Fraud detection in social networks.

People share opinions, exchange information, and create content on large, distributed, and often unprotected platforms. Reputation mechanisms are used to assess trustworthiness and are key in shaping interactions between individuals. Because of the large number of users, social networks are ideal targets for abusive activity trying to exploit these mechanisms. Fraudsters are generally characterized by their interaction with a small set of other fraudsters, which mutually boost their reputation to raise the perception of being trusted. Having gained credibility, fraudsters eventually engage in fraudulent transactions to the detriment of regular members of the network. This research proposes the development of a framework to identify the collaborative nature of transactions involving deception and to detect fraud based on spectral analysis techniques. In particular, our approach uses the spectra of the adjacency matrices of social networks to analyze topological deviations in both degree distribution and clustering characteristics that point to possible evidence of fraud from the available data.