P2P网络信任机制及其信任推荐模型研究

发布时间：2018-05-20 05:28

  本文选题：对等网络 + 可信度　； 参考：《南京航空航天大学》2016年硕士论文

【摘要】：区别于传统的客户端/服务器(Client/Server)模式,P2P(Peer-to-Peer)网络坚持“人人为我,我为人人”的原则,具有动态性、高度共享性、公平性等特点,带给人们全新的网络共享体验。与此同时,大量匿名节点的随意进出,使网络容易受到不同类型恶意用户的攻击,其中共谋团体对网络秩序的破坏更为严重,还需承受理性用户自私行为的影响。因此,为了减少以上安全风险对网络的危害,建立一个有效、合理和可靠的信任模型是及其重要的。本文建立了基于节点偏好相似度的混合推荐信任模型和基于聚类与激励机制的混合推荐信任模型,具体研究工作与创新如下:(1)提出了一个基于节点偏好相似度的混合推荐信任模型(PSRTrust)P2P网络中大量新进节点的加入造成可信矩阵变得稀疏,导致依据可信矩阵迭代计算的节点的全局可信度不够准确,使得交易成功率较低,针对此问题,提出相似随机游走策略(Similarity Random Walk,SRW)预测缺省的可信数据,提高交易成功率;已存在信任模型中的不合理假设造成了网络中绝大多数交易发生在极少数具有较高可信度节点上,针对此问题,提出了多层次选择策略扩大节点局限的选择范围,且在减少节点负载的同时增加了网络资源利用率;针对网络中可信数据分布式存储的安全问题,提出了基于改进Chord协议的多信任管理者的可信数据管理机制,避免恶意节点随意篡改可信数据,并给出可信数据存储计算的分布式算法;为防治共谋团体对网络的严重危害,提出基于节点行为相似的聚类方法识别网络中参与共谋团体的恶意节点,该方法较为简单易行。(2)提出了一个基于聚类与激励机制的混合推荐信任模型(IPSRTrust)改进了PSRTrust模型中基于节点单属性行为相似的简单聚类方法,提出了基于节点多属性行为相似的蚁群聚类方法以提高共谋团体识别的准确率和稳定性;运用基于节点双层贡献度和动态规划的激励机制以减少网络中理性用户的数量,稳定网络秩序。仿真实验表明所提出的PSRTrust、IPSRTrust两种模型的性能与经典的信任模型EigenTrust、PowerTrust相比较,在可信数据稀疏情况下可使网络保持较稳定秩序,且对于共谋团体的遏制效果尤为突出,并且IPSRTrust中提出的激励机制可有效减少网络中自私理性用户的数量。
[Abstract]:Different from the traditional client / Server (client / Server) model, the P2P Peer-to-Peer network adheres to the principle of "everyone for me, I for everyone", which has the characteristics of dynamic, high sharing and fairness, and brings people a new network sharing experience. At the same time, the random access of a large number of anonymous nodes makes the network vulnerable to attack by different types of malicious users, among which the collusion group has more serious damage to the network order and has to bear the influence of rational user selfishness. Therefore, it is very important to establish an effective, reasonable and reliable trust model in order to reduce the harm of the above security risks to the network. In this paper, a hybrid recommendation trust model based on node preference similarity and a hybrid recommendation trust model based on clustering and incentive mechanism are established. Specific research work and innovation are as follows: (1) A hybrid recommendation trust model based on node preference similarity is proposed. The addition of a large number of new nodes in PSRTrustN P2P network causes the trust matrix to become sparse. As a result, the global credibility of nodes calculated by trust matrix iteration is not accurate enough, and the transaction success rate is low. In view of this problem, a similar random walk strategy is proposed to predict the default trusted data to improve the transaction success rate. The unreasonable assumption in the existing trust model causes most of the transactions in the network to occur on a very small number of highly reliable nodes. In order to solve this problem, a multi-level selection strategy is proposed to expand the selection range of node limitation. In order to solve the security problem of distributed storage of trusted data in the network, a new trusted data management mechanism based on improved Chord protocol for multi-trust managers is proposed, which can reduce the load of nodes and increase the utilization of network resources. To prevent malicious nodes from tampering with trusted data at will, and to provide a distributed algorithm for computing trusted data storage, in order to prevent the serious harm of collusion groups to the network, A clustering method based on similarity of node behavior is proposed to identify malicious nodes participating in collusion groups in the network. A hybrid recommendation trust model based on clustering and incentive mechanism is proposed, which improves the simple clustering method based on the similarity of node single attribute behavior in PSRTrust model. An ant colony clustering method based on node multi-attribute behavior similarity is proposed to improve the accuracy and stability of collusion group identification, and to reduce the number of rational users in the network by using the incentive mechanism based on node two-level contribution and dynamic programming. Stable network order. The simulation results show that compared with the classical trust model EigenTrusti PowerTrust, the proposed two models can keep the network stable and orderly under the condition of sparse trusted data, and the containment effect of the two models is especially prominent for the collusion group, and the simulation results show that the proposed two models can keep the network stable and orderly under the condition of sparse trusted data. And the incentive mechanism proposed in IPSRTrust can effectively reduce the number of selfish and rational users in the network.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3
，

本文编号：1913364