Research On Privacy Protection Methods For Social Networks

Thanks to the rapid development of Internet and big data technology,the massive data contained in social networks can bring great value to the production and life of society,but in the process of data publishing and data mining of social networks,it may lead to the problem of privacy leakage.Therefore,how to achieve safe and effective data releasing and data mining is a hot research topic at present without disclosing the privacy information of social network.Differential Privacy,as an important and effective privacy protection method,has been applied to social network privacy protection.This thesis studies the combination of social network privacy protection and differential privacy,and does the following work:(1)This thesis introduces the concept and characteristics of social networks and the basic theory of differential privacy,analyzes the privacy leakage problems in social networks,explains the privacy protection requirements of social networks,summarizes the common privacy protection methods in social networks,expounds the data protection ability of differential privacy,and reviews relevant academic achievements in the field in recent years.(2)Aiming at the privacy leakage and low query accuracy of social network data in the process of histogram publishing,this paper proposes an Adjacent Group Bucket Dividing(AGBD)method based on the differential privacy protection model.The graph mapping method is used to perform node differential privacy processing on social networks,and for the problem of introducing excessive noise during the histogram publishing process of the mapping method The AGBD method proposed in this paper uses a greedy strategy and combines the Laplace mechanism to group adjacent buckets to reduce the impact of adding excessive noise on the quality of histogram publishing.At the same time,the ranking and order-preserving method is used to optimize the histogram release and improve the accuracy of the histogram release query.The experimental results show that the AGBD method proposed in this thesis can improve the query accuracy after the histogram is released.(3)Aiming at the privacy leakage problem faced in the training and testing of the classification model in social network data,this paper uses the differential privacy protection model and the adaptive enhanced ensemble learning strategy to propose a Differential Privacy Ensemble Learning(DPEL)method.The core idea of this method is that after the construction of individual classifier based on decision tree,combined with the noise adding mechanism,the pre allocated privacy budget is added to the combination process of individual classifiers,so as to obtain a strong classifier with privacy protection,and the ?-differential privacy proof of DPEL is carried out.The experimental results show that the DPEL method proposed in this thesis can make the model still have high classification accuracy under the premise of ensuring certain data privacy information.