Research On Trust Mechanisms And Recommendation-based Trust Model For P2P Networks

Different from the traditional Client/Server model, the features of P2 P network, dynamic, highly sharing, fairness, bring people a new shared Web experience. Meanwhile, a large number of anonymous nodes accessing to the network, make the network vulnerable to the attacks of different types of malicious users, in which colluding clique become more serious treats to the network. In addition, the existing rational users cause free-riding and the tragedy of the commons in the networks. Therefore, in order to reduce the security risks above, an effective, reasonable and reliable trust evaluation model is crucial. In view of these security issues above, this paper establishes two trust models: Preference Similarity-Based Mixed Recommendation Trust Model(PSRTrust), Clustering and Incentive Mechanism-Based Mixed Recommendation Trust Model(IPSRTrust). The specific research and novel contributions are as follows:(1) Preference Similarity-Based Mixed Recommendation Trust ModelDue to the emergence of a large number of new nodes in peer-to-peer network, the trust matrix is sparse and data is insufficient. Therefore, global trusts of peers are inaccurate which are computed by trust matrix iteration and the success rates of transactions become low. This paper proposes a Similarity Random Walk(SRW) strategy to restore the default trust data to improve the success rates of transactions.The unreasonable hypothesis of the existing trust model leads to the Power-law distribution of feedbacks that the vast majority of transactions occur in a few peers that have higher global trust. This model puts forward a multi-level selection strategy to expand the limited range of options, reduce the load of partial nodes and increase the utilization of resource in the network.In order to solve the security issues of distributed data storage, this model proposes trust data management mechanism with multiple managers based on improved Chord protocol to avoid malicious nodes tampering with the trust data.To prevent the serious treats of colluding clique, this model proposes a simple clustering method based on behavior similarity of nodes to identify the colluding clique in the networks.(2) Clustering and Incentive Mechanism-Based Mixed Recommendation Trust ModelThis model puts forward an ant clustering method based on behavior similarity with multiple attributions to improve the clustering method with single attribution in PSRTrust model. This strategy improves the accuracy and stability of identifying colluding clique.To reduce the number of rational users and stabilize the network order, this model proposes an incentive mechanism based on two-layer contributions of nodes and dynamic programming.(3) Simulation ExperiencesComparing to the classic trust model, EigenTrust and PowerTrust, the simulation results show that the proposed trust models, PSRTrust and IPSRTrust, can maintain the stable network order and identify the colluding clique accurately. Moreover, the incentive mechanism in IPSRTrust model can effectively reduce the number of rational users in the networks.