Discovering Opinion Spammer Groups by Network Footprints

摘要: Online reviews are an important source for consumers to evaluate products/services on the Internet (e.g. Amazon, Yelp, etc.). However, more and fraudulent reviewers write fake mislead users. To maximize their impact share effort, many spam attacks organized as campaigns, by a group of spammers. In this paper, we propose new two-step method discover spammer groups targeted products. First, introduce NFS (Network Footprint Score), measure that quantifies likelihood products being campaign targets. Second, carefully devise GroupStrainer cluster spammers 2-hop subgraph induced top ranking Our approach has four key advantages: (i) unsupervised detection; both steps require no labeled data, (ii) adversarial robustness; quantify statistical distortions in review network, which have only partial view, avoid any side information can easily evade, (iii) sensemaking; output facilitates exploration nested hierarchy (i.e., organization) among spammers, finally (iv) scalability; complexity linear network size, moreover, operates subnetwork. We demonstrate efficiency effectiveness our synthetic real-world datasets from two different domains with millions reviewers. Moreover, interesting strategies employ through case studies detected groups.