面向金融行业的领域词汇关联-研究

《面向金融行业的领域词汇关联-研究》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。

1、万方数据面向金融行业的领域词汇关联研究ABSTRACTFinancialindustryhasgreatdemandoninformationacquisitionandprocessing．AlthoughInteractactsasahugefinancialinformationplatform，theexplosivevolumealsobringsgreatchallengesforpeopletOuseit．Tosolvethisproblem，manytechnologiesappear,suchastextmining，amongwhichwordre

2、lationminingisanimportanttask．However,existingmethodsheavilyrelyonmanualworkandlackofbenchmarkdatasetforcomparisonandvalidation．Besides．mostofcurrentworkiScarriedonspecificdomainandthelanguageisonlyforEnglish．Thispaperhascarriedoutaresearchonwordrelationminingforfinancialdomain，whichmai

3、nlycoverstwokindsoftasks．Onetaskisbringingoutanovelmeasureofrelationshipbetweensecurityentities；theotheroneistodetecttherelationshipbetweenfinancialterminologies．Firstly,fromtheperspectiveofthetopicsemanticsandwordCO—occurrence，weestablishedamodelformeasuringthecorrelationbetweensecurit

4、yentitiesusingtopicmodelandcosineSimilarity．Theexperimentresultshowedsomeusefulcorrelationbetweensecurityentities．Secondly,wefoundthatsomeconceptsofacorpuswillsplitwhilechangingthecountoftopicsintopicmodel。Accordingly,weproposedallautomatichierarchicalontologylearningalgorithrn．Totestth

5、eproposedalgorithm，webroughtoutanadaptivequeryexpansionmethodbasedonitandquerylogs．TheexperimentresultprovedtheproposedmethodwasbetterthanothermethodwhichWaSsuitableforthesituationthatislackofdomalnontology．Themainworkofthispaperincludes：1)weproposeanentitymeasuremodelbasedontopiccorrel

6、ationandCO-occurrence．ThismodelevaluatesthecorrelationfromtheperspectiveoftopicsemanticandwordCO-occurrenceanddiscoveriesthehiddencorrelationsbetweenentitiesandmeasuresthemthroughunsupervisedmethod；2)weapplythismodeltosecuritydomainbycomparisonwiththepriceserieswhichisobjectivestandard；

7、3)weproposeanautomaticdomain-specifichierarchicalontologylearningmethodbasedondomaincorpus．4)weproposeanadaptivehybridqueryexpansionmodelincludingdomainontologyandquerylogs，whichcanenhancetheexistingdomainontologythroughquerylogs；5)weprovideanapplicationsampleoftopicmodelbasedo