抗CFP攻击的社交网络隐私保护算法研究

发布时间：2018-10-20 17:05

【摘要】：随着互联网和大数据时代的到来,互联网给人们带来了巨大生活便利,但也使得人们的隐私保护受到很大程度上的威胁。因为相对于数据信息传播速度不那么发达的时期,在现有网络环境下收集、整合、分析和传播用户信息要容易的多,所以会导致用户信息更易于泄露。因此在互联网上如何保护个人隐私成为研究的热点问题。目前,已有很多关于社交网络隐私保护的方法和模型,其中最经典的是k-匿名社交网络隐私保护算法。它要求在k-匿名数据集里,每条具有标识的记录至少有k-1个记录与之相同。因此,k-匿名社交网络隐私保护算法在一定程度上保护了个人隐私。但是,现有的k-匿名技术在进行隐私保护时,将社交网络中的节点全部设为私有,忽略了实际网络中存在大量公有节点。这些公有节点身份是公开的,攻击者可以利用它们与私有节点之间的连接作为背景知识对私有节点进行再识别攻击,即Connection Fingerprint(CFP)攻击。原有的抗CFP抗击隐私保护算法很好地保护了公有节点的中心性,但是仍有不足之处,没有尽可能多地考虑社交网络图性质。本文在此基础上提出了一种改进的抗CFP社交网络隐私保护算法。主要工作有:第一,分析原有的抗CFP攻击的社交网络隐私保护算法。针对CFP攻击,现有的社交网络隐私保护算法在实施边替换时随机选取等价组中的私有节点,忽略了网络图中各私有节点的中心性等。第二,针对原有的抗CFP攻击隐私保护算法,即K-anony算法考虑图性质的不足,提出一种改进的抗CFP攻击隐私保护算法——N-hop-K-anony算法。其思想是:在n跳范围内,对任意私有节点v都至少有其余k-1个节点与其所连接的公共节点相同。N-hop-K-anony在进行节点边替换时,从社交网络图性质的几个评价标准出发,最终选取网络聚集系数作为其理论依据,对原有算法进行改进。改进后的算法在边替换上做出处理,并编码实现改进前后的算法。第三,在email-Eu-core、College Msg、Facebook和ca-Gr Qc四个真实有效的数据集上进行改进前后算法的对比实验。通过对比实验可以发现:在时间性能基本一致的情况下,算法改进后在一定程度上比改进前更能够保护节点中心性,尤其是紧密中心性和介数中心性;在网络聚集系数上,算法改进后也比改进前具有较好的实验效果。
[Abstract]:With the advent of the Internet and big data era, the Internet has brought great convenience to people, but also make people's privacy protection is threatened to a great extent. It is much easier to collect, integrate, analyze and disseminate user information in the existing network environment than in the period when the speed of data dissemination is not so developed. Therefore, how to protect personal privacy on the Internet has become a hot issue. At present, there are many methods and models of social network privacy protection, among which the most classical is the k-anonymous social network privacy protection algorithm. It requires at least K-1 records to be identical to each identified record in a k- anonymous dataset. Therefore, k-anonymous social network privacy protection algorithm to a certain extent to protect personal privacy. However, the existing k- anonymity technology in privacy protection, all the nodes in the social network are set private, ignoring the existence of a large number of public nodes in the actual network. The identity of these public nodes is public and the attacker can use the connection between them and the private node as the background knowledge to re-identify the private node attack, that is, the Connection Fingerprint (CFP) attack. The original anti-CFP privacy protection algorithm protects the centrality of public nodes well, but there are still some shortcomings, and the nature of social network graph is not considered as much as possible. In this paper, an improved privacy protection algorithm against CFP social networks is proposed. The main work is as follows: first, the original privacy protection algorithm against CFP attack is analyzed. For CFP attacks, the existing privacy protection algorithms of social networks randomly select the private nodes in the equivalent group when implementing edge substitution, ignoring the centrality of each private node in the network diagram. Secondly, an improved privacy protection algorithm (N-hop-K-anony) against CFP attacks is proposed, which is an improved privacy protection algorithm against CFP attacks, that is, K-anony algorithm takes into account the shortcomings of graph properties. The idea is: in the n-hop range, at least the remaining k-1 nodes for any private node v are the same as the public nodes connected there.When N-hop-K-anony performs node side substitution, it starts from several evaluation criteria of the nature of social network graph. Finally, the network aggregation coefficient is selected as the theoretical basis to improve the original algorithm. The improved algorithm deals with edge substitution and encodes the improved algorithm. Thirdly, the contrast experiment of the improved algorithm is carried out on the four real and effective data sets of email-Eu-core,College Msg,Facebook and ca-Gr Qc. Through comparison experiments, we can find that the improved algorithm can protect node centrality to some extent, especially tight centrality and medium centrality, in the case of basically consistent time performance, and in the network aggregation coefficient, the improved algorithm can protect node centrality to a certain extent, especially the close-centrality and intermediate-centrality. The improved algorithm also has better experimental results than before.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP309

【参考文献】