Abstract: A new security management mechanism based on the classification control of trade nodes was proposed. The model gives over the concepts of local trust and global trust and takes the classification control of trade nodes. This pa- per's innovation is as follows;l) Through classification and quantification of failure events in trade between nodes,ac- cording to the severity of damage and the size of the trade, the trade failure events arc divided into malicious attacks, bad quality and so on. 2)The rough set classifier and I3ayes classifier are used and the nodes are divided into trust nodes, strange nodes and malicious nodes in a peer-to-peer network. "hhe trust node list and the malicious node list are built. hhe malicious nodes arc excluded from trading. 3) hhe rough set classifier and 13ayes classifier arc used to integrate the feedback recommendation and to decide the type of the recommended node. hhe feedback behaviors arc divided into the honest feedback, the malicious feedback and so on. The experiment indicates that compared with the existing trust mo- dcl, the model may obtain higher examination rate over malicious acts with the higher transaction success ratio, and has the better feedback information synthesizing capacity.

Key words: Security model, Peer-to-peer network, Rough set, Bayes classifier, Simulation experiment