Research On Personalized K-anonymity Model And MDAV Algorithm Based On Gray Relational Analysis

With the advent of the era of big data,more frequent interaction in data,the problem of privacy disclosure in the process of data transferring is becoming increasingly prominent.Therefore,how to ensure that private information in the process of interacting and sharing can be protected effectively becomes a hot research issue,k-anonymity is an effective method of privacy preserving,its core idea is to requires that,for each tuple,there are at least k-1 tuples can not be distinguished from it after k-anonymity,making the attacker can not distinguish the private information to a specific individual.There are many algorithms for k-anonymity.Microaggregation algorithm MDAV(Maximum Distance to Average Vector)is one of the more commonly used algorithms for k-anonymous models.Its core idea is to divide the mass data into several categories fast according to the closeness of tuples,then achieve k-anonymity by the class centroid values.The algorithm act good performance in efficiency,data availability,data security and other aspects.This paper focuses on k-anonymity model and its algorithm MDAV of privacy protection technology.Realizing specific personalized privacy preserving though the two dimension of the quasi-identifier attributes and sensitive attribute values.Algorithm and model improvement research details is as follows.(1)For the algorithm MDAV of k-anonymity,focus on the importance of the quasi-identifier attributes,propose a microaggregation algorithm called GMDAV(Grey Maximum Distance to Average Vector).The algorithm use the grey relational analysis to measure the important attributes,focuses on the similarity of the important attributes,so that enhance information retention rates of the important attribute values.Experimental result shows that,the algorithm can effectively reduce information disclosure risk of the important attribute values.(2)Based on previous studies,focus on the distribution of the sensitive attribute values,propose(k,l,gp)-anonymity model.The model uses quantile and sensitive value frequency to describe the sensitive attribute values distribution of equivalence classes and the population,requires the balanced adjacent degree between the two be greater than gp(gp is a parameter).In addition,the model has the new condition that tuples add to equivalence class,which requires the balanced adjacent degree to be improved.Experimental result shows that,the model can ensure less information loss,and reduce the privacy disclosure risk at the same time.(3)To make further improvement to the(k,l,gp)-anonymity model,focus on the sensitivity of the sensitive attribute values,propose(k,l,gp,a)-anonymity model.The model puts the sensitivity into(k,l,gp)-anonymity model and combines with grey relational analysis,makes the sensitive attribute values distribution of equivalence classes and the population reflect personalized sensitivity of the sensitive attribute values.Experimental result shows that,the improved model not only achieve personalized privacy preserving oriented to sensitive attribute values,effectively guarantee the security of the high the sensitivity of sensitive values,but also have a better comprehensive privacy protection ability.