基于知網(wǎng)的中文問答系統(tǒng)問題理解與研究

1、大連理工大學(xué)碩士學(xué)位論文基于知網(wǎng)的中文問答系統(tǒng)問題理解與研究姓名：姜東洋申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用技術(shù)指導(dǎo)教師：王興元20071216大連理工大學(xué)碩學(xué)位論文ResearchonQuestionUndstandingofChineseQuestionAnsweringowingtoHowNetAbstraetQuestionAnswering(QA)systemisalladvancedformoftheinformationre

2、trievalsystemAutomaticQuestionAnsweringSystem(QA)callsatisfytheusers’retrievalcommandbetteranditcan西vetheanswerthatuserneedindeed，SOithasthecharacteristicsofconvenience、celerityandhighefficiencyThereasonthatQAtechnologyt

3、hrivingisthedemandthatpeoplehopefindtheinformationquicklyandaccuratelyQAsystemispaidmoreattentionanditisapromisingresearchdirectioninthefieldofArtificialIntelligenceandNaturalLanguageProcessing(1)TheQuestionUnderstanding

4、(QU)technologyofQAisstudiedinthispaperQUisthefoundationofQAAccurateresultscanbereturnedonlybyanalyzingandunderstandingthequestionsproperlyQuestionclassificationisthecoreoftheQu1hordertounderstandquestionsdeeplyandgraspth

5、equestionssemantic，asemanticdescribingmodelnamedEventFrameModeliSadvancedAndwetakeHOWNeteventsememesandattributesememesasthecentralpartoftheEventFrame(2)Onbasisoftheideaof“question’Styperepresentsquestionintention”，quest

6、ionintentionbasedquestionclassificationisproposedinthispaperAndintheEventFramesystem，theformalquestionintentiondefinitionisshownasfollowNamelyquestionintentionisrepresentwiththecoreoftheeventflame，aswellastheanswertypean

7、dthesemanticroleofthequery—focusPlacedqueriesraisedthequestionofintentclassificationsystemAtthesametime，thispaperpresentsaEnventFramebasedontheissueoftheincidentsemanticdescriptionmodel，inthismodel，wearegivenformaldefini

8、tiondoubttheintentionThroughthehelpofsemanticHowNetresources，thesizeoftheproblemspacetocarryoutevaluation(3)ThestudyofautomaticrecognitionofthequestionintentionAtwolevelMaximumEntropy(ME)modelisproposedtecognizetheanswer

9、type，whoseresultis2％higherthanotherapproachesonthesametrainingtestingcorpusAHownetconfidenceevaluationbasedmethodtoextractthecentralpartoftheEventFrameisadvanced，whichcouldextractthesamecentralpartinquestionswhichhavethe

10、sameintentionbutindifferentexpressionformsMEmodelisusedforthequery—focusSemanticRoleLabeling(SRL、InSRLprocessfeatureselectionandoptimizationiSstudieddeeplyandmodifiedMutuallnformationmodelbasedfeatureselectionmethodandad