Taming Recurrent Neural Networks for Better Summarization _ Abigail See英文学习资料

《Taming Recurrent Neural Networks for Better Summarization _ Abigail See英文学习资料》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、TamingRecurrentNeuralNetworksforBetterSummarizationAPRIL16,2017Thisisablogpostaboutourlatestpaper,GetToThePoint:SummarizationwithPointer-GeneratorNetworks,toappearatACL2017.Codewillbereleasedshortly.Theinternetagehasbroughtunfathomablymassiveamountsofinform

2、ationtothefingertipsofbillions–ifonlywehadtimetoreadit.Thoughourliveshavebeentransformedbyreadyaccesstolimitlessdata,wealsofindourselvesensnaredbyinformationoverload.Forthisreason,automatictextsummarization–thetaskofautomaticallycondensingapieceoftexttoasho

3、rterversion–isbecomingincreasinglyvital.TwotypesofsummarizationTherearebroadlytwoapproachestoautomatictextsummarization:extractiveandabstractive.Extractiveapproachesselectpassagesfromthesourcetext,thenarrangethemtoformasummary.Youmightthinkoftheseapproaches

4、aslikeahighlighter.Abstractiveapproachesusenaturallanguagegenerationtechniquestowritenovelsentences.Bythesameanalogy,theseapproachesarelikeapen.Thegreatmajorityofexistingapproachestoautomaticsummarizationareextractive–mostlybecauseitismucheasiertoselecttext

5、thanitistogeneratetextfromscratch.Forexample,ifyourextractiveapproachinvolvesselectingandrearrangingwholesentencesfromthesourcetext,youareguaranteedtoproduceasummarythatisgrammatical,fairlyreadable,andrelatedtothesourcetext.Thesesystems(severalareavailableo

6、nline)canbereasonablysuccessfulwhenappliedtomid-lengthfactualtextsuchasnewsarticlesandtechnicaldocuments.Ontheotherhand,theextractiveapproachistoorestrictivetoproducehuman-likesummaries–especiallyoflonger,morecomplextext.ImaginetryingtowriteaWikipedia-style

7、plotsynopsisofanovel–say,GreatExpectations–solelybyselectingandrearrangingsentencesfromthebook.Thiswouldbeimpossible.Foronething,GreatExpectationsiswritteninthefirstpersonwhereasasynopsisshouldbeinthethirdperson.Moreimportantly,condensingwholechaptersofacti

8、ondowntoasentencelikePipvisitsMissHavishamandfallsinlovewithheradopteddaughterEstellarequirespowerfulparaphrasingthatispossibleonlyinanabstractiveframework.Inshort:abstractivesummarizationmaybediicult