Abstract

Fake user accounts are a growing problem today for numerous online social networks. For years, researchers have relied on community detection algorithms to propose algorithms and systems that detect these Sybil accounts. However, recent measurement work showed that attackers intentionally avoid forming communities, calling the efficacy of these systems into question.

In this talk, I will present results of two projects focused on using novel methods to detect fake Sybil accounts without relying on social graph structures. First, I will talk about our work exploring the use of crowdsourcing as a core component in a scalable Sybil detection system. We carry out a large user study analyzing the ability of crowdsourcing workers to quickly and cheaply detect fake account profiles, using a large corpus of ground-truth Sybil accounts from the Facebook and Renren networks. We analyze detection accuracy by both “experts” and “turkers” under a variety of conditions, and find that while turkers vary significantly in their effectiveness, experts consistently produce near-optimal results. We use these results to drive the design of a multi-tier crowdsourcing Sybil detection system. Using our user study data, we show that this system is scalable, and can be highly effective either as a standalone system or as a complementary technique to current tools. Second, I will present early results of a new study on Sybil detection using models of user clickstream events. We show that legitimate and Sybil users differ dramatically in user-generated events, and propose a unsupervised learning system for effectively identifying Sybil users based on user actions.