摘要
Althoughk-anonymityisagoodwayofpublishingmicrodataforresearchpurposes,itcannotresistseveralcommonattacks,suchasattributedisclosureandthesimilarityattack.Toresisttheseattacks,manyrefinementsofk-anonymityhavebeenproposedwitht-closenessbeingoneofthestrictestprivacymodels.Whilemostexistingt-closenessmodelsaddressthecaseinwhichtheoriginaldatahaveonlyonesinglesensitiveattribute,datawithmultiplesensitiveattributesaremorecommoninpractice.Inthispaper,wecoverthisgapwithtwoproposedalgorithmsformultiplesensitiveattributesandmakethepublisheddatasatisfyt-closeness.Basedontheobservationthatthevaluesofthesensitiveattributesinanyequivalenceclassmustbeasspreadaspossibleovertheentiredatatomakethepublisheddatasatisfyt-closeness,bothofthealgorithmsusedifferentmethodstopartitionrecordsintogroupsintermsofsensitiveattributes.Oneusesaclusteringmethod,whiletheotherleveragestheprincipalcomponentanalysis.Then,accordingtothesimilarityofquasi-identifierattributes,recordsareselectedfromdifferentgroupstoconstructanequivalenceclass,whichwillreducethelossofinformationasmuchaspossibleduringanonymization.Ourproposedalgorithmsareevaluatedusingarealdataset.Theresultsshowthattheaveragespeedofthefirstproposedalgorithmisslowerthanthatofthesecondproposedalgorithmbuttheformercanpreservemoreoriginalinformation.Inaddition,comparedwithrelatedapproaches,bothproposedalgorithmscanachievestrongerprotectionofprivacyandreduceless.
出版日期
2018年06月16日(中国期刊网平台首次上网日期,不代表论文的发表时间)