Research On Privacy Analysis Methods Of Graph Data In Social Networks

With the popularity of social software such as Facebook,QQ and WeChat,it brings great convenience to people's life.In general,the network formed by social software is often referred to as a social network,which is a way to establish connections between people.Meanwhile,when people make friends by using social software which will record a lot of personal information.Therefore,while many people enjoy this convenience,some personal privacy information will be leaked.In recent years,researchers have proposed many privacy preserving methods,such as the methods based on encryption,anonymity,data distortion and so on.At present,researchers often use data distortion which mainly implements privacy preserving by adding noise,but this method cannot fully preserve privacy of social network,there is also a lot of privacy leakage problems,for example,thousands of network fraudulent phone calls.Therefore,in addition to designing better privacy protection methods,how to illustrate the effectiveness of these privacy preserving methods is also a hot topic of current research.Generally,social networks are abstracted into graphs,for graph data,researchers at home and abroad have proposed different privacy preserving methods in the social network scenario,but there are also some researchers considering the effect of privacy preserving from the opposite direction.Owing to the privacy analysis methods of social network is very similar to privacy attack methods,thus,this paper mainly uses the privacy attack methods to study the privacy analysis methods for graph data in social network.On the one hand,adding some background knowledge,we use Bayesian inference to solve how to recognize an individual in social network.On the other hand,without considering background knowledge,we refer to denoising method of image processing based on wiener filtering which can automatically suppress noise to solve how to analyze the whole social network.The primary research work is as follows.(1)Summarizes the process of graph privacy analysis in social networks,and concludes the existing methods of graph privacy attack and privacy preserving in social networks.In addition,the common methods of graph privacy attack are summarized.And,the privacy attack method and privacy analysis method are compared in detail.(2)A graph privacy analysis method based on Bayesian reasoning to recognize single-node is proposed.Aiming at identifying a single node in social network,a privacy analysis model based on Bayesian reasoning is designed.Under this model,a Bayesian-based graph privacy analysis algorithm PABR(Privacy Analysis Based on Bayes Reasoning)is proposed.After data validation,it is found that the designed Bayesian method can identify the node with some probability.Eventually,the algorithm is compared with methods which have the same background knowledge.It is found that the recognition rate of PABR algorithm is higher than other methods,but more attention is paid to the entire social network in real life.This algorithm only studies a single node and does not analyze the entire social network.(3)A graph privacy analysis method based on Wiener filtering is proposed.In order to analyze the whole social network,we referring to the method of denoising in image processing,we choose wiener filtering which can automatically suppress noise,and propose a graph privacy analysis model based on filtering.Under this model,a graph privacy analysis algorithm GPAF(Graph Privacy Analysis on Filtering)based on filtering is designed.In the end,we use nodes,edges,average degree(AD),average clustering coefficient(ACC),betweenness centrality(BC)and degree distribution to analyze statistical data.The experimental results show that the metrics is similar with original graph data,which shows that the method can remove a part of noise in noise graph,and also provides theoretical guidance for the research of privacy preserving.