基于圖和轉(zhuǎn)移算法相結(jié)合的中文依存關(guān)系解析.pdf

1、大連理工大學(xué)碩士學(xué)位論文基于圖和轉(zhuǎn)移算法相結(jié)合的中文依存關(guān)系解析姓名：劉春紅申請學(xué)位級別：碩士專業(yè)：計算機應(yīng)用技術(shù)指導(dǎo)教師：黃德根20091216大連理工大學(xué)碩士學(xué)位論文IntegratingGraphBasedandTransitionBasedChineseDependencyParserAbstractSyntaxparsingisoneofthemostimportanttasksofnaturallanguageprocess

2、ing，Dependencyanalysisiswidelyusedinmachinetranslation，informationretrievalandautomaticabstract。Dependencyrelationspresentrelationsbetweenwords，andareeasytobeconvertedintosemanticdependencyWordisthesmallestelementofsente

3、nce，andthedependencybasedonwordsanalysiscarlrepresentdeepsyntaxrelation，ThisthesisresearchesChinesedependencybetweenwordsAtpresent，Nivre’SalgorithmhasbeenusedforEnglishandJapanesedependencyanalysis，andhasachievedgoodrese

4、archresultsThegrammaticalstructureofChineseisfardifferentfromotherlanguagesanddependencyanalysisisverycomplexCurrentlythemethodsofChinesedependencyparsingcallbedividedintotransitionbasedapproachandgraphbasedapproach，them

5、ainrepresentativeoftransition—basedapproachisNivre’Salgorithmandthemainrepresentativeoftransition—basedapproachisMaximumspanningtree(MST)algorithmNivre’Salgorithmbelongstodeterministicdependencyparsingandparsesdependency

6、relationwitllthefeaturesaroundthewordsItperformsparsingbygreedilytakingthehighestscoringtransitionoutofeveryparserstateandtheparsinghistoryCanbeusedinnextparsinguntilwehavederivedacompletedependencygraphMSTalgorithmgetam

7、odelforscoringpossibledependencygraphsforagivensentence，performparsingbysearchingforthehighestscoringgraphWeCan’tgetthepartparsingresultuntilthesentenceisparsedoverIIlthispaper，weproposetwokindsofintegratingmethodsaccord

8、ingtothecomplementaryrelationshipofNivre’SalgorithmandmaximumspanningtreealgorithmOnemethodisbasedontheimpactfactorsofexistencethatMaximumspanningtreealgorithmactasthebasicalgorithmmodifiedbytheexistenceoftheresultsofNiv

9、realgorithmTheothermethodisbasedontheimpactfactorsofdependencedegreethatMaximumspanningtreealgorithmactasthebasicalgorithmmodifiedbythedependencydegreeoftheresultsofNivrealgorithmExperimentsusingPennChineseTreebank50show

10、thattheproposedtwocombinationmethodsarebetterthantheoriginalsinglealgorithmandthemodelbasedontheimpactfactorsofdependencedegreereachesthehighestaccuracywhichis8687％KeyWords：ChineseDependencyAnalysis；MaximumSpanningTreeAl