资源描述:
《efficient online spherical k-means clusteringnew》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库。
1、EfficientOnlineSphericalK-meansClusteringShiZhongDepartmentofComputerScienceandEngineeringFloridaAtlanticUniversity,BocaRaton,FL33431Email:zhong@cse.fau.eduAbstract—Thesphericalk-meansalgorithm,i.e.,thek-meansOurmainobjectiveinthispaperistosignificantlyimprovealgorithmwit
2、hcosinesimilarity,isapopularmethodforcluster-theclusteringqualityfork-means,yettoretainitsefficiency.inghigh-dimensionaltextdata.Inthisalgorithm,eachdocumentAtraditionalwayofimprovingthequalityofk-meansistoaswellaseachclustermeanisrepresentedasahigh-dimensionalemploybett
3、eroptimization(annealing)techniques(e.g.,[4],unit-lengthvector.However,ithasbeenmainlyusedinbatchmode.Thatis,eachclustermeanvectorisupdated,onlyafter[5])toavoid(bad)localoptima.However,thesemethodsarealldocumentvectorsbeingassigned,asthe(normalized)averageusuallytootime
4、-consumingtobevaluableinpractice.Localofallthedocumentvectorsassignedtothatcluster.Thispapersearchwasusedin[6]toimprovetheobjectivefunctionvalueinvestigatesanonlineversionofthesphericalk-meansalgorithmforsphericalk-means.Thealgorithmtriestofindadatapoint,basedonthewell-k
5、nownWinner-Take-Allcompetitivelearning.andbyswitchingittoadifferentcluster,toimprovetheInthisonlinealgorithm,eachclustercentroidisincrementallyupdatedgivenadocument.Wedemonstratethattheonlinesphericalk-meansobjectivefunction.Thisapproach,however,sphericalk-meansalgorith
6、mcanachievesignificantlybetterworksfavorablyonlyforsmalltextdatasets.Forlargeclusters,clusteringresultsthanthebatchversion,especiallywhenanthe“firstvariation”techniqueisunlikelytobeeffective,annealing-typelearningratescheduleisused.Wealsopresentasadmittedin[6].Inaddition,
7、theproposedlocalsearchheuristicstoimprovethespeed,yetalmostwithoutlossofalgorithmseemstobecomputationallycostlyforclusteringclusteringquality.largetextdatasets.I.INTRODUCTIONAnotherapproachistoreplacethebatchupdateofclusterDocumentclusteringhasbecomeanincreasinglyimport
8、antcentroidsbyanonlinelearningstrategy.Thestandardk-meanstechniqueforunsuperviseddocumentorganization,automati