Research On Differentially Private Network Data Release Based On Hierarchical Structure

Data collection and deep analysis techniques for social network now has caused great concern in the information industry. The rapid development of data mining technology is a double-edged sword and the privacy leak which it lead to has caused a serious threat to the privacy of citizens. These problems become the limitations of application and management of the big data. The concept of privacy has become major practical problem in big data ecosystems.Network structure is an important feature of social networks. The network hierarchy reflects the regional characteristics of individual behavior and the important relationship between the network groups. In this paper, I research on privacy issues of social networks and regard the perspective that network hierarchy as a starting point, introducing hierarchical random graph model to describe the characteristics of social network data, and use differential privacy mathematical model to protect the privacy of social network. In the analysis of social network, transformed a social network into hierarchical random graph firstly, and then using the Markov chain Monte Carlo method to combine with the exponential mechanism to obtain a group of hierarchical random graph sample, and for each tree binding Laplace mechanism to make noise processing. Finally, to meet the differential privacy, HRGs which are added noise will eventually tranformed into a lower triangular matrix, figure out the mean value of these matrix and then based on the matrix elements that is the internal node connecting probability value of hierarchical random graph to generated purified social network data. The proposed method applied the differential privacy technology to social network privacy protection, cover interlinkages between nodes, so as to achieve the effect of privacy protection; At the same time, through the hierarchical random graph that the abstract approximately conversion, greatly reduces the amount of data processing without destroying the structure characteristics of the network, and can still be used on data analysis and mining after the publication. Compared to previous research, this method greatly improved the availability of differential privacy, and the utility of the released version. Finally, given the sensitivity analysis and privacy algorithm analysis, combined with the real network data to make the analysis of the effectiveness of the algorithm and data availability to verify that this method has a good level of privacy protection.