Research Of Privacy Protection Technology Based On Non-interactive Data Release

In the background of big data booming, the occasions and the opportunities that people interact with massive data become more and more. While the process may be related to high-privacy data such as population, medical or financial and so on, t will cause a lot of data leakage of sensitive information if releasing such linear statistical data directly. So, it is very important that how to protect personal privacy information contained in the linear statistical results of the data releasing.By the linear statistical result of data releasing, the attackers with rich background knowledge can infer the target individual sensitive information with a high probability, thus this inference attacks of high degree of confidence become one of the important reasons for current data privacy leakage of linear statistical. Privacy differential privacy has been proposed as a new model of privacy protection because traditional anonymous models can not defend against such attacks, and makes up for lack of anonymous models fundamentally. For the implementation of differential privacy in line statistical data releasing, the main works in the paper as follows:Firstly, this paper conducts a theoretical study on differential privacy with a detailed analysis of the model to achieve privacy protection principles by data distortion, gives a comparative analysis based on the advanced technology of privacy protection, and sums up the advantages and disadvantages of the model scope.Secondly, this paper briefly describes technologies related to implementation process of differential model, such as chain attach model、Laplace Mechanism and data releasing technology.In addition, this paper gives a detailed analysis of Laplace noise distribution parameters for the protective effect of privacy protection. In view of single-dimensional data release, privacy protection is mainly a combination of Structure First model and isotonic regression algorithm,while introduce the dynamic programming algorithm to the histogram reconstruction. Additionally,the released histogram achieves minimum perturbation error by re-allocating the privacy budget εin process of histogram reconstruction. In view of multi-dimensional data release, privacy protection is to implement differential privacy in strategy matrix method and receive the disturbance result in vector form on the foundation of converting database batch queries to algebra form based on matrix-vector pattern and extending Laplace mechanism for the single query to that for batch queries.Then, the research about non-interactive privacy data release is divided into two parts according to dimension of the data releasing, proposing two solutions of privacy data release and giving a detailed demonstration. The research results are as follows.Finally, this paper conducts the experiments by different data sets about differential privacy data releasing, and verifies the availability and effectiveness of research methods via the analysis of specific data release error. Difference privacy in this research is implemented on scene for single and multi-dimensional data, which improves effectively the accuracy and ensuring the releasing data privacy at the same time.