简介:Inthispaper,aneweffectivemethodisproposedtofindclassassociationrules(CAR),togetusefulclassassociationrules(UCAR)byremovingthespuriousclassassociationrules(SCAR),andtogenerateexceptionclassassociationrules(ECAR)foreachUCAR.CARmining,whichintegratesthetechniquesofclassificationandassociation,isofgreatinterestrecently.However,ithastwodrawbacks:oneisthatalargepartofCARsarespuriousandmaybemisleadingtousers;theotheristhatsomeimportantECARsaredifficulttofindusingtraditionaldataminingtechniques.Themethodintroducedinthispaperaimstogetovertheseflaws.Accordingtoourapproach,ausercanretrievecorrectinformationfromUCARsandknowtheinfluencefromdifferentconditionsbycheckingcorrespondingECARs.Experimentalresultsdemonstratetheeffectivenessofourproposedapproach.
简介:Softwaremodularizationisatechniqueusedtodivideasoftwaresystemintoindependentmodules(packages)thatareexpectedtobecohesiveandlooselycoupled.However,assoftwaresystemsevolveovertimetomeetnewrequire-ments,theirmodularizationsbecomecomplexandgraduallyloosetheirquality.Thus,itischallengingtoautomaticallyoptimizetheclasses'distributioninpackages,alsoknownasremodularization.Toalleviatethisissue,weintroduceanewapproachtooptimizesoftwaremodularizationbymovingclassestomoresuitablepackages.Inadditiontoimprovingdesignqualityandpreservingsemanticcoherence,ourapproachtakesintoconsiderationtherefactoringeffortasanobjectiveinitselfwhileoptimizingsoftwaremodularization.WeadapttheElitistNon-dominatedSortingGeneticAlgorithm(NSGA-Ⅱ)ofDebetal.tofindthebestsequenceofrefactoringsthat1)maximizestructuralquality,2)maximizesemanticcohesivenessofpackages(evaluatedbyasemanticmeasurebasedonWordNet),and3)minimizetherefactoringeffort.Wereporttheresultsofanevaluationofourapproachusingopen-sourceprojects,andweshowthatourproposalisabletoproduceacoherentandusefulsequenceofrecommendedrefactoringsbothintermsofqualitymetricsandfromthedeveloper'spointsofview.
简介:在这份报纸,我们在基于社区的问答处理答案检索问题。为了充分捕获在问题答案之间的相互作用,配对,我们建议原来的张肌为在他们之间的关联建模的神经网络。问题和候选人答案独立被嵌进不同潜伏的语义空格,并且3方法张肌然后被利用为在潜伏的语义之间的相互作用建模。适当地初始化网络层,我们建议一个新奇算法打电话降噪张肌autoencoder(DTAE),然后实现用降噪嵌入张肌层上的层和DTAE的词上的autoencoders(DAE)的layerwisepretraining策略。试验性的结果显示出那我们神经网络超过的张肌有另外的竞争神经网络方法的各种各样的基线,和我们的pretrainingDTAE策略改进系统性能和坚韧性。
简介:Weintroduceanalmost-automatictechniqueforgenerating3Dcarstylingsurfacemodelsbasedonasingleside-viewimage.Ourapproachcombinesthepriorknowledgeofcarstylinganddeformablecurvenetworkmodeltoobtainanautomaticmodelingprocess.Firstly,wedefinetheconsistentparameterizedcurvetemplatefor2Dand3Dcaserespectivelybyanalyzingthecharacteristiclinesforcarstyling.Then,asemi-automaticextractionfromaside-viewcarimageisadopted.Thirdly,statisticmorphablemodelof3Dcurvenetworkisusedtogettheinitialsolutionwithsparsepointconstraints.Withonlyafewpost-processingoperations,theoptimizedcurvenetworkmodelsforcreatingsurfacesareobtained.Finally,thestylingsurfacesareautomaticallygeneratedusingtemplate-basedparametricsurfacemodelingmethod.Morethan503Dcurvenetworkmodelsareconstructedasthemorphabledatabase.Weshowthatthisintelligentmodelingtoolsimplifiestheexhaustedmodelingtask,andalsodemonstratemeaningfulresultsofourapproach.