Research On Multi-Dimensional Range Queries Key Techniques Based On Local Differential Privacy

With the rapid development of artificial intelligence technology,information systems generate large amounts of user data.These data are highly dimensional and contain rich value.The collection and analysis of users’ data can help merchants improve the quality of products and services.However,in the process of collecting and analyzing user data,it is necessary to protect the privacy of user data.The problem of multi-dimensional range query based on local differential privacy technology has been widely concerned by researchers.For existing algorithms combined with hierarchical tree structure,hierarchical tree root nodes leak user privacy.With the increase of data dimension,query error increases rapidly and response efficiency is low.Therefore,based on local differential privacy model,a multi-dimensional range query algorithm(H4MDA)combining user grouping strategy and local perturbation mechanism is proposed.H4MDA algorithm makes full use of horizontal and vertical structure characteristics of hierarchical tree,and combines user grouping strategy to design three local perturbation mechanism.They are HGRR,LGRR-FD and LGRR.HGRR perturbation mechanism is based on the horizontal structure of hierarchical tree,dismisses the combination of hierarchical root nodes,and combines GRR mechanism to perform local perturbation for user tuple data.Different from HGRR mechanism,LGRR-FD mechanism is based on the vertical structure of hierarchical tree,combined with the GRR mechanism to perturb the local data of users,and at the same time by adding fake data to avoid the leakage of user privacy of hierarchical tree leaf nodes.The LGRR mechanism also combines the vertical structure of the hierarchical tree and the GRR mechanism.After discarding the leaf node layer,the tuple data of the user is perturbed locally.The collector uses all the user data perturbed by the LGRR mechanism to reconstruct the leaf node layer of the hierarchical tree through local consistency processing,and the count sum of virtual leaf nodes is equal to the count value of the parent node.The three perturbation algorithms make use of hierarchical tree structure to respond to multi-dimensional range queries and avoid the risk of privacy leakage.Theoretical analysis and experimental results show that HGGR,LGRR-FD and LGRR mechanisms can achieve the effect of privacy protection,and the accuracy of multi-dimensional range query response is better than that of similar algorithms.