustWalker: a random walk model for combining trust-based and item-based recommendation , 2009, KDD.

[2]  Zunping Cheng,et al.  Statistical attack detection , 2009, RecSys '09.

[3]  Padraig Cunningham,et al.  Unsupervised retrieval of attack profiles in collaborative recommender systems , 2008, RecSys '08.

[4]  Martin Ester,et al.  A matrix factorization technique with trust propagation for recommendation in social networks , 2010, RecSys '10.

[5]  Bamshad Mobasher,et al.  Classification features for attack detection in collaborative recommender systems , 2006, KDD '06.

[6]  Bamshad Mobasher,et al.  Towards Trustworthy Recommender Systems : An Analysis of Attack Models and Algorithm Robustness , 2007 .

[7]  Tomoharu Iwata,et al.  Strength of social influence in trust networks in product review sites , 2011, WSDM '11.

[8]  Wolfgang Nejdl,et al.  Preventing shilling attacks in online recommender systems , 2005, WIDM '05.

[9]  Bamshad Mobasher,et al.  Defending recommender systems: detection of profile injection attacks , 2007, Service Oriented Computing and Applications.

[10]  Li Chen,et al.  A user-centric evaluation framework for recommender systems , 2011, RecSys '11.

[11]  Bhaskar Mehta,et al.  Unsupervised strategies for shilling detection and robust collaborative filtering , 2009, User Modeling and User-Adapted Interaction.

[12]  Robin Burke,et al.  Effective Attack Models for Shilling Item-Based Collaborative Filtering Systems , 2005 .

[13]  Krishna P. Gummadi,et al.  Measuring User Influence in Twitter: The Million Follower Fallacy , 2010, ICWSM.

[14]  Neil J. Hurley,et al.  Utility-based neighbourhood formation for efficient and robust collaborative filtering , 2004, EC '04.

[15]  Zheng Chen,et al.  Finding group shilling in recommendation system , 2005, WWW '05.