Study On Privacy Preserving Technology For Social Network Data Publishing

Various types of published data in social networks imply plenty of valuable information.Development of data mining technology enables users to collect and analyze these data for commercial or research purpose.However,some private information is also disclosed and illegally used by malicious party,which has raised people's concern on information sharing.In order to securely share and analyze data while not revealing users'privacy,privacy preserving technology is emerging.It aims to protect data privacy and meanwhile maintain data utility for particular purpose.Existing privacy preserving methods for social networks mainly focus on the model's establishment and implementation,without taking characteristics of social network into account,such as power-law distribution and integrity.This results in damage of the overall structure when publishing these social networks' data,thus affecting the utility of data for analysis.Therefore,this thesis studies the structure characteristics of social networks,and proposes an anonymized network reconstruction algorithm based on node's betweenness centrality.This algorithm protects the node degree data in social networks' publishing and maintain the structure property to the maximal degree.In protection on weight matrix publishing in social networks this thesis proposes a differential privacy preserving method based on diffusion wavelet.The contribution is listed as follows:(1)In order to retain the core structure of the original network,this paper establishes the k-anonymity model and proposes an anonymized network re-construction algorithm based on betweenness centrality property.In this algorithm,betweenness centrality is employed to evaluate the nodes' contribution to the topology connectivity.It may reconstruct the subgraph of the core nodes with low degrees,which is a deficiency of the existing network reconstruction methods.The proposed algorithm may effectively improve the data utility and preserve data privacy.(2)In order to avoid weights attack on the weighted undirected graph of social networks,this thesis combines the differential privacy technique with diffusion wavelet,and proposes an algorithm which satisfies ?-differential privacy.It converts the weight matrix from time domain to frequency domain,and injects Laplace noise into the coefficient matrix flexibly.This algorithm avoids the data distortion caused by adding noise directly to the local link in the existing differential privacy algorithms.With the same privacy budget,the published networks protected by the proposed algorithm in this thesis have higher data utility.