Detection of Misbehaviors in Clone Identities on Online Social Networks

Abstract

The account registration steps in Online Social Networks (OSNs) are simple to facilitate users to join the OSN sites. Alongside, Personally Identifiable Information (PII) of users is readily available online. Therefore, it becomes trivial for a malicious user (attacker) to create a spoofed identity of a real user (victim), which we refer to as clone identity. While a victim can be an ordinary or a famous person, we focus our attention on clone identities of famous persons (celebrity clones). These clone identities ride on the credibility and popularity of celebrities to gain engagement and impact. In this work, we analyze celebrity clone identities and extract an exhaustive set of 40 features based on posting behavior, friend network and profile attributes. Accordingly, we characterize their behavior as benign and malicious. On detailed inspection, we find benign behaviors are either to promote the celebrity which they have cloned or seek attention, thereby helping in the popularity of celebrity. However, on the contrary, we also find malicious behaviors (misbehaviors) wherein clone celebrities indulge in spreading indecent content, issuing advisories and opinions on contentious topics. We evaluate our approach on a real social network (Twitter) by constructing a machine learning based model to automatically classify behaviors of clone identities, and achieve accuracies of 86%, 95%, 74%, 92% & 63% for five clone behaviors corresponding to promotion, indecency, attention-seeking, advisory and opinionated.