Visible to the public TWC: Medium: Collaborative: Online Social Network Fraud and Attack Research and IdentificationConflict Detection Enabled

Project Details

Lead PI

Performance Period

Jul 01, 2016 - Jun 30, 2020

Institution(s)

University of Arkansas

Award Number


Online social networks (OSNs) face various forms of fraud and attacks, such as spam, denial of service, Sybil attacks, and viral marketing. In order to build trustworthy and secure OSNs, it has become critical to develop techniques to analyze and detect OSN fraud and attacks. Existing OSN security approaches usually target a specific type of OSN fraud or attack and often fall short of detecting more complex attacks such as collusive attacks that involve many fraudulent OSN accounts, or dynamic attacks that encompass multiple attack phases over time. This research, dubbed oSAFARI (Online SociAl network Fraud and Attack Research and Identification), models, analyzes and characterizes OSN frauds and attacks; designs, develops, and evaluates a new approach to detecting static OSN frauds and attacks; and further enhances the approach to handle dynamic attacks with multiple phases. The research team plans to develop a new course focused on OSN attacks and defenses, which has the potential to be offered across many institutions. To increase public security awareness, the team also plans to develop tutorial courses on typical OSN attacks and their defense and offer them at popular public events and in freshman classes. The research team will broadly disseminate their results, tools, software, and documents to the research community, IT industries, and to OSN companies.

This project embraces a systematic, comprehensive study of OSN frauds and attacks. It models OSN threats by viewing an OSN as a graph embedded with attacker nodes and edges, identifies and analyzes specific forms of frauds and attacks, and evaluates state-of-the-art attack analysis and defense approaches. It develops a spectral-analysis-based framework for OSN fraud and attack detection. The framework transforms topological information of an OSN graph into patterns formed by spectral coordinates in the spectral space, and introduces the use of the spectral graph perturbation theory to more easily model and capture changes of spectral coordinates for attacker, victim, and regular nodes. Further, this research develops spectral-analysis-based detection approaches for complex networks where nodes can carry attributes and edges can be negative, weighted, or asymmetric. Through a novel combination of the network dynamics and the vector autoregressive model, it develops an automatic spectral-analysis-based approach to detecting dynamic attacks while avoiding the high cost and low accuracy of traditional approaches. It also transforms attack characteristics from high-dimensional spectral spaces into distinctive visual patterns, and develops interactive mechanisms for analysts to incorporate domain knowledge and flexibly handle attacks. The research team will build a simulation framework to evaluate the detection approaches against different types of OSN attacks, where one can plug in different OSN datasets to evaluate and compare different detection approaches. Moreover, the research team will build a prototype oSAFARI on top of an OSN, and evaluate how oSAFARI can withstand various attacks in a real setting.