基于多项式核的结构化有向树数据聚类算法.pdfVIP

ISSN 1000-9825，CODEN RUXUfEW E—mail：jOS@iscas．ac．cn JournalofSoftware，Vo1．19，No．12，December2008，PP．3147—3160 http：／／wwwjos．org．cn DOI：10．3724／SRJ．1001．2008．03147 Tbl／Fax：+86—10 @2008byJournalofSoftware．Allrightsreserved． 基于多项式核的结构化有向树数据聚类算法 丁军娣1,2，马儒宁，陈松灿 。(南京理工大学 计算机科学与技术学院，江苏 南京 210094) (南京航空航天大学 信息科学与技术学院，江苏 南京 210016) (南京航空航天大学 理学院，江苏 南京 210016) PolynomialKernelBasedStructuralClusteringAlgorithm byBuildingDirectedTrees DING Jun．Di，-， MA Ru-Ning3 CHEN Song．CanE ， (SchoolofComputerScienceandTechnology，NnajingUniversityofComputerSciencenadTechnoloyg，Nanjing210094，China) (DepartmentofComputerSciencenadEngineering，NnajingUniversiytofAeronauticsandAstronautics，Nnajing210016，China) (DepartmentofScience，NanjingUniversiytofAeronauticsAstronautics，Nanjing210016，China) +Correspondingauthor：E—mail：s．chen@nuaa．edu．cn DingJD，M aRN，Chen SC．Polynomialkernelbased structuralclusteringalgorithm bybuilding directed trees．JournalofSoftware，2008，19(12)：3147—3160．hrtp：／／www．jos．org．cn／1000-9825／19／3147．htm Abstract： Withintheinternalorganizationofthedata，thedatapointsrespectivelyplaythreedifferentstructural roles：thehub，centroidandoutlier．Theneighborhood-baseddensityfactor(NDE)usedintheneighborhoodbased clustering(NBC)algorithm hastheabilityofidentifyingwhichpointsactashubs，centriodsoroutliersin separated-welldataset．However,NDF oftenworkspoorly in thecircumstancesofnoiseandoverlapping．This paperintroducesapolynomialkernelbasedneighborhooddensiytfactor(PKNDF)toaddressthisissue．Relyingon thePKNDF,astructuraldata clusteringalgorithm isfurtherpresentedwhich can findallsalientclusterswith arbitraryshapesandunbalancedsizesinanoisyoroverlappingdataset．Itbuildsclustersintotheframeworkof directedtreesingraphtheory andtherebyeachpointisscannedonlyonceinthep