Identifying Spam Activity on Public Facebook Pages

Abstract

Since their emergence, online social networks (OSNs) keep gaining popularity. However, many related problems have also arisen, such as the use of fake accounts for malicious activities. In this paper, we focus on identifying spammers among users that are active on public Facebook pages. We are specifically interested in identifying groups of spammers sharing similar URLs. For this purpose, we built an initial dataset based on all the content that has been posted upon feed posts on a set of public Facebook pages with high numbers of subscribers. We assumed that such public pages, with hundreds of thousands of subscribers and revolving around a common attractive topic, make an ideal ground for spamming activity. Our first contribution in this paper is a reliable methodology that helps in identifying potential spammer and non-spammer accounts that are likely to be tagged as, respectively, spammers/non-spammers upon manual verification. For that aim, we used a set of features characterizing spam activity with a coring method. This methodology, combined with manual human validation, successfully allowed us to build a dataset of spammers and non-spammers. Our second contribution is the analysis of the identified spammer accounts. We found that these accounts do not display any community-like behavior as they rarely interact with each other, and are slightly more active than non-spammers during late-night hours, while slightly less active during daytime hours. Finally, our third contribution is the proposal of a clustering approach that successfully detected 16 groups of spammers in the form of clusters of spam accounts sharing similar URLs.