Research On Privacy Protection Methods For Social Networks

With the rapid development of mobile Internet,social network has become an important way of communication between people.More and more individuals are participating in social networking activities,forming a huge amount of personal and social relationship information.Such information contains great research value,but it also contains a lot of sensitive information.For example,the weight of the network can indicate the frequency of contact or transaction price between two commercial organizations.An attacker with background knowledge usually collects specific information and analyzes it through data mining techniques to re-identify information such as nodes and connections,resulting in the disclosure of sensitive private information,which will cause extremely serious consequences.Therefore,social network data needs to be protected before being published and shared,so that social network data with data mining and analytics value is released without revealing privacy.However,the existing privacy protection technologies such as anonymization cannot completely resist the privacy attack with strong background knowledge,and the availability of data after privacy protection is also relatively insufficient.Therefore,this paper aims to use differential privacy technology to solve the privacy leakage problem in social network publishing while improving the availability of published data.This paper expounds the research status of traditional social network privacy protection technology,and summarizes the research techniques of social network differential privacy protection.Based on the trusted third-party centralized system architecture,combined with differential privacy technology for different types of social networks,the key technologies of privacy protection published by social networks are studied.The main work of this paper is as follows:(1)Aiming at the problem of edge weight privacy leakage in the process of weighting social network publishing,combined with differential privacy technology,a weighted social network publishing algorithm DWT-DP based on community discovery and discrete wavelet transform is proposed.The algorithm first divides the social network into communities to reduce the network size,and uses discrete wavelet transform to perform multi-resolution analysis on the weight matrix of each community after partitioning.Then,the high-frequency detail matrix of each layer after the discrete wavelet transform and the low-frequency approximation matrix of the last layer are adaptively assigned a privacy budget and Laplacian noise is added,and the weight matrix is reconstructed.Finally,connect the community and post a privacy-protected social network.(2)Aiming at the leakage of social relationship privacy in dynamic social network evolution and publishing,combined with differential privacy technology,a dynamic social network publishing algorithm DP-DSNP based on differential privacy was proposed.The algorithm first performs dynamic community discovery for dynamic social networks and then uses Jaccard correlation coefficients to track the evolution of the community.For the evolving community to transform into an adjacency matrix and perform selective differential privacy perturbation,the community that has not evolved retains the perturbation of the network at the last moment.Finally,the community is dynamically connected to generate a complete social network and publish it.(3)In order to verify the effectiveness of the weighted social network publishing algorithm DWT-DP and the dynamic social network publishing algorithm DP-DSNP,theoretical analysis is performed on these two algorithms,including time complexity analysis,privacy analysis and utility analysis.The results show that both algorithms satisfy the differential privacy protection and have better execution efficiency.At the same time,the social network important indicators such as node degree distribution,weighted clustering coefficient and average clustering coefficient are used to conduct experiments on real social network datasets.Verify and compare experiments with the privHRG privacy protection algorithm.The experimental results show that the two algorithms retain good network characteristics and have high data utility while satisfying the privacy protection requirements.