A Research On Non-interactive Differentially Private Anonymity Algorithm Based On Consistent Constraint

The rapidly developing data applied technology takes full advantage of the potential value of data,but it's not so safe and reliable for traditional privacy preserving technology after the emerging technology of data mining and attack means on privacy.The data owner needs to protect private information while publishing datasets on the one hand,on the other hand,the released data is under threats of various data mining applications and attacks on privacy.At this moment,it is a difficult problem on how to design a algorithm with strong guarantee for privacy.For the above conditions,differential privacy solves the problem fundamentally.It doesn't care about the specific applied background for released data.Even in the worst case that a attacker has acquired all of the records except for one,and the attacker couldn't infer the private information of that one.However,the distortion-based operation of adding noise in differential privacy affects the availability of released data,which reduces classification accuracy in the subsequent application.This is making new demands on the design of differentially private algorithm.Aimed at range-count queries towards to classification applications in differential privacy,we design and realize a non-interactive differentially private anonymity algorithm DiffCon,which is based on a fundamental algorithm and a least square method with consistent constraint.DiffCon effectively enhances the availability of released data,on the premise of privacy protection.The main work of this dissertation includes: 1)Background introduction.Based on the relationship among privacy protection,anonymity algorithm and classification application in data mining,we introduce the project background and research status.2)Problem statement.In the traditional differentially private algorithm towards to frequency matrix,the independent noise-adding way and the rough inquiry-response pattern cause linear superposition of noise by an equal amount.This problem reduces the usability of released data in higher query dimension.3)Solution description.Based on fundamental algorithm and least square method with consistent constraint,we design a new inquiry-response pattern and redefine sensibility to realize DiffCon algorithm.4)Experimental evaluation.We validate the algorithm performance and then testing the classification accuracy by a real dataset and a classifier.The experiment results show that DiffCon significantly improves the classification accuracy and the query-dimension expansibility for released data.