Graph Model Based C2C E-business Fraud Detection Research

As C2C e-commerce is one of the important patterns of e-commerce, it becomesmore and more popular with consumers. Nowadays online reputation systems have beenused by most of the C2C e-commerce trading platform. User’s current feedback score iscalculated by online reputation system based on user’s trading history, and user’sfeedback score is proportional to user’s reliability. Online reputation system has solvedthe problem of trust in C2C e-commerce to some extent, but it cannot identify fraud ofC2C e-commerce. A malicious user of C2C e-commerce enhances its feedback score bysome ways disguising as an honest user, and takes fraud. Fraud has become one of therestricting factors for C2C e-commerce development.This paper divides users’ roles of C2C e-commerce into fraudest, accomplice,co-accomplice and honest. This paper also proposes a three phase fraud detectionalgorithm. In first phase, build user hybird social network graph is built. In secondphase, Bayesian Network is applied for fraud detection of C2C e-commerce. In thirdphase, Markov Random Field is used for fraud detection of C2C e-commerce.Black and white list can be accessed by artificial recognition and bulletin ofthird-party trading platform. By a certain number of black and white list users, it canidentify fraud by Bayesian Network method. First of all, select part of attributes fromuser’s feedback related attributes, price related attributes, and social related attributes asvariables in the Bayesian Networks. By the Bayesian Network learning method, theC2C e-commerce Bayesian Network is constructed. Then it is used to identify user’srole for fraud detection. At the end, the result shows combining user’s feedback relatedattributes, price related attributes, and social related attributes together for frauddetection performs better than one related attributes individually.Reach a large number of black and white lists is a difficult and complex job, but itis easy to access a little black and white list. This paper proposes combining user hybridsocial network graph and Markov Random Field together for fraud detection in the caseof little black white list. Firstly, building hybrid social network graph with user’strading network and user’s social network. Secondly, expand black and white list withuser’s similarity. Then, Markov Random Field is applied to predicting user’ role. At theend, the effectiveness and efficient of this method is proved by experiment.