Online auction networks often use reputation-based systems to help users assess each other's honesty and integrity. Fraudsters, however, can collude with accomplices to accumulate bogus positive feedback to manipulate the reputation systems. In this paper, we model an online auction network with fraudsters as a random network with hidden communities (fraudsters and associated accomplices), and propose a maximum likelihood framework to detect the fraudsters. We develop an iterative message passing algorithm to heuristically solve the maximum likelihood detection problem. This algorithm identifies fraudsters and accomplices in a distributed fashion and is a scalable solution. The algorithm converges in a finite number of iterations and has very high detection rates according to our simulations.