Local Differential Privacy Protection Technology Of Social Network Based On Hierarchical Random Graph

With the popularity of social network services,various types of social media can obtain a large amount of personal data and related information from users for data mining and analysis.But at the same time,it carries a risk of privacy leakage.Therefore,the issue of personal privacy protection has become increasingly important.At present,differential privacy as a rigorously theoretically validated and mathematically proven model of privacy protection,has attracted attention and research in many fields.However,the privacy protection of relational data in social networks is still in its infancy,and the existing differential privacy publishing technology is mainly focused on a centralized model,that is,the data collector is trusted by default.However,this assumption does not meet the reality.In order to solve the above problems,this thesis proposes the following two parts of research:(1)We propose a hierarchical random graph model that satisfies local differential privacy for privacy protection.First,in order to achieve the purpose of protecting privacy without losing its usefulness,we retain the statistical characteristics by converting the original graph into a hierarchical random graph model,combining the original evaluation index with the exponential mechanism to form a new model selection scheme,and selecting the final Model structure.The obtained model is then combined with the connection probability with Laplacian noise to obtain a noise graph.In the iterative selection process of the model,we borrowed the Monte Carlo Markov chain method to improve the efficiency and accuracy.In addition,we propose to apply predictive models to correct lost connections.We performed theoretical analysis and proof,and verified it on real social network data.(2)We propose a local differential privacy community detection algorithm based on Louvain algorithm.In the first stage of the Louvain algorithm,we introduced an exponential mechanism and combined the module gain to determine the community to which the node belongs.When we acquire a community,the edge connections within the community are closer,so we need to provide further privacy protection.We use a hierarchical random graph model to encode the structure information of the graph according to the edge probability,thereby transforming each community into a hierarchical random graph model.The generated hierarchical random graph model is then combined with the edge connection probability with Laplace noise.We validate our algorithm on two real social network datasets of different sizes,and the results show that our algorithm retains a certain utility while satisfying the differential privacy conditions.