資源描述:
《Hierarchical Text Classification and Evaluation》由會(huì)員上傳分享,免費(fèi)在線(xiàn)閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、HierarchicalTextClassi?cationandEvaluationAixinSunandEe-PengLimCenterforAdvancedInformationSystemsNanyangTechnologicalUniversityNanyangAvenue,Singapore639798,Singaporesunaixin@pmail.ntu.edu.sgaseplim@ntu.edu.sgAbstractclassi?cationproblemusingadivide-and-conquerap-proach.Attherootlevelinthecate
2、goryhierarchy,adocu-HierarchicalClassi?cationreferstoassigningofoneormentcanbe?rstclassi?edintooneormoresub-categoriesmoresuitablecategoriesfromahierarchicalcategoryspaceusingsome?atclassi?cationmethod(s).Theclassi?ca-toadocument.Whilepreviousworkinhierarchicalclassi-tioncanberepeatedonthedocum
3、entineachofthesub-?cationfocusedonvirtualcategorytreeswheredocumentscategoriesuntilthedocumentreachessomeleafcategoriesareassignedonlytotheleafcategories,weproposeatop-orcannotbefurtherclassi?edintoanysub-categories.Adownlevel-basedclassi?cationmethodthatcanclassifyfewhierarchicalclassi?cationm
4、ethodshavebeenproposeddocumentstobothleafandinternalcategories.Asthestan-recently[1,2,10,13,15,17].Inmostofthehierarchicaldardperformancemeasuresassumeindependencebetweenclassi?cationmethods,thecategoriesareorganizedintree-categories,theyhavenotconsideredthedocumentsincor-likestructures.Onthewh
5、ole,wecanidentifyfourdistinctrectlyclassi?edintocategoriesthataresimilarornotfarcategorystructuresfortextclassi?cation.Theyare:fromthecorrectonesinthecategorytree.Wetherefore1.Virtualcategorytree:Inthiscategorystructure,cate-proposetheCategory-SimilarityMeasuresandDistance-goriesareorganizedasa
6、tree.Eachcategorycanbe-BasedMeasurestoconsiderthedegreeofmisclassi?cationlongtoatmostoneparentcategoryanddocumentscaninmeasuringtheclassi?cationperformance.Anexperi-onlybeassignedtotheleafcategories[2].menthasbeencarriedouttomeasuretheperformanceofourproposedhierarchicalclassi?cationmethod.Ther
7、e-2.Categorytree:Thisisanextensionofthevirtualcat-sultsshowedthatourmethodperformswellforReuterstextegorytreethatallowsdocumentstobeassignedintocollectionwhenenoughtrainingdocumentsaregivenandbothinternalandleafcategories[15].then