基于文本语料库中分类法学习的综述：问题、资源和必威体育精装版进展A Short Survey on Taxonomy Learning from Text Corpora - Issues, Resources and Recent Advances.pdfVIP

AShortSurveyonTaxonomyLearningfromTextCorpora:

Issues,ResourcesandRecentAdvances

∗

ChengyuWang,XiaofengHe,AoyingZhou

ShanghaiKeyLaboratoryofTrustworthyComputing,

SchoolofComputerScienceandSoftwareEngineering,EastChinaNormalUniversity

chywang2013@,{xfhe,ayzhou}@

Abstractcorporamayvaryinsize,topicandquality.It

isunlikelytodevelopa“one-size-fits-all”solu-

Ataxonomyisasemantichierarchy,con-

tionforallscenarios.Forexample,givenanex-

sistingofconceptslinkedbyis-arelations.

tremelylargecorpus,Hearst-patternbasedmethod

Whilealargenumberoftaxonomieshave

isemployedtobuildProbase(Wuetal.,2012).

beenconstructedfromhuman-compiled

Fordomain-specificcorpora,learninghypernymy

resources(e.g.,Wikipedia),learningtax-

embeddingismoresuitable(Luuetal.,2016b).

onomiesfromtextcorporahasreceiveda

ii)Theaccuracyoffree-texttaxonomiesisusu-

growinginterestandisessentialforlong-

allylowerthanmanyWikipedia-basedtaxonomies

tailedanddomain-specificknowledgeac-

becauseitisdifficul