信息抽取中的文本挖掘.pdf

《信息抽取中的文本挖掘.pdf》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、MultilingualismandElectronicLanguageManagement:Proceedingsofthe4thInternationalMIDPColloquium,September2003,Bloemfontein,SouthAfrica,Daelemans,W.,duPlessis,T.,Snyman,C.andTeck,L.(Eds.)pp.141-160,VanSchaikPub.,SouthAfrica,2005TextMiningwithInformationExtractionRaymondJ.MooneyandUnYongNahmDepartmen

2、tofComputerSciences,UniversityofTexas,Austin,TX78712-1188mooney,pebronia
@cs.utexas.eduAbstractTextminingconcernslookingforpatternsinunstructuredtext.TherelatedtaskofInforma-tionExtraction(IE)isaboutlocatingspecificitemsinnatural-languagedocuments.Thispaperpresentsaframeworkfortextmining,calledDI

3、SCOTEX(DiscoveryfromTextEXtraction),usingalearnedinformationextractionsystemtotransformtextintomorestructureddatawhichisthenminedforinterestingrelationships.TheinitialversionofDISCOTEXintegratesanIEmoduleacquiredbyanIElearningsystem,andastandardruleinductionmodule.Inaddition,rulesminedfromadataba

4、seextractedfromacorpusoftextsareusedtopredictadditionalinformationtoextractfromfuturedocuments,therebyimprovingtherecalloftheunderlyingextractionsystem.EncouragingresultsarepresentedonapplyingthesetechniquestoacorpusofcomputerjobannouncementpostingsfromanInternetnewsgroup.1IntroductionTheproblemo

5、ftextmining,i.e.discoveringusefulknowledgefromunstructuredorsemi-structuredtext,isattractingincreasingattention[4,18,19,21,22,27].ThispapersuggestsanewframeworkfortextminingbasedontheintegrationofInformationExtraction(IE)andKnowledgeDiscov-eryfromDatabases(KDD),a.k.a.datamining.KDDandIEarebothtop

6、icsofsignificantrecentinterest.KDDconsiderstheapplicationofstatisticalandmachine-learningmethodstodiscovernovelrelationshipsinlargerelationaldatabases.IEconcernslocatingspecificpiecesofdatainnatural-languagedocuments,therebyextractingstructuredinformationfromfreetext.However,therehasbeenlittleifany

7、researchexploringtheinteractionbetweenthesetwoimportantareas.Inthispaper,weexplorethemutualbenefitthattheintegrationofIEandKDDfortextminingcanprovide.Traditionaldataminingassumesthattheinformationtobe“mined”isalreadyint