Research On Dynamic Data Publishing Method Based On Differential Privacy

Differential privacy has become a standard for data publishing privacy of protection because it's strong privacy protection and rigorous mathematical proof.In recent years,many data publishing methods based on differential privacy have been proposed,but most of the methods focus on the release of static data,and there are few studies on dynamic data publishing.In addition,privacy budget as an indicator to measure the intensity of privacy protection,how to properly allocate a limited privacy budget to each time point for releasing data is essential to improve the utility of differential privacy preserving dynamic data publishing.Therefore,proposing a new dynamic data publishing algorithm that can allocate the limited privacy budget reasonably,and improve the utility of published data is the research focus of dynamic data publishing based on differential privacy.In order to solve the above problems,this dissertation proposes the following two algorithms:Firstly,a new differential privacy dynamic data publishing method is proposed: a greedy grouping algorithm based on KL(Kullback-Leibler)divergence(GGA).The algorithm uses KL divergence to calculate the difference between the two adjacent time points,and then publishes the data by comparing the KL divergence value with the threshold.At the same time,we also give the privacy analysis and proof of GGA algorithm.In addition,we also adopt three data publishing strategies in the data releasing process of the algorithm.Among these strategies,our proposed greedy group publishing strategy can effectively reduce the global sensitivity and improve the effectiveness of data publishing.Finally,the GGA algorithm is verified on the real dataset.The experimental results show that the GGA algorithm can publish data more effectively than the existing differential privacy dynamic data publishing algorithm and improve the utility of data publishing under the condition of satisfying differential privacy.Secondly,a dynamic data publishing algorithm based on reinforcement learning(DDPA)is proposed.Under the condition of satisfying the differential privacy,the algorithm combines the idea of reinforcement learning with the changing characteristics of dynamic data,and uses the Markov Decision Process to formalize the allocation of privacy budget.Finally,we get a reasonable privacy budget allocation scheme in the process of dynamic data publishing,and then use this allocation scheme to publish dynamic data.In addition,a new dynamic data publishing strategy is proposed in this dissertation,which is different from the existing data publishing method.Finally,the experiments on the real dataset show that the DDPA algorithm can reasonably allocate limited privacy budgets and effectively improve the utility of dynamic data publishing by combining the new data publishing strategy.