資源描述:
《Convolutional Neural Networks for Sentence Classification 》由會員上傳分享,免費在線閱讀,更多相關內容在學術論文-天天文庫。
1、ConvolutionalNeuralNetworksforSentenceClassi?cationYoonKimNewYorkUniversityyhk255@nyu.eduAbstractlocalfeatures(LeCunetal.,1998).Originallyinventedforcomputervision,CNNmodelshaveWereportonaseriesofexperimentswithsubsequentlybeenshowntobeeffectiveforNLPconvolutionalneuralnetwo
2、rks(CNN)andhaveachievedexcellentresultsinsemantictrainedontopofpre-trainedwordvec-parsing(Yihetal.,2014),searchqueryretrievaltorsforsentence-levelclassi?cationtasks.(Shenetal.,2014),sentencemodeling(Kalch-WeshowthatasimpleCNNwithlit-brenneretal.,2014),andothertraditionalNLPt
3、lehyperparametertuningandstaticvec-tasks(Collobertetal.,2011).torsachievesexcellentresultsonmulti-Inthepresentwork,wetrainasimpleCNNwithplebenchmarks.Learningtask-speci?conelayerofconvolutionontopofwordvectorsvectorsthrough?ne-tuningoffersfurtherobtainedfromanunsupervisedneu
4、rallanguagegainsinperformance.Weadditionallymodel.ThesevectorsweretrainedbyMikolovetproposeasimplemodi?cationtothear-al.(2013)on100billionwordsofGoogleNews,chitecturetoallowfortheuseofbothandarepubliclyavailable.1Weinitiallykeepthetask-speci?candstaticvectors.TheCNNwordvecto
5、rsstaticandlearnonlytheotherparam-modelsdiscussedhereinimproveupontheetersofthemodel.Despitelittletuningofhyper-stateofthearton4outof7tasks,whichparameters,thissimplemodelachievesexcellentincludesentimentanalysisandquestionresultsonmultiplebenchmarks,suggestingthatclassi?cat
6、ion.thepre-trainedvectorsare‘universal’featureex-1Introductiontractorsthatcanbeutilizedforvariousclassi?ca-tiontasks.Learningtask-speci?cvectorsthroughDeeplearningmodelshaveachievedremarkable?ne-tuningresultsinfurtherimprovements.Weresultsincomputervision(Krizhevskyetal.,?na
7、llydescribeasimplemodi?cationtothearchi-2012)andspeechrecognition(Gravesetal.,2013)tecturetoallowfortheuseofbothpre-trainedandinrecentyears.Withinnaturallanguageprocess-task-speci?cvectorsbyhavingmultiplechannels.ing,muchoftheworkwithdeeplearningmeth-odshasinvolvedlearningwo
8、rdvectorrepresenta-OurworkisphilosophicallysimilartoRazaviantionsthroughneu