Abstract_Adaptive_Filtering_for_Efficient_Record_Linkage

《Abstract_Adaptive_Filtering_for_Efficient_Record_Linkage》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、AdaptiveFilteringforEfficientRecordLinkageLifangGuRohanBaxterCSIROICTCentreCSIROICTCentreGPOBox664GPOBox664CanberraACT2601,AustraliaCanberraACT2601,AustraliaLifang.Gu@csiro.auRohan.Baxter@csiro.auAbstractBlocking/RecordComparisonComparisonTheprocessofi

2、dentifyingrecordpairsthatrepresentSearchingPairsVectorsthesamereal-worldentityinmultipledatabases,com-Recordsmonlyknownasrecordlinkage,isoneoftheimportantRecordDecisioninitialstepsinmanydataminingapplications.RecordStandardisationPairsModellinkageofm

3、illionsofrecordsisacomputationallyex-MatchingRecordsStatuspensivetask.VariousblockingmethodshavebeenusedinrecordlinkagesystemstoreducethenumberofrecordDatapairsforcomparison.AgoodblockingkeyiscriticaltothesuccessofaMeasurementblockingmethodandwillide

4、allyresultinlotofsmallblocks.However,inpractice,therearealmostalwaysFigure1:Informationflowdiagramofarecordlinkagelargeblocksnomatterhowgoodtheblockingkeyis.system.Forexample,whenblockingonsurnameforanAnglo-Celticpopulation,SmithandTaylorarepopulousan

5、dresultinverylargeblocksizes.Theefficiencyofaandgovernmentadministration.Theseapplicationscanblockingmethodishinderedbytheselargeblockssincebeclassedas‘administrative’,becauserecordlinkageistheresultingnumberofrecordpairsisdominatedbyusedtomakedecision

6、sandtakeactionsregardinganthesizesoftheselargeblocks.Inthispaper,wepresentindividualentity.afilteringalgorithmtopost-processlargeblockstoInmanydataminingprojectsitisnecessarytoenhancetheblockingefficiency.collateinformationaboutanentityfrommorethanExper

7、imentalresultsshowthatourfilteringalgo-onedatasource.Ifauniqueidentifierorkeyoftherithmcanreducethenumberofrecordpairsproducedentityofinterestisavailableinallofthedatasourcesbythestandardblockingmethodby88%onasmalltobelinked,conventional‘join’operation

8、scanbeusedreal-worlddataset.Thealgorithmalsoreducestheforrecordlinkage,whichassumeserror-freeidentifyingnumberofrecordpairsgeneratedbya3-passstandardfieldsandlinksrecordsthatexactlymatchontheseblockingmethodby50%onseveralsynthetictestdataidentifyingfie