Hierarchical Text Classification and Evaluation

《Hierarchical Text Classification and Evaluation》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、HierarchicalTextClassificationandEvaluationAixinSunandEe-PengLimCenterforAdvancedInformationSystemsNanyangTechnologicalUniversityNanyangAvenue,Singapore639798,Singaporesunaixin@pmail.ntu.edu.sgaseplim@ntu.edu.sgAbstractclassificationproblemusingadivide-and-conquerap-proach.Attherootlevelinthecate

2、goryhierarchy,adocu-HierarchicalClassificationreferstoassigningofoneormentcanbefirstclassifiedintooneormoresub-categoriesmoresuitablecategoriesfromahierarchicalcategoryspaceusingsomeflatclassificationmethod(s).Theclassifica-toadocument.Whilepreviousworkinhierarchicalclassi-tioncanberepeatedonthedocum

3、entineachofthesub-ficationfocusedonvirtualcategorytreeswheredocumentscategoriesuntilthedocumentreachessomeleafcategoriesareassignedonlytotheleafcategories,weproposeatop-orcannotbefurtherclassifiedintoanysub-categories.Adownlevel-basedclassificationmethodthatcanclassifyfewhierarchicalclassificationm

4、ethodshavebeenproposeddocumentstobothleafandinternalcategories.Asthestan-recently[1,2,10,13,15,17].Inmostofthehierarchicaldardperformancemeasuresassumeindependencebetweenclassificationmethods,thecategoriesareorganizedintree-categories,theyhavenotconsideredthedocumentsincor-likestructures.Onthewh

5、ole,wecanidentifyfourdistinctrectlyclassifiedintocategoriesthataresimilarornotfarcategorystructuresfortextclassification.Theyare:fromthecorrectonesinthecategorytree.Wetherefore1.Virtualcategorytree:Inthiscategorystructure,cate-proposetheCategory-SimilarityMeasuresandDistance-goriesareorganizedasa

6、tree.Eachcategorycanbe-BasedMeasurestoconsiderthedegreeofmisclassificationlongtoatmostoneparentcategoryanddocumentscaninmeasuringtheclassificationperformance.Anexperi-onlybeassignedtotheleafcategories[2].menthasbeencarriedouttomeasuretheperformanceofourproposedhierarchicalclassificationmethod.Ther

7、e-2.Categorytree:Thisisanextensionofthevirtualcat-sultsshowedthatourmethodperformswellforReuterstextegorytreethatallowsdocumentstobeassignedintocollectionwhenenoughtrainingdocumentsaregivenandbothinternalandleafcategories[15].then