Two-stage Unsupervised Approach for Combating Social Spammers

Spammers use Online Social Networks (OSNs) as a popular platform for spreading malicious content and links. The nature of OSNs allows the spammers to bypass the combating techniques by changing their behaviours. Classification based approaches are the most common technique for spam detection. “Data labelling” “spam drift” “imbalanced datasets” and “data fabrication” are the most common limitations of classification techniques that hinder the accuracy of spam detection. The paper presents a two-stage fully unsupervised approach using a user’s peer acceptance within OSN to distinguish spammers from genuine users. User’s common shared interest over multiple topics and the mentioning behaviour are used to derive the peer acceptance. The contribution of the paper is a pure unsupervised method to detect spammers based on users’ peer acceptance without labelled datasets. Our unsupervised approach is able to achieve 95.9% accuracy without the need for labelling.