Social Spammer Detection in Microblogging

Abstract

The availability of microblogging, like Twitter and Sina Weibo, makes it a popular platform for spammers to unfairly overpower normal users with unwanted content via social networks, known as social spamming. The rise of social spamming can significantly hinder the use of microblogging systems for effective information dissemination and sharing. Distinct features of microblogging systems present new challenges for social spammer detection. First, unlike traditional social networks, microblogging allows to establish some connections between two parties without mutual consent, which makes it easier for spammers to imitate normal users by quickly accumulating a large number of human friends. Second, microblogging messages are short, noisy, and unstructured. Traditional social spammer detection methods are not directly applicable to microblogging. In this paper, we investigate how to collectively use network and content information to perform effective social spammer detection in microblogging. In particular, we present an optimization formulation that models the social network and content information in a unified framework. Experiments on a real-world Twitter dataset demonstrate that our proposed method can effectively utilize both kinds of information for social spammer detection.