Research On Personalized Pre-trust Based Trust Management

As a consequence of the great growth of service-oriented P2P networks in the past few years, reputation management becomes more and more essential. Due to the anonymity, dynamic and scalability in these networks, normal users face a serious threat from malicious ones. A good reputation management system can provide peers in the network with reliable information about other peers, enable them to evaluate each other accurately, aid establish trusted interactions, and finally reduce the numbers of bad services them receive.There are a lot of trust management models proposed up to date, such as EigenTrust, PeerTrust, GossipTrust, PowerTrust, etc. Nevertheless, trust management still faces some fundamental challenges, for example, effectively identify those malicious peers in the network, especially those strategic malicious peers; combat with malicious collision attacks, especially those sophisticated ones; guarantee the service requesters interact with trusted services providers. Focus on these questions, we propose a personalized pre-trust based trust management model, PersonalTrust. PersonalTrust utilizes personalized pre-trust peers and the transitivity of pre-trust among these peers. We assume that except for those pre-trusted peers selected by a system, every peer can select his own personalized pre-trusted peers based on his own interactions with other peers in the network. Utilizing the transitivity of pre-trust, we can obtain the pre-trust matrix which contains the information to ensure that the trust of peers will only propagate within the circle of trusted friends. In addition, we incorporate a mechanism into our PersonalTrust to automatically update and optimize the pre-trust matrix. Therefore, with more accurate information about other peers, a peer can interact with those peers who are more reputable, which greatly reduces bad services the peer receives.The results of our simulations and experiments reveal that our proposal performs better than EigenTrust in all scenarios, better than EigenTrust++ and Pos&Neg EigenTrust in most scenarios, and outperforms EigenTrust when threat models consist of not only pure malicious peers but also disguised peers or spies.