Research And Application Of Privacy Protection Technology Based On Correlation Statistical Model

The rapid development of network and information technology has triggered an explosive growth in the scale and variety of data.A large amount of data information has been collected by various information carriers.On the one hand,the open sharing and development of data resources can activate the hidden value behind the data and promote the scientific development of all walks of life;On the other hand,the data set usually contains a lot of personal sensitive information,and there are various correlations between sensitive information and sensitive information or non-sensitive information.The lack of effective protection measures or excessive data mining can lead to privacy disclosure.Aiming at the privacy protection in the process of data flow and analysis,this thesis studies the privacy protection technology based on Association statistical model.The main research work is as follows:(1)This thesis proposes a differential privacy data publishing algorithm via principal component analysis based on maximum information coefficient.By introducing the maximum information coefficient to improve the traditional PCA method,and based on the ability of reducing dimensions to generate the information content of the PCA data,this thesis quantifies the introduced noise.Finally,the perturbed low rank approximation matrix is restored and published based on the data characteristics.This method combines the maximum information coefficient and differential privacy budget optimization allocation,and realizes the original dimension privacy data publishing in the centralized scenario.The experimental results show that the algorithm can achieve efficient information carrying of various data relationships,and the noise introduced is lower than other algorithms.(2)This thesis proposes a sparse PCA algorithm based on local differential privacy.In the framework of distributed optimization,the main subspace is estimated by alternating direction multiplier method and sparse orthogonal iteration pursuit method.According to the characteristics of data structure and local differential privacy,this method constructs a privacy protection model in distributed scenario,which can realize the collaborative analysis of privacy data in distributed scenario.The experimental results show that the algorithm can improve the estimation accuracy on the premise of data privacy.(3)This thesis designs a privacy protection scheme for wireless medical treatment in centralized and distributed scenarios.Through the experimental comparison and analysis of the privacy data utility of similar privacy protection schemes in the same scenario,it is verified that the proposed scheme can maintain higher data utility under the same level of privacy constraints,and has more advantages than other schemes.