Local Differentially Privacy For Social Network Publishing Based On Uncertain Graph

With the growing popularity of the Internet technology,people's social style is developing gradually from traditional mode to online communication platform,day and night increases in various online social platform for new users to produce more and more social network data.Social network data besides can bring huge commercial interests,for its particular structure and data analysis of various existing patterns are widely used in public opinion analysis,group activities and so on various aspects.However,the new technology has brought great convenience to our daily life and exposed the problems of this convenience.Social network data often contain a large number of personal sensitive information,and the release of these network data will pose a serious threat to personal privacy.Privacy protection of social network data has become a hot topic in the field of privacy protection.First of all,based on local disturbance and k-anonymous method of social network data privacy protection only against certain types of attacks and data for the attacker's background knowledge need strict assumptions.This privacy protection model only works for a specific type of attacker,and the privacy protection is not strong enough.Is second,based on finite difference method of social network data privacy is mostly combined to generate graph model is reconstructed by means of sampling to generate the original network,and add a noise in the process of reconstruction,so as to realize difference data privacy.This data publishing privacy protection method is destructive to the structure information of social network.Is a social network structure of social network analysis,however,information is very valuable,for example,are widespread in the social network of community structure and so on,to further excavate subgroup user behavior,such as attribute information play an important role.Therefore,it is very necessary to protect the topology of social network when the social network privacy protection data publishing method can resist all kinds of attacks.Aiming at the social network problem with structural information,this paper proposes a privacy protection method based on local differential privacy to design privacy protection method to realize data release.The details are as follows:(1)analysis of the existing global finite difference model of social network structure bigger destructiveness information privacy issues,put forward two aspects reasons: on the one hand,the difference of global privacy protection method of the attacker's background knowledge is not accord with practical significance;On the other hand,the existing methods of network users and the connection of the individual is only expressed in simple Boolean variables,privacy in injection difference noise,excessive noise will be added to a single edge,thus affect the liquidity of the figure.(2)aimed at undermining Social Network data structure information of the two aspects of reasons,put forward the Privacy of the partial differential Social Network data Privacy protection method(Local Differentially Privacy for Social Network publishing-based on all Graph).From the two aspects,we solved the problem that the existing method had great damage to the structure information of social network: on the one hand,it set up a more realistic social network data to release the privacy protection scene;On the other hand,combining with the generation graph model,the local edge probability of the social network with the community structure is reconstructed and the Laplace noise is injected.(3)to publish the privacy protection method(LDP-USN)for the social network data of the proposed partial differential privacy,which proves that the method satisfies local differential privacy by rigorous mathematical formula.The algorithm is designed and implemented on three real data sets.The system framework of this method is constructed,and the algorithm flow of each function submodule and each module is described in detail.Finally,the time complexity of three main algorithms is analyzed.(4)through three real social network data sets(Web KB,Cora,Citation)simulation experiment,combining with the commonly used social network structural properties testing standards,this article selects the impassability privacy under the budget of aggregation coefficient(Clustering coefficient),the structure entropy(structure entropy)as well as the number of edges the structure of the three performance indicators to verify this method the degree of protection.The results show that the average aggregation coefficient on the three data sets is close to the original graph,and the aggregation coefficient is closer to the original graph as the privacy budget becomes larger.Structure entropy in privacy budget ? 0.2 or higher,with the original figure approaching;The number of edges is less than the original graph in the sparse data set Web KB,but in the other two data sets,as the privacy budget increases,the number of edges is gradually closer to the original graph.By comparing the experiment with the original social network,this paper analyzes the performance of this method in network structure protection,and verifies the effectiveness of the proposed method.To some extent,it protects the structure information of social network data,and provides effective privacy protection for the published social network data.