Ecsy-Recsy: Considering Sybil attack with time dynamics and economics in recommender system

There is no doubt on the popularity of online social networks (OSNs) nowadays. With the exponential growth of the OSNs, various recommender systems (RSs) are widely deployed onto the OSNs. The RSs provide users with customized information to the RSs' users. Meanwhile, nefarious users attempt to compromise the RSs at the same time. There have been a lot of researches to block or detect such the nefarious users in the last decade. However, almost all studies aim to prevent such an anomaly under assumptions that the attackers have unlimited resources with static time condition. We consider the attackers' economics and time dynamics by introducing the concept of stickiness and persistence in the RSs. Furthermore, we suggest an anomaly detection scheme named Ecsy-Recsy (Economics-based sybil detection in Recommender system) leveraging the stickiness and persistence. We evaluate the stickiness and persistence from experiment with our dataset in which we crawled from a real world RS. Our experiment shows that the stickiness and persistence are the powerful measures to reveal the condition of RS in dynamically changing time domain. The Ecsy-Recsy shows the good detection performance under the various anomalies (naive, random, and average Sybil attack).

[1]  Chao Liu,et al.  Recommender systems with social regularization , 2011, WSDM '11.

[2]  P. Massa,et al.  Trust-aware Bootstrapping of Recommender Systems , 2006 .

[3]  Jon Kleinberg,et al.  Differences in the mechanics of information diffusion across topics: idioms, political hashtags, and complex contagion on twitter , 2011, WWW.

[4]  Fengyuan Xu,et al.  SybilDefender: Defend against sybil attacks in large social networks , 2012, 2012 Proceedings IEEE INFOCOM.

[5]  Michael Kaminsky,et al.  SybilGuard: defending against sybil attacks via social networks , 2006, SIGCOMM.

[6]  Zunping Cheng,et al.  Robust Collaborative Recommendation by Least Trimmed Squares Matrix Factorization , 2010, 2010 22nd IEEE International Conference on Tools with Artificial Intelligence.

[7]  Karl Aberer,et al.  Impact of Trust Management and Information Sharing to Adversarial Cost in Ranking Systems , 2010, IFIPTM.

[8]  Thomas Hofmann,et al.  Lies and propaganda: detecting spam users in collaborative filtering , 2007, IUI '07.

[9]  Bhaskar Mehta,et al.  Attack resistant collaborative filtering , 2008, SIGIR '08.

[10]  Feng Xiao,et al.  SybilLimit: A Near-Optimal Social Network Defense Against Sybil Attacks , 2010, IEEE/ACM Trans. Netw..

[11]  Neil J. Hurley.  Robustness of recommender systems , 2011, RecSys '11.

[12]  L. Capra,et al.  Temporal Defences for Robust Recommendations , 2010 .

[13]  Ido Guy,et al.  Social recommender system tutorial , 2014, RecSys '14.

[14]  Bamshad Mobasher,et al.  Model-Based Collaborative Filtering as a Defense against Profile Injection Attacks , 2006, AAAI.

[15]  Lakshminarayanan Subramanian,et al.  Optimal Sybil-resilient node admission control , 2011, 2011 Proceedings IEEE INFOCOM.

[16]  John R. Douceur,et al.  The Sybil Attack , 2002, IPTPS.

[17]  Feng Xiao,et al.  DSybil: Optimal Sybil-Resistance for Recommendation Systems , 2009, 2009 30th IEEE Symposium on Security and Privacy.

[18]  Lakshminarayanan Subramanian,et al.  Sybil-Resilient Online Content Voting , 2009, NSDI.