Machine learning meets social networking security
Guofei Gu · 2012
While normal users are enjoying the services provided by Online Social Networking websites (OSNs), cyber criminals are now using OSNs as their new weapon in their attacks (e.g., spamming, malware spreading). Although OSNs provide a new attack platform for cyber criminals, it is also a new opportunity to study this malicious community as a whole group, i.e., the malicious social network, a research topic that previous studies have not touched yet. We believe that it is interesting, meaningful, and timely to study the peculiarities of malicious social networks that are composed of malicious identities. In particular, in this talk we will focus on Twitter-like OSNs as the starting point. We will examine current evasion tactics used by malicious Twitter identities, and discuss how to design new machine learning detection features and how to evaluate their robustness in the context of OSNs. Furthermore, we will empirically study the cyber criminal ecosystem on Twitter, show several interesting observations about the malicious social networks, and discuss how we can design new inference algorithms to defend against criminal accounts using the information we have learned from them.