面向問答系統(tǒng)的實(shí)體識(shí)別與分類研究.pdf

1、沈陽(yáng)航空工業(yè)學(xué)院碩士學(xué)位論文面向問答系統(tǒng)的實(shí)體識(shí)別與分類研究姓名：周波申請(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：蔡?hào)|風(fēng)20090327沈Ij航空丁業(yè)學(xué)院碩十!學(xué)位論文AbstractQuestionAnsweringSystem(QAS)isbothhigh—levelformofinformationretrievalandresearchemphasisandhotspotinthisfieldItisanintegrated

2、technicalprocessingsystemwhichissupportedbythetechniqueofwordsegment，lexicalanalysis，retrieval，entityrecognitionandanswerextractionOneofthekeytechniqueofQASisentityrecognitionandtaggingwhichdirectlydecidethejudgementofqu

3、estiontypeandtheextractionofanswersThejudgementofquestiontypeisveryrelatedtoentityrecognitioninQASAccordingtothedetaildemands，questiontypehasdifferentclassificationsIngeneral，thefactoidquestionsarealwaysrelatedtotheperso

4、n，time，place，quantityandetcButforsometypetherearedifferentlevels，forexampleaplacecanbesubdividedintocountryprovince，citymountain，riverlakeandSOonThepaperemphasizesentityrecognitionandclassificationmethodbythelevelsanddiv

5、ersityoftheentityMainworkis1Thisthesisproposesamethodforentityrecognitionandclassificationbasedonrulesandstatistic，whichismainlycombinedtheentityclassifieddictionaryandconditionalrandomfields(CRFs)，throughtheresearchonqu

6、estionclassificationofQASofuniversalfieldandtheentityclassifiedsystem2ThisthesisdoesresearchonloggingwordsrecognitionbasedontheentityclassfieddictionaryestablishesaentityclassfieddictionaryforQASwhichincludesalmost300tho

7、usandwordsbythewaysoftheopendatabaseoftheChineseWikipediaandetc，thedictionaryisstoredinmemoryastheformofindextree，thenusethisdictionarytogoestherecognitionandclassificationofloggingwords，improvestheeffectoftheentityrecog

8、nitionandclassification3Thisthesisprovidesatwo—stagesmethodfornamedentityrecognitionbasedonCRFs，theFscoreoftwostagesis8630％inclosetestwhichislower15％thantheone—stageresult8801％，howeveritcanreducethetemporalcomplexityto20