Poster: Effort-based Detection of Comment Spammers

Social media has become ubiquitous and important for content sharing. A typical example of how users contribute content to a social media platform is through comment threads in online articles. Unfortunately, there is an increasing prevalence of malicious activity in these threads by spammers through comment messages. The existing approaches tackling comment spam are comment-level in that they attempt to classify a comment message as spam or not spam. We propose EDOCS, a graph-based userlevel approach that quantifies how much effort a user exerted over his or her comments, to detect if the user is a comment spammer or not. We conjecture that spammers can only exert limited effort in terms of time and money over preparing and disseminating their comments, hence their effort scores are expected to be lower than those of the legitimate users. Our experimental evaluation of EDOCS shows its effectiveness in detecting comment spammers accurately with 95% true positive rate at 3% false positive rate as well as preemptively.