Identify Online Fraudster with Extended Cellular Automata

Auction fraud and anomalies detection is proved to be a troublesome task. The fraudulent cliques can easily forge lots of unreal transactions, to improve their reputation and cheat the auction reputation system. By applying cellular automata reference to the transaction graph, which is constructed from users and their transaction feedbacks, we introduced an graphic model of cellular automata, which is named GCA, and a parallel analysis algorithm was implemented on Hadoop. Series of data preprocessing were performed beforehand, including data cleaning, selection and manual verification. By the definition of valid state transition rules, GCA functioned well in state prediction under the influence of its neighbors. Finally, several groups of experiments were carried out, and the results demonstrated that GCA is both effective and efficient, as it was able to predict the credible state of the tens of thousands of traders in less than one second.