Research On Trust Mechanism In Service-Oriented Network Environment

Service-oriented Network has become a key issue in the development of Future Internet."Everything is service" is permeating people’s thinking. But "lack of trust" is still the shackles of the development of Service-oriented Network. With the growing number of network scale, peers in the network could hardly select a trusted and reliable service provider since most the providers are strangers without prior experience. Many regard trust management models as one of the most important measures for sharing information and developing new consumer-provider relationships. However, due to the existence of malicious peers, especially strategy malicious peers, trust models are vulnerable when facing cheating, collection, spy and camouflage attacks, which fails the trust models to metric the reputation of users and to provide a valuable reference for service selecting. Considering the malicious behaviors of malicious peers, this thesis proposes related theory and addressing solutions in terms of description of trust, filtering malicious feedback and recommendations, trust propagation and service selection. The main contributions are as follows:(1) The trust entity is separated into multi-attributed and fine grained parts in trust models. We also describe trust relationship from the perspective of fuzzy set theory and introduce the concept of weighted multi-attribute and context, which makes the direct trust and recommendation trust evaluation process more accordance with the subjective feature of trust. We filter malicious recommendations and feedbacks by utilizing the pair-wise similarity of recommendation’s weight coefficient and lattice degree of nearness of each recommendation vector. At last, we improve the derivation rules of trust recommendation relationships by multiple multivariable fuzzy reasoning mode and fuzzy uncertainty reasoning theory and give an integrated trust model in service-oriented network.(2) This thesis investigates the classical EigenTrust model deeply in terms of rating mechanism, local trust computing, trust propagation mechanism and network density. We point out that essential causes of inherent vulnerabilities in EigenTrust exist in its mechanism of local trust computing and uniform trust propagation, which are less attack resilient. This thesis proposes an attack resilient local trust computing model based on multi-dimensional rating mechanism and variant of service quality. We also introduce the similarity weighted uniform trust propagation and similarity threshold based conditional trust propagation. Besides, we also do a deep analysis on the factors in trust propagation process such as decay factor, jump strategies and so on. Experiment results show that the proposed local trust computing model and trust propagation mechanism are attack resilient under sophisticated attack models. (3) Considering about the importance of trust propagation mechanism in trust models, this thesis analyzes and compares all kinds of trust propagation models in detail under the real dataset-Epinions. We also proposed linear threshold value based trust propagation model. Experiment results show that similarity as weight and threshold reference in conditional trust propagation both are important for improving the attack resilience of trust model. Based on the study on trust propagation models, we found that the linear threshold value based conditional trust propagation leads to a new service selection strategy:activation value based service selection. The experiments in Epinions network show that this selection strategy and reputation based service selection could be complementary to each other. The hybrid selection strategy can improve the attack resilience of trust model and decline time complexity.