問答系統(tǒng)中信息檢索和答案抽取的研究.pdf

1、963881論文題目：問答系統(tǒng)中信息檢索和答案抽取的研究StudiesonInformationRetrievalandAnswerExtractioninQuestion—answeringSystem研究生姓名：蔡建山指導教師姓名：堡呈墓單位：夔出疆煎盔堂論文提交日期學位授予口期授予單位：蝗出型撻虛堂論文評閱人：王盍蟄單位：丕jb厶堂亙疊單位：夔出登遮丕堂鞍山科技火學頌：l論文AbstractAbstractSearchengine

2、providesaninformationretrievalinterfaceofalarge—scalecorpusKeywordlist，whichistranslatedfromtheuserinformationneed，isprovidedtosearchengine，whichreturnsrelevantdocumentsQuestion—answeringsystemisanewgenerationofsearcheng

3、ineItacceptsquestionsthataredescribedinnaturallanguageandreturnsthejustrightanswerItconcernsmorewithretrievingthejustrightanswerthanwithretrievingdocumentswhichsatisfiesagivenqueryQ—KillerisaChinesequestion—answeringsyst

4、em，whichextractanswerusingNLPandclassicIRQ—Killeriscomposedbyquestionanalysis，IRandanswerextractionThisthesispresentsourstudyon1RandanswerextractionIntheIRmodel，Q—KillerresolvestwomainquestionsofinvertedfileFirstly，wedes

5、ignathree—layerindexingfile，whichisbasedonChinesedictionaryanalysisItisproventobeeffectiveonlargescaleChinesetextretrievalSecondly，wepresentaselectivedocumentindexingmethod，whichonlyindexhighweightdocumentstesolvethehigh

6、frequentwordproblemItisalsoproventobeeffectiveonhigh—frequentwordretrievalAnanswerofe—KillerismadeupofananswerandinformationtosupporttheanswerTheanswerisextractedfromquery—biasedsummarizationofrelevantdocuments，andthesup

7、portinformationisquerybiasedsummarizationofrelevantdocumentInthispaperwepresentanapproachforautomaticallylearningquestionpatterns(alongwithdeterminingtheirprecision)fromtheweb，forgiventypesofquestionsOurmethodusesthemach

8、inelearningtechniqueofbootstrappingtobuildquestionpatternsstartingwithonlyafewexamplesofQApairsWeproposeasnippetextractionalgorithmusingslidewindowsWealsopresentaperformanceevaluationofdynamicsurmnaryUsingthesamedocument

9、set，weperformanexperimentcontrastwithGoogleandBaiduResultsshowthatthesummarygeneratedbythisalgorithmconciselyoutlinesthecontentofatext，and，comparedwithGoogleandBaidushowsabetteraverageperformance5％thanGoogleand1O％thanBai