Interest Group-based Trust Model For Peer-to-Peer Networks

Trust evaluation model could solve some of the security problems in the network, but when there are a very large number of nodes in the network, computation and communication costs of the existing global trust model are relatively large, which affect the promotion and application of the trust model.Based on this problem, trust mechanism in the social relations is exploited, large-scale networks are divided logically into a number of small-scale interest group which is based on the similarity of nodes, and then this thesis constructed a hierarchical unstructured overlay with small-world network characteristics.This thesis proposes a self-organized algorithm of interest group construction and recommendation, the nodes which have the similar interests are found according to the recommendation request, and then the nodes are gradually gathered into the same group using the trust based incentives and dynamic exchange mechanism. At last, the connections between the groups are established, which makes the constructed hierarchical network have the small-world network characteristics. The simulation results indicate that, this algorithm can complete the construction of interest groups in a short recommended period. It is an effective solution to reduce the computation and communication costs, and the recommendation information among the similar nodes is more credible.Because of the subjectivity and uncertainty with the trust, it is difficult to describe trust with the accurately mathematical model. In this thesis, fuzzy theory is exploited to improve the trust evaluation model. In the trust evaluation, we highlight the subjective factor of the nodes. In the recommendation of the trust, the direct interaction experience between the intermediate nodes is exploited, combined with the factor of path attenuation, a new fuzzy reasoning method is proposed. In the trust updating, we use the cosine function to simulate the history forgetting to the trust information. Analyses and simulations show that the model is exact and robust, malicious oscillating behaviors can be quickly identified. The model is scalable because of its computation and communication costs improvement in the large-scale networks.