Research Of The Key Techniques Based On Bayesian Inference In Trust Prediction In P2P Networks

P2P networks (Peer-to-Peer Networks) has become an important and new network organization form in Internet. Compared with C/S (Client/Server) network, there is no concept of server and client in P2P network and the nodes are peer, which could be server or client. Due to the characteristics of anonymity, openness and loose coupling in P2P network, it is a serious threat to the security of P2P network. In order to provide a secure and reliable network environment, trust could be used to measure the credible degree of network nodes. This paper is focus on how to compute the trust of a node to the unknown to provide a basis for decision-making.This paper first proposed trust inference algorithm based on trust evidence chain, which established a trust relation network according to the nodes and conditional probability table, set filter conditions and got the trust computing network. Based on the characteristic of the network structure, the structure is simplified, level and confidence factor are proposed to calculate trust inference values. This method is on the basis of the traditional trust inference methods and made node information more refined to compute trust prediction more accurately and objectively.This paper emphasis on a new type of trust inference method, a trust inference model based on Bayesian inference—BITrust (Bayesian Inference Trust Model). Junction Tree is the basis of Bayesian inference and the way of message transmit is used to compute trust inference. The triangulation is a critical step in junction tree and this paper proposed a triangulation based on TSP (Travel Salesman Problem) to identify the only sequence of the nodes eliminated and to establish the sole junction tree. Junction tree method based on LAZY-ARVE updates the trust potential of node and computes the node marginal probability according to the way of message transmit, which reduces the number of calculation in the process of operation and complete the trust prediction. Moreover, this method for evidence node newly joined updates the CPT information to compute the dynamic trust prediction better.Different from traditional trust inference models that focus on the external factors in the process of trust inference, BITrust emphasis on the optimization of the inference process itself and it is a pure trust prediction algorithm. This algorithm settled the issue of trust inference in P2P network to a certain extent, but there are other improvements. Node secure storage, identification of the malicious nodes, large datasets and sparse node information are still to continue research.