基于特征強化的中文產(chǎn)品評論挖掘研究

1、合肥工業(yè)大學碩士學位論文基于特征強化的中文產(chǎn)品評論挖掘研究姓名：史麗君申請學位級別：碩士專業(yè)：計算機軟件與理論指導教師：胡學鋼201104IIIResearchonProductReviewsMininginChinesebasedonReinfcementofFeaturesAbstractWiththerapiddevelopmentoftheelectroniccommercebusinesstheincreaseofonline

2、shoppingcustomersonlineproductreviewsasawayofreflectisdramaticallyincreasing.Therefetheautomaticminingonthesereviewsisimptantfthepotentialcustomersmerchantsproductmanufacturersithasbecomethetrendtousethemachinelearningme

3、thodstominingthereviews.Howevertheshtlengthweaksignaltodescribetheacteristicsofreviewsmakesapplicationofmachinelearningmethodsdifficulttoapplyinproductreviewmining.MotivatedbythiswefocusonminingChineseproductreviewsbased

4、onReinfcementofFeaturesindocumentlevelsentimentclassificationfeaturebasedproductopinionmining.Themaincontentsareasfollows:Basedontheacteristicsthatfeatureinfmationofreviewisweakamethodbasedoncrelationfeatureswasintroduce

5、dtoanalysissentiment.InthismethodtoenhancethefeaturesofreviewsweusedthecrelationrulesbetweenfeatureitemswhichwereminedbyFPGrowthalgithm.FinallyIGfeatureionSVMclassifyalgithmwereusedtoclassifysentimentofreviews.Experiment

6、alresultsshowthattheproposedmethodperfmswellitsMicroF1MacroF1arehigherthanthoseoftraditionalapproach.Thenduetothatthereisnotenoughfeaturesinfmationaboutthosepairstraditionalsingleclassifierisnotsatisfied.Therefewederived

7、sententialinfmationasfeatureofpairsfromcontextbesidestraditionalfeaturessuchaslexicalpartofspeechsemanticpositionalfeatures.Besidesofusingthesefeaturesofpairswebuiltanensembleclassifierwithweightedvotingmechanismtoidenti

8、fythesubjectiverelationshipbetweenfeatureopinionpairs.ExtensivestudiesoncpusofbookphonereviewsinChinesedemonstratethattheintroductionofsententialfeaturecouldimprovetherecallrateofclassifieraweightedensembleclassifieralso

9、couldachievethehighervalueofFmeasurewiththetradeoffbetweentheaccuracyrecallrateofsubclassifiers.Basedonabovemethodsaproductreviewminingsystemisconstructedtoanalysisthereviewwebpagereviewtext.Keywds:productreviewminingopi