面向服务网络环境中信任机制的研究

发布时间：2018-02-13 08:12

本文关键词： 信任管理信任传播网络安全模糊理论　出处：《大连理工大学》2014年博士论文　论文类型：学位论文

【摘要】：面向服务的网络已经成为下一代互联网发展的重中之重,“一切皆服务”的思想也已经深入人心,而信任的缺乏仍然是影响面向服务网络发展的桎梏之一。在网络规模逐渐增大的情况下,由于绝大多数的服务提供者都是缺乏先验信息的陌生人,导致网络中节点在选择服务对象时无法确定该对象的可信性和可靠性。信任管理被认为是帮助服务消费者和提供者之间建立信任关系和共享信息的重要措施之一。但是,由于恶意节点的存在,尤其是策略型恶意节点的存在,使得信任模型很容易受到合谋、欺骗、间谍和伪装等行为的攻击,导致信任模型无法正确地衡量节点的信誉值,从而不能给服务请求节点一个有价值的服务选择参考。针对这些恶意行为,本文从信任的描述、恶意推荐和反馈的过滤、信任传播和服务选择等方面出发,提出了相关理论和解决方案,其主要贡献如下： (1)在信任模型中,将信任主体划分为多属性的细粒度主体,通过模糊集合理论对信任关系进行描述,引入属性权系数以及上下文的概念,使直接信任以及推荐信任的评估更符合信任的特性和准确。通过对各推荐权系数相似度和各推荐路径格贴近度的计算,有效地屏蔽了恶意推荐。最后通过多重多维模糊推理和模糊可能推理理论解决了推荐信任向量的传递和合成,给出了一个面向服务网络环境下比较完整的信任评估模型。 (2)从评价机制、局部信任值计算、信任传播以及信任网络的密度等四个方面,对经典的EigenTrust模型进行了分析和研究,并指出：EigenTrust设计缺陷的本质原因在于其缺乏可抗攻击性的局部信任值计算和统一的信任传播机制。本文提出了基于多维评价机制和服务质量方差的局部信任值计算模型,并在信任传播过程中引入了基于相似度权重统一信任传播模型和相似度阂值的条件信任传播模型。此外,本文还着重分析了衰减因子、跳转策略等对信任传播机制的影响。实验表明：本文所提出的局部信任值计算模型和信任传播机制都能在较为复杂的攻击模型下展现出良好的可抗攻击性。 (3)鉴于信任传播机制对于信任模型性能影响的重要性,本文以Epinions网络中采集到的数据为实验数据集,对多种信任传播机制进行了统一的分析对比和验证,并提出了新的基于线性阈值的条件信任传播机制。实验表明：在条件信任传播机制中同时引入相似度权重和相似度闽值是提高信任模型可抗攻击性的重要举措。在对信任传播机制的研究基础上,我们发现基于线性阈值的条件信任传播产生了一种新的服务选择策略-基于激活值的服务选择策略。实际网络数据集下实验表明：该策略和基于信誉值的服务选择策略可以相辅相成,很好地提高信任模型的可抗攻击性和降低时间复杂度。
[Abstract]:Service-oriented networks have become the top priority of the next generation of Internet development, and the idea of "all services" has been deeply rooted in the hearts of the people. However, the lack of trust is still one of the shackles that affect the development of service-oriented networks. With the increasing of network size, most service providers are strangers who lack prior information. Trust management is considered to be one of the important measures to help build trust relationship and share information between service consumers and providers. Because of the existence of malicious nodes, especially the existence of policy malicious nodes, the trust model is vulnerable to collusion, deception, espionage and camouflage, resulting in the trust model can not correctly measure the credibility of the nodes. In view of these malicious behaviors, this paper starts from the description of trust, the filtering of malicious recommendation and feedback, trust propagation and service selection, etc. The related theories and solutions are put forward, and their main contributions are as follows:. In the trust model, the trust agent is divided into fine-grained agents with multiple attributes, the trust relationship is described by fuzzy set theory, and the concept of attribute weight coefficient and context is introduced. The evaluation of direct trust and recommendation trust is more consistent with the characteristics and accuracy of trust. Finally, the transfer and composition of recommendation trust vector are solved by multi-multi-dimensional fuzzy reasoning and fuzzy possible reasoning theory, and a more complete trust evaluation model under service-oriented network environment is presented. 2) the classical EigenTrust model is analyzed and studied from four aspects: evaluation mechanism, local trust value calculation, trust propagation and the density of trust network. It is also pointed out that the essential reason for the defect of the design of the: EigenTrust is that it lacks the local trust value calculation and the unified trust propagation mechanism that can resist aggression. In this paper, a local trust value calculation model based on multidimensional evaluation mechanism and quality of service variance is proposed. In addition, a conditional trust propagation model based on similarity weighted unified trust propagation model and similarity threshold is introduced in the process of trust propagation. In addition, the attenuation factor is also analyzed in this paper. Experiments show that both the local trust value calculation model and the trust propagation mechanism presented in this paper can show good anti-aggression under the more complex attack model. 3) in view of the importance of trust propagation mechanism to the performance of trust model, this paper takes the data collected in Epinions network as the experimental data set, and makes a unified analysis and comparison and verification of various trust propagation mechanisms. A new conditional trust propagation mechanism based on linear threshold is proposed. The experimental results show that the introduction of similarity weight and similarity threshold in conditional trust propagation mechanism is an important measure to improve the robustness of trust model. Based on the research of trust communication mechanism, We find that conditional trust propagation based on linear threshold produces a new service selection strategy, which is based on the activation value. The experiments in the actual network data set show that the strategy and the service selection based on the credit value are presented. Alternative strategies can complement each other, It can improve the anti-aggression and reduce the time complexity of the trust model.
【学位授予单位】：大连理工大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP393.08

【参考文献】