Speaker: Binghui Wang, ECpE Graduate Student
Advisor: Neil Gong
Title: Structure-based Sybil Detection in Online Social Networks
Abstract: Online social networks (OSNs) are known to be vulnerable to the so-called Sybil attack, in which an attacker maintains massive Sybils and uses them to perform various malicious activities, e.g., disrupting democratic election, influencing financial market, distributing spams and phishing attacks, and harvesting private user data, etc.. Therefore, Sybil detection in OSNs is a basic and important security research problem and has attracted increasing attention from multiple research communities including networking, security, and data mining. Among various methods, structure-based methods have demonstrated promising results, e.g., SybilRank was deployed to detect a large amount of Sybils in Tuenti, the largest OSN in Spain. However, existing structure-based methods suffer from several limitations. For instance, random walk (RW)-based methods can only leverage either labeled benign users or labeled Sybils in the training dataset, but not both; Loopy Belief Propagation (LBP)-based methods are not scalable and cannot be applied to directed graphs. In this talk, we propose two methods to address key limitations. The first method unifies RW-based and LBP-based methods into a local rule-based framework. Under the framework, we develop an undirected graph-based Sybil detection method, called SybilSCAR, that can maintain the advantages and overcome the disadvantages of existing undirected graph-based methods. The second method, called GANG, performs Sybil detection in directed OSNs. GANG is designed to capture unique characteristics of the Sybil detection problem in directed OSNs and generalizes LBP on directed graphs.