資源描述：
《Convolutional Neural Networks for Sentence Classification》由會員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、ConvolutionalNeuralNetworksforSentenceClassi?cationYoonKimNewYorkUniversityyhk255@nyu.eduAbstractlocalfeatures(LeCunetal.,1998).Originallyinventedforcomputervision,CNNmodelshaveWereportonaseriesofexperimentswithsubsequentlybeenshowntobeeffectiveforNLPconvolutionalneuralnetworks(CNN)andhave
2、achievedexcellentresultsinsemantictrainedontopofpre-trainedwordvec-parsing(Yihetal.,2014),searchqueryretrievaltorsforsentence-levelclassi?cationtasks.(Shenetal.,2014),sentencemodeling(Kalch-WeshowthatasimpleCNNwithlit-brenneretal.,2014),andothertraditionalNLPtlehyperparametertuningandstati
3、cvec-tasks(Collobertetal.,2011).torsachievesexcellentresultsonmulti-Inthepresentwork,wetrainasimpleCNNwithplebenchmarks.Learningtask-speci?conelayerofconvolutionontopofwordvectorsvectorsthrough?ne-tuningoffersfurtherobtainedfromanunsupervisedneurallanguagegainsinperformance.Weadditionallym
4、odel.ThesevectorsweretrainedbyMikolovetproposeasimplemodi?cationtothear-al.(2013)on100billionwordsofGoogleNews,chitecturetoallowfortheuseofbothandarepubliclyavailable.1Weinitiallykeepthetask-speci?candstaticvectors.TheCNNwordvectorsstaticandlearnonlytheotherparam-modelsdiscussedhereinimpro
5、veupontheetersofthemodel.Despitelittletuningofhyper-stateofthearton4outof7tasks,whichparameters,thissimplemodelachievesexcellentincludesentimentanalysisandquestionresultsonmultiplebenchmarks,suggestingthatclassi?cation.thepre-trainedvectorsare‘universal’featureex-1Introductiontractorsthatc
6、anbeutilizedforvariousclassi?ca-tiontasks.Learningtask-speci?cvectorsthroughDeeplearningmodelshaveachievedremarkable?ne-tuningresultsinfurtherimprovements.Weresultsincomputervision(Krizhevskyetal.,?nallydescribeasimplemodi?cationtothearchi-2012)andspeechrecognition(Gravesetal.,2013)tecture
7、toallowfortheuseofbothpre-trainedandinrecentyears.Withinnaturallanguageprocess-task-speci?cvectorsbyhavingmultiplechannels.ing,muchoftheworkwithdeeplearningmeth-odshasinvolvedlearningwordvectorrepresenta-OurworkisphilosophicallysimilartoRazaviantionsthroughneu
