Social spammer detection: A multi-relational embedding approach

Since the relation is the main data shape of social networks, social spammer detection desperately needs a relation-dependent but content-independent framework. Some recent detection method transforms the social relations into a set of topological features, such as degree, k-core, etc. However, the multiple heterogeneous relations and the direction within each relation have not been fully explored for identifying social spammers. In this paper, we make an attempt to adopt the Multi-Relational Embedding (MRE) approach for learning latent features of the social network. The MRE model is able to fuse multiple kinds of different relations and also learn two latent vectors for each relation indicating both sending role and receiving role of every user, respectively. Experimental results on a real-world multi-relational social network demonstrate the latent features extracted by our MRE model can improve the detection performance remarkably. Copyright © 2018 Springer International Publishing AG, part of Springer Nature.