Research On Range Query Method Based On Differential Privacy

Differential privacy technology is a privacy protection technology proposed by Dwork et al.in 2006.Its original intention is to allow data analysts to understand group-related statistical information while ensuring the security of individual privacy data of users.According to whether the data collector is trustworthy,the differential privacy model can be divided into the centralized differential privacy model and the localized differential privacy model.Among them,the localized differential privacy assumes that the untrustworthy setting of the data collector is more in line with the real application scenario,and is therefore widely used.This thesis mainly analyzes the two commonly used perturbation mechanisms for localized differential privacy:k-RR mechanism and k-RAPPOR mechanism,and summarizes the shortcomings of these two mechanisms.Aiming at the problem of low availability of data collected by data collectors in the localized differential privacy model,this thesis combines the existing localized differential privacy disturbance mechanism with the personalized privacy protection model,and proposes a personalized local differential privacy model based on a hierarchical structure HPLDP.Then this thesis compares the data availability of HPLDP model and LDP model on the problem of frequency statistics through experiments.Experiments show that in the three privacy protection modes of high,medium,and low,the HPLDP model can obtain more accurate frequency statistics on both the real data set and the synthetic data set.Aiming at the problem of low availability of query results for range queries on localized differential privacy summary data,this thesis proposes a range query algorithm based on the HPLDP model,which uses the statistical results of the safe region frequency submitted by users in the HPLDP model to calculate the maximum value and the minimum value of the range query results.This thesis compares the availability of the range query algorithm based on the HPLDP model and the existing range query algorithm based on the hierarchical extraction model through experiments.Experiments show that among the three privacy protection modes of high,medium,and low,the HPLDP model has better availability of range query results on the real data set.In summary,the HPLDP model proposed in this thesis can improve the availability of frequency statistics and the range query results on the basis of existing algorithms.The related algorithm proposed in this thesis has certain guiding significance for the frequency statistics problem and range query problem in localized differential privacy.