Biblio

The rising popularity of social network services, such as Twitter, has attracted many spammers and created a large number of fake accounts, overwhelming legitimate users with advertising, malware and unwanted and disruptive information. This not only inconveniences the users' social activities but causes financial loss and privacy issues. Identifying social spammers is challenging because spammers continually change their strategies to fool existing anti-spamming systems. Thus, many researchers have tried to propose new classification systems using various types of features extracted from the content and user's information. However, no comprehensive comparative study has been done to compare the effectiveness and the efficiency of the existing systems. At this stage, it is hard to know what the best anti spamming system is and why. This paper proposes a unified evaluation workbench that allows researchers to access various user and content-based features, implement new features, and evaluate and compare the performance of their systems against existing systems. Through our analysis, we can identify the most effective and efficient social spammer detection features and help develop a faster and more accurate classifier model that has higher true positives and lower false positives.