Evaluating Algorithms for Detection of Compromised Social Media User Accounts

Abstract

Hijacking of user accounts on Online Social Networks (OSNs) such as Twitter is increasingly used for purposes such as large-scale spam campaigns and cyber crime-related scams and phishing attacks, but also for purposes such as spreading terrorism propaganda and as part of information operations in political and military conflicts. We present a novel method for evaluation of the performance of algorithms for detection of compromised social media accounts. In short, we create two artificially "hijacked accounts" by randomly selecting two genuine user accounts and after a number of successive posts switch the remaining posts between them. This allows us to create large amounts of training and/or test data, something which is not easy to find for this problem. Further, the developed method is utilized to evaluate the performance of a modified version of the existing COMPA system.