Colluder Detection Based on Hypergraph Decomposition

In this paper, a new model for reputation collusion detection is established based on hyper graph theory. Users of a e-commerce system may have some kind of relationship according to the corresponding application. Such kind of connected users can be viewed as vertices jointed by hyper-edges, and thus formed a hyper graph. Colluders are those clusters in the hyper graph that all of their vertices are closely connected via hyper edges. Thus the task of detecting colluders from common users is converted to be a problem of finding those tightly connected clusters, which can be found by splitting the hyper graph according to modularity defined in this paper. Experiment shows that such modularity attribute of colluder groups are generally of large values while are of little value for common user groups, which demonstrates the effectiveness of our proposed model and algorithm.