欢迎来到天天文库
浏览记录
ID:34620492
大小:184.66 KB
页数:11页
时间:2019-03-08
《relevance feedback and inference networksnew》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、RelevanceFeedbackandInferenceNetworksDavidHainesandW.BruceCroftDepartmentofComputerScienceUniversityofMassachusettsAmherstMA01003internet:fhaines,croftg@cs.umass.eduAbstractTheinferencenetmodelrecentlydescribedbyTurtleandCroft[19,20]hasbeenshowntobeaneectiveandRelevancefeedback,whichmo
2、diesqueriesus-generalbasisforaninformationretrievalsystem.Oneofingjudgementsoftherelevanceofafew,thepurposesofthispaperistoshowhowfeedbacktech-highly-rankeddocuments,hashistoricallybeenniquescanbeusedwiththismodel.Thisincludesbothanimportantmethodforincreasingtheper-simpletechniques,asd
3、escribedbySaltonandBuckleyformanceofinformationretrievalsystems.In[15],andtechniquesthatexploittheabilityoftheinfer-thispaper,weextendtheinferencenetworkencenetmodeltorepresentstructureinthequery.modelintroducedbyTurtleandCrofttoin-Theothermajortopicaddressedinthispaperisthecluderelevanc
4、efeedbacktechniques.Thedif-eectoffulltextcollectionsonrelevancefeedbacktech-ferencebetweenrelevancefeedbackontextab-niques.Virtuallyallofthepreviousrelevancefeed-stractsandfulltextcollectionsisstudied.Pre-backexperimentshavebeendoneusingcollectionsofliminaryresultsforrelevancefeedbackon
5、thedocumentabstracts.Fulltextcollectionsarebecom-structuredqueriessupportedbytheinferenceingincreasinglyimportant,andthereisthepossibilitynetmodelarealsoreported.thattheincreasedamountoftextintheidentiedrel-evantdocumentswillmaketheselectionandweightingoftermsmoredicult.1Introduction1.
6、1PriorWorkRelevancefeedbackmethodsininformationretrievalat-tempttoimproveperformanceforaparticularquerybyWorkonrelevancefeedbackmethodsininformationre-modifyingthequery,basedontheuser'sreactiontotrievalhasalonghistory[16,4].Rocchio[13]describestheinitialretrieveddocuments.Specically,the
7、user'sanelegantapproachtorelevancefeedbackinthevectorjudgementsoftherelevanceornon-relevanceofsomeofspacemodel.Heshowshowtheoptimalvectorspacethedocumentsretrievedareusedtoaddnewtermstoquerycanbederivedusingvectoradditionandsub-thequeryandtoreweightqueryterms.Forexample,i
此文档下载收益归作者所有