A sock puppet detection algorithm on virtual spaces

On virtual spaces, some individuals use multiple usernames or copycat/forge other users (usually called ''sock puppet'') to communicate with others. Those sock puppets are fake identities through which members of Internet community praise or create the illusion of support for the product or one's work, pretending to be a different person. A fundamental problem is how to identify these sock puppets. In this paper, we propose a sock puppet detection algorithm which combines authorship-identification techniques and link analysis. Firstly, we propose an interesting social network model in which links between two IDs are built if they have similar attitudes to most topics that both of them participate in; then, the edges are pruned according a hypothesis test, which consider the impact of their writing styles; finally, the link-based community detection for pruned network is performed. Compared to traditional methods, our approach has three advantages: (1) it conforms to the practical meanings of sock puppet community; (2) it can be applied in online situation; (3) it increases the efficiency of link analysis. In the experimental work, we evaluate our method using real datasets and compared our approach with several previous methods; the results have proved above advantages.