资源描述：
《Bag of Tricks for Efficient Text Classification》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、BagofTricksforEfficientTextClassificationArmandJoulinEdouardGravePiotrBojanowskiTomasMikolovFacebookAIResearch{ajoulin,egrave,bojanowski,tmikolov}@fb.comAbstractextensionofthesemodelstodirectlylearnsentencerepresentations.WeshowthatbyincorporatingThispaperproposesasimpleandefficientap-additionalstatist
2、icssuchasusingbagofn-grams,proachfortextclassificationandrepresenta-tionlearning.OurexperimentsshowthatourwereducethegapinaccuracybetweenlinearandfasttextclassifierfastTextisoftenonpardeepmodels,whilebeingmanyordersofmagnitudewithdeeplearningclassifiersintermsofac-faster.curacy,andmanyordersofmagnitude
3、fasterOurworkiscloselyrelatedtostan-fortrainingandevaluation.Wecantraindardlineartextclassifiers(Joachims,1998;fastTextonmorethanonebillionwordsMcCallumandNigam,1998;Fanetal.,2008).inlessthantenminutesusingastandardmul-SimilartoWangandManning(2012),ourmoti-ticoreCPU,andclassifyhalfamillionsen-tencesa
4、mong312Kclassesinlessthanavationistoexploresimplebaselinesinspiredbyminute.modelsusedforlearningunsupervisedwordrepre-sentations.AsopposedtoLeandMikolov(2014),1Introductionourapproachdoesnotrequiresophisticatedinfer-enceattesttime,makingitslearnedrepresentationsBuildinggoodrepresentationsfortextclas
5、si-easilyreusableondifferentproblems.Weevaluateficationisanimportanttaskwithmanyap-thequalityofourmodelontwodifferenttasks,plications,suchaswebsearch,informationnamelytagpredictionandsentimentanalysis.retrieval,rankinganddocumentclassifica-tion(Deerwesteretal.,1990;PangandLee,2008).2Modelarchitecturea
6、rXiv:1607.01759v2[cs.CL]7Jul2016Recently,modelsbasedonneuralnetworkshavebecomeincreasinglypopularforcomputingAsimpleandefficientbaselineforsentencesentencerepresentations(Bengioetal.,2003;classificationistorepresentsentencesasbagofCollobertandWeston,2008).Whilethesewords(BoW)andtrainalinearclassifier,f
7、ormodelsachieveverygoodperformanceinexamplealogisticregressionorsupportvec-practice(Kim,2014;ZhangandLeCun,2015;tormachine(Joachims,1998;Fanetal.,2008).Zhangetal.,2015),theytendtoberelativelyslowHowev
