面向社区检测的局部随机化匿名方法的研究

发布时间：2018-07-09 17:02

本文选题：社会网络 + 隐私安全　；参考：《广西师范大学》2017年硕士论文

【摘要】：近年来,随着社会的快速发展,尤其是社交平台的兴起,越来越多的用户加入到线上社交平台,使得社会网络数据量迅猛增加,不但方便了个人及团体更好的进行交流,也方便相关科研机构对社会网络进行更细致的研究分析。但是,有些数据却不能直接提供给相关科研人员使用,因为数据中可能包含一些敏感信息,比如姓名、用户关系、工资等。如果直接对原始数据发布,有可能泄露用户隐私。因此我们在发布数据前必须对相应敏感数据进行隐私保护处理,但是这些隐私保护策略可能会很大程度上修改原始社会网络中的结构信息。因此,如何在数据的隐私保护和数据的效用性取得平衡是社会网络的一个热点问题。目前的社会网络数据发布方法主要是给定一个原始社会网络图,进行相应的隐私匿名保护后,直接发布。但是忽略了如下问题:(一)仅仅考虑隐私安全,没有考虑到社会网络图的信息改变量;(二)原始社会网络图中可能包含各个子社区,没有对相应的子社区的隐私安全做详细的考虑。以上两点使数据的实用性降低。发布的社会网络图越精细对于相关的社会网络分析者越有利,本文通过社区检测算法对划分后的子社区结构(结点的度)做相应的隐私保护分析,对面向社区检测的社会网络隐私保护做相应的研究。本文的主要工作如下:首先,通过分析当前的社会网络隐私保护方法,发现其不足。本文使用社会网络结构(结点的度)作为攻击者背景知识,传统的k度匿名方法以及随机化方法在隐私保护的过程中没有充分考虑原始社会网络的结构图,以及原始社会网络图中存在多个子社区,一些边的连接关系可以分为子社区内部的连接,子社区之间的连接。在进行隐私保护的过程中,可能会破坏原始社会网络的结构,比如一些社会网络结点的添加删除或者边的添加删除。在k度匿名以及随机化的过程中,可能会产生很多的不确定图,破坏了数据的实用性。其次,对于上述两种传统方法存在的不足,本文提出了一种新的保护子社区结构信息的局部随机化扰动方法。在该方法中:首先基于社区检测过程中,记录边介数。当社区检测完成之后,根据攻击者的背景知识(结点的度),判断是否有隐私泄露,如果没有,说明该子社区不存在隐私泄露问题,不做处理;如果有隐私泄露,判断隐私泄露结点所连接的边是否都在子社区内部,如果是社区内部的边,属于社区内的隐私泄露,则等概率删除或者添加边,运用随机化进行扰动;如果隐私泄露的结点所连接的边与另一个社区相连,说明该结点存在边介数,调整边介数被删除的概率,使其被删除的概率增大,然后在子社区间进行随机化处理操作。通过该方法,很大程度上保证了社会网络图的原始面貌,对于任意的子社区,在保证隐私要求的前提下,该子社区社会网络结构也得到了更好的保证,方便相关科研人员对发布的社会网络图进行相关的研究分析。最后,我们使用真实的数据集来验证本文所提方法的可行性和效用性,用此方法在保证隐私安全的同时可以更好的保证社会网络的结构特征。
[Abstract]:In recent years, with the rapid development of society, especially the rise of social platform, more and more users have joined the online social platform, making the social network data increasing rapidly. It not only facilitates the better communication between individuals and groups, but also facilitates the more careful research and analysis of the social network. It can not be directly provided to relevant researchers, because the data may contain sensitive information, such as name, user relationship, salary, etc. if the data are issued directly to the original data, it is possible to disclose the privacy of the user. So we must protect the corresponding sensitive data before publishing the data, but these privacy guarantees. The protection strategy may modify the structure information in the original social network to a great extent. Therefore, how to balance the data privacy and the utility of the data is a hot issue in the social network. The current social network data publication method is to give a original social network map and carry out the corresponding privacy protection. After that, direct release. But the following questions are ignored: (I) only consideration of privacy security, no consideration of the amount of information change in social network maps; (two) the original social network map may contain all subcommunities, and no detailed consideration of the privacy security of the corresponding sub community. The above two points make the data practical. The more detailed the network graph is, the more favorable for the relevant social network analysts, this paper makes the corresponding privacy protection analysis by community detection algorithm to the sub community structure (node degree), and makes the corresponding research on the social network privacy protection against community detection. The main work of this paper is as follows: first, through the analysis of the current society. This paper uses the social network structure (node degree) as the attacker background knowledge. The traditional k degree anonymous method and the randomization method do not fully consider the structure of the original social network in the process of privacy protection, and there are a number of sub communities in the original social network map, and some sides. Connections can be divided into sub community connections and connections between sub communities. In the process of privacy protection, the structure of the original social network may be destroyed, such as the addition and deletion of some social network nodes or the addition and deletion of the edges. In the process of anonymity and randomization of K degrees, there may be a lot of uncertain graphs. Secondly, for the shortcomings of the two traditional methods, a new local randomization disturbance method for protecting the structure information of the sub community is proposed. In this method, the number of sides is recorded in the process of community detection. When the community detection is completed, the background knowledge of the attacker (node) is based on the background knowledge of the attacker If there is no privacy disclosure, if there is privacy disclosure, whether the edge of the privacy disclosure node is within the sub community, if it is the side of the community, the privacy disclosure within the community, then the probability delete or add the edge, use the following. If the nodes connected by the privacy leaked nodes are connected to another community, it shows that the node exists the boundary number, adjusts the probability of the deletion of the boundary, increases the probability of the deletion, and then carries out randomization processing between the subcommunities. By this method, the original surface of the social network map is guaranteed to a large extent. For any sub community, the social network structure of the sub community has also been better guaranteed under the premise of ensuring privacy, and it is convenient for relevant researchers to carry out relevant research and Analysis on the published social network. Finally, we use a real data set to verify the feasibility and utility of the proposed method. While ensuring privacy, the law can better guarantee the structural characteristics of social networks.
【学位授予单位】：广西师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP309

【相似文献】