Detecting spammers on social networks

Abstract

Social networking has become a popular way for users to meet and interact online. Users spend a significant amount of time on popular social network platforms (such as Facebook, MySpace, or Twitter), storing and sharing a wealth of personal information. This information, as well as the possibility of contacting thousands of users, also attracts the interest of cybercriminals. For example, cybercriminals might exploit the implicit trust relationships between users in order to lure victims to malicious websites. As another example, cybercriminals might find personal information valuable for identity theft or to drive targeted spam campaigns.In this paper, we analyze to which extent spam has entered social networks. More precisely, we analyze how spammers who target social networking sites operate. To collect the data about spamming activity, we created a large and diverse set of "honey-profiles" on three large social networking sites, and logged the kind of contacts and messages that they received. We then analyzed the collected data and identified anomalous behavior of users who contacted our profiles. Based on the analysis of this behavior, we developed techniques to detect spammers in social networks, and we aggregated their messages in large spam campaigns. Our results show that it is possible to automatically identify the accounts used by spammers, and our analysis was used for take-down efforts in a real-world social network. More precisely, during this study, we collaborated with Twitter and correctly detected and deleted 15,857 spam profiles.

Die Free or Live Hard? Empirical Evaluation and New Design for Fighting Evolving Twitter Spammers

“…In terms of Twitter, most existing detection work can be classified into two categories. The first category of work, such as [32,22,35,34], mainly utilizes machine learning techniques to classify legitimate accounts and spam accounts according to their collected training data and their selections of classification features. The second category of work, e.g.…”

“…[28], detects spam accounts by examining whether the URLs or web domains posted in the tweets are tagged as malicious by the public blacklists. Especially, to collect training data, both [32] and [34] utilize social honey accounts to identify Twitter spammers.…”

“…Finally, we formalize the robustness of 24 detection features that are commonly utilized in the literature as well as our proposed ones. Through our experiments, we show that our new designed features are effective to detect Twitter spammers, achieving a much higher detection rate than three state-of-the-art approaches [35,32,34] while keeping an even lower false positive rate. …”

“…Any account engaging in the abnormal activities is subject to temporary or even permanent suspension by Twitter. Meanwhile, many existing research studies, such as [25,32,22,35,34], also utilize machine learning techniques to detect Twitter spammers.…”

“…To achieve our research goals, we collect and analyze around 500,000 Twitter accounts and more than 14 million tweets using Twitter API [18], and identify around 2,000 Twitter spammers by using blacklist and honeypot techniques. Then, we describe and validate current evasion tactics by both showing some case studies and examining three existing state-of-the-art approaches [35,32,34] on our collected data set. Based on the in-depth analysis of those evasion tactics, we design ten new features including graph-based features, neighbor-based features, timing-based features, and automation-based features to detect Twitter spammers.…”