“hg3088培训课件壹word避免费下载“

> 国际 > 作者:locoy 2019-08-29 13:05 编辑:周建平
0

  hg3088培训课件壹龙星方案课程:信息检索StatisticalLanguageModelsforIROutlineMoreaboutstatisticallanguagemodelsingeneralSystematicreviewoflanguagemodelsforIRThebasiclanguagemodelingapproachAdvancedlanguagemodelsKLdivergenceretrievalmodelandfeedbackLanguagemodelsforspecialretrievaltasksMoreaboutstatisticallanguagemodelsingeneralWhatisaStatisticalLMAprobabilitydistributionoverwordsequencesp“TodayisWednesday”p“TodayWednesdayis”p“Theeigenvalueispositive”Contexttopicdependent!Canalsoberegardedasaprobabilisticmechanismfor“generating”text,thusalsocalleda“generative”modelWhyisaLMUsefulProvidesaprincipledwaytoquantifytheuncertaintiesassociatedwithnaturallanguageAllowsustoanswerquestionslike:Giventhatwesee“John”and“feels”,howlikelywillwesee“happy”asopposedto“habit”asthenextwordspeechrecognitionGiventhatweobserve“baseball”threetimesand“game”onceinanewsarticle,howlikelyisitabout“sports”textcategorization,informationretrievalGiventhatauserisinterestedinsportsnews,howlikelywouldtheuseruse“baseball”inaqueryinformaodelorrelevancemodelandthecollectionLMAnenrichedqueryLMcanbeestimatedbyexploitingpseudofeedbackeg,relevancemodelCorrelationbetweentheclarityscoresandretrievalperformanceisfoundExpertFindingBalogetal,FangZhaiTask:GivenatopicT,alistofcandidatesCi,andacollectionofsupportdocumentsSDi,rankthecandidatesaccordingtothelikelihoodthatacandidateCisanexpertonTRetrievalanalogy:QuerytopicTDocumentCandidateCRankaccordingtoPR|T,CSimilarderivationstothoseonslides,canbemadeCandidategenerationmodel:Topicgenerationmodel:SummarySLMsvsTraditionalIRPros:StatisticalfoundationsbetterparametersettingMoreprincipledwayofhandlingtermweightingMorepowerfulformodelingsubtopics,passages,LeverageLMsdevelopedinrelatedareasEmpiricallyaseffectiveaswelltunedtraditionalmodelswithpotentialforautomaticparametertuningCons:LackofdiscriminationacommonproblemwithgenerativemodelsLessrobustinsomecaseseg,whenqueriesaresemistructuredComputationallycomplexEmpirically,performanceappearstobeinferiortowelltunedfullfledgedtraditionalmethodsatleast,noevidenceforbeatingthemChallengesandFutureDirectionsChallenge:EstablisharobustandeffectiveLMthatOptimizesretrievalparametersautomaticallyPerformsaswellasorbetterthanwelltunedtraditionalretrievalmethodswithpseudofeedbackIsasefficientastraditionalretrievalmethodsChallenge:DemonstrateconsistentandsubstantialimprovementbygoingbeyondunigramLMsModellimiteddependencybetweentermsDerivemoreprincipledweightingmethodsforphrasesChallengesandFutureDirectionscontChallenge:DevelopLMsthatcansupport“lifetimelearning”DevelopLMsthatcanimproveaccuracyforacurrentquerythroughlearningfrompastrelevancejudgmentsSupportcollaborativeinformationretrievalChallenge:DevelopLMsthatcanmodeldocumentstructuresandsubtopicsRecognizequeryspecificboundariesofrelevantpassagesPassagebasedsubtopicbasedfeedbackCombinedifferentstructuralcomponentsofadocumentChallengesandFutureDirectionscontChallenge:DevelopLMstosupportpersonalizedsearchInferandtrackauser’sinterestswithLMsIncorporateuser’spreferencesandsearchcontextinretrievalCustomizeorganizesearchresultsaccordingtouser’sinterestsChallenge:GeneralizeLMstohandlerelationaldataDevelopLMsforsemistructureddataeg,XMLDevelopLMstohandlestructuredqueriesDevelopLMsforkeywordsearchinrelationaldatabasesChallengesandFutureDirectionscontChallenge:DevelopLMsforhypertextretrievalCombineLMswithlinkinformationModelingandexploitinganchortextDevelopaunifiedLMforhypertextsearchChallenge:DevelopLMsforretrievalwithcomplexinformationneeds,eg,SubtopicretrievalReadabilityconstrainedretrievalEntityretrievalegexpertsearchWhatYouShouldKnowGeneralpictureoflanguagemodelsforIRTheKLdivergenceretrievalformulaasageneralizationofthequerylikelihoodmethodHowthemixturemodelforfeedbackworksKnowhowtoestimatethesimplemixturemodelusingEMReferencesAgichteinCucerzanEAgichteinandSCucerzan,Predictingaccuracyofextractinginformationfromunstructuredtextcollections,ProceedingsofACMCIKMpagesBaezaYatesRibeiroNetoRBaezaYatesandBRibeiroNeto,ModernInformationRetrieval,AddisonWesley,BaietalJingBai,DaweiSong,PeterBruza,JianYunNie,GuihongCao,Queryexpansionusingtermrelationshipsinlanguagemodelsforinformationretrieval,ProceedingsofACMCIKM,pagesBalogetalKBalog,LAzzopardi,MdeRijke,Formalmodelsforexpertfindinginenterprisecorpora,ProceedingsofACMSIGIR,pagesBergerLaffertyABergerandJLaffertyInformationretrievalasstatisticaltranslationProceedingsoftheACMSIGIR,pagesBergerABergerStatisticalmachinelearningforinformationretrievalPhDdissertation,CarnegieMellonUniversity,BleietalDBlei,ANg,andMJordanLatentdirichletallocationInTGDietterich,SBecker,andZGhahramani,editors,AdvancesinNeuralInformationProcessingSystems,Cambridge,MA,MITPressCaoetalGuihongCao,JianYunNie,JingBai,Integratingwordrelationshipsintolanguagemodels,ProceedingsofACMSIGIR,Pages:CarbonellandGoldsteinJCarbonellandJGoldstein,TheuseofMMR,diversitybasedrerankingforreorderingdocumentsandproducingsummariesInProceedingsofSIGIR',pagesChenGoodmanSFChenandJTGoodmanAnempiricalstudyofsmoothingtechniquesforlanguagemodelingTechnicalReportTR,HarvardUniversityCollinsThompsonCallanKCollinsThompsonandJCallan,Queryexpansingusingrandomwalkmodels,ProceedingsofACMCIKM,pagesCronenTownsendetalSteveCronenTownsend,YunZhou,andWBruceCroftPredictingqueryperformanceInProceedingsoftheACMConferenceonResearchinInformationRetrievalSIGIR,CroftLaffertyWBCroftandJLaffertyed,LanguageModelingandInformationRetrievalKluwerAcademicPublishersFangetalHFang,TTaoandCZhai,Aformalstudyofinformationretrievalheuristics,ProceedingsofACMSIGIRpagesReferencescontFangZhaiHFangandCZhai,Probabilisticmodelsforexpertfinding,ProceedingsofECIRFoxEFoxExpendingtheBooleanandVectorSpaceModelsofInformationRetrievalwithPNormQueriesandMultipleConceptTypesPhDthesis,CornellUniversityFuhrNFuhrLanguagemodelsanduncertaininferenceininformationretrievalInProceedingsoftheLanguageModelingandIRworkshop,pagesGaoetalJGao,JNie,GWu,andGCao,Dependencelanguagemodelforinformationretrieval,InProceedingsofACMSIGIRGoodIJGoodThepopulationfrequenciesofspeciesandtheestimationofpopulationparametersBiometrika,and:,GreiffMorganWGreiffandWMorgan,ContributionsofLanguageModelingtotheTheoryandPracticeofIR,InWBCroftandJLaffertyeds,LanguageModelingforInformationRetrieval,KluwerAcademicPubGrossmanFriederDGrossmanandOFrieder,InformationRetrieval:AlgorithmsandHeuristics,ndEd,Springer,HeOunisBenHeandIadhOunis,AstudyoftheDirichletpriorsfortermfrequencynormalisation,ProceedingsofACMSIGIR,PagesHiemstraKraaijDHiemstraandWKraaij,TwentyOneatTREC:AdhocandCrosslanguagetrack,InProceedingsoftheSeventhTextREtrievalConferenceTREC,HiemstraDHiemstraUsingLanguageModelsforInformationRetrievalPhDdissertation,UniversityofTwente,Enschede,TheNetherlands,JanuaryHiemstraDHiemstraTermspecificsmoothingforthelanguagemodelingapproachtoinformationretrieval:theimportanceofaquerytermInProceedingsofACMSIGIR,HiemstraetalDHiemstra,SRobertson,andHZaragozaParsimoniouslanguagemodelsforinformationretrieval,InProceedingsofACMSIGIRHofmannTHofmannProbabilisticlatentsemanticindexingInProceedingsonthendannualinternationalACMSIGIR,pagesJarvelinKekalainenCumulatedgainbasedevaluationofIRtechniques,ACMTOIS,Vol,No,,JelinekFJelinek,StatisticalMethodsforSpeechRecognition,Cambirdge:MITPress,JelinekMercerFJelinekandRLMercerInterpolatedestimationofmarkovsourceparametersfromsparsedataInESGelsemaandLNKanal,editors,PatternRecognitioninPracticeAmsterdam,NorthHolland,ReferencescontJeonetalJJeon,VLavrenkoandRManmatha,AutomaticImageAnnotationandRetrievalusingCrossmediaRelevanceModels,InProceedingsofACMSIGIRJinetalRJin,AHauptmann,andCZhai,Titlelanguagemodelsforinformationretrieval,InProceedingsofACMSIGIRKaltTKaltAnewprobabilisticmodeloftextclassicationandretrievalUniversityofMassachusettsTechnicalreportTR,KatzSMKatzEstimationofprobabilitiesfromsparsedataforthelanguagemodelcomponentofaspeechrecognizerIEEETransactionsonAcoustics,SpeechandSignalProcessing,volumeASSP:KraaijetalWKraaij,TWesterveld,DHiemstra:TheImportanceofPriorProbabilitiesforEntryPageSearchProceedingsofSIGIR,ppKraaijWKraaijVariationsonLanguageModelingforInformationRetrieval,PhDthesis,UniversityofTwente,,KurlandLeeOKurlandandLLeeCorpusstructure,languagemodels,andadhocinformationretrievalInProceedingsofACMSIGIRKurlandetalOrenKurland,LillianLee,CarmelDomshlak,Betterthantherealthing:iterativepseudoqueryprocessingusingclusterbasedlanguagemodels,ProceedingsofACMSIGIRpagesKurlandLeeOrenKurlandandLillianLee,PageRankwithouthyperlinks:structuralrerankingusinglinksinducedbylanguagemodels,ProceedingsofACMSIGIRpagesLaffertyandZhaiaJLaffertyandCZhai,ProbabilisticIRmodelsbasedonqueryanddocumentgenerationInProceedingsoftheLanguageModelingandIRworkshop,pagesLaffertyZhaibJLaffertyandCZhaiDocumentlanguagemodels,querymodels,andriskminimizationforinformationretrievalInProceedingsoftheACMSIGIR,pagesLavrenkoCroftVLavrenkoandWBCroftRelevancebasedlanguagemodelsInProceedingsoftheACMSIGIR,pagesLavrenkoetalVLavrenko,MChoquette,andWCroftCrosslingualrelevancemodelsInProceedingsofSIGIR,pagesLavrenkoVLavrenko,AgenerativetheoryofrelevancePhDthesis,UniversityofMassachusettsLiCroftXLi,andWBCroft,TimeBasedLanguageModels,InProceedingsofCIKM',LiuCroftXLiuandWBCroftPassageretrievalbasedonlanguagemodelsInProceedingsofCIKM,pagesReferencescontLiuCroftXLiuandWBCroftClusterbasedretrievalusinglanguagemodelsInProceedingsofACMSIGIRMacKayPetoDMacKayandLPetoAhierarchicalDirichletlanguagemodelNaturalLanguageEngineering,:MaronKuhnsMEMaronandJLKuhns,Onrelevance,probabilisticindexingandinformationretrievalJournaloftheACM,:McCallumNigamAMcCallumandKNigamAcomparisonofeventmodelsforNaveBayestextclassificationInAAAILearningforTextCategorizationWorkshop,pagesMilleretalDRHMiller,TLeek,andRMSchwartzAhiddenMarkovmodelinformationretrievalsystemInProceedingsofACMSIGIR,pagesMinkaLaffertyTMinkaandJLafferty,Expectationpropagationforthegenerativeaspectmodel,InProceedingsoftheUAI,pagesNallanatiAllanRameshNallapatiandJamesAllan,CapturingtermdependenciesusingalanguagemodelbasedonsentencetreesInProceedingsofCIKMNallanatietalRNallanati,WBCroft,andJAllan,Relevantqueryfeedbackinstatisticallanguagemodeling,InProceedingsofCIKMNeyetalHNey,UEssen,andRKneserOnStructuringProbabilisticDependenciesinStochasticLanguageModelingComputSpeechandLang,,NgKNgAimumlikelihoodratioinformationretrievalmodelInVoorhees,EandHarman,D,editors,ProceedingsoftheEighthTextREtrievalConferenceTREC,pagesOgilvieCallanPOgilvieandJCallanCombiningDocumentRepresentationsforKnownItemSearchInProceedingsofthethAnnualInternationalACMSIGIRConferenceonResearchandDevelopmentinInformationRetrievalSIGIR,ppPonteCroftJMPonteandWBCroftAlanguagemodelingapproachtoinformationretrievalInProceedingsofACMSIGIR,pagesPonteJMPonteAlanguagemodelingapproachtoinformationretrievalPhddissertation,UniversityofMassachusets,Amherst,MA,SeptemberReferencescontPonteJPonteIsinformationretrievalanythingmorethansmoothingInProceedingsoftheWorkshoponLanguageModelingandInformationRetrieval,pages,RobertsonSparchJonesSRobertsonandKSparckJonesRelevanceWeightingofSearchTermsJASIS,,RobertsonSERobertsonTheprobabilityrankingprincipleinIRJournalofDocumentation,:,RobertsonWalkerSERobertsonandSWalker,SomesimpleeffectiveapproximationstothePoissonmodelforprobabilisticweightedretrievalProceedingsofACMSIGIRpagesRosenfeldRRosenfeld,Twodecadesofstatisticallanguagemodeling:wheredowegofromhereInProceedingsofIEEE,volume~SaltonetalGSalton,AWongandCSYang,AvectorspacemodelforautomaticindexingCommunicationsoftheACM,:SaltonBuckleyGSaltonandCBuckley,Termweightingapproachesinautomatictextretrieval,InformationProcessingandManagement,,ShannonShannon,CEAmathematicaltheoryofcommunicationBellSystemTechJ,,ShenetalXShen,BTan,andCZhaiContextsensitiveinformationretrievalwithimplicitfeedbackInProceedingsofACMSIGIRSietalLSi,RJin,JCallanandPlOgilvieALanguageModelFrameworkforResourceSelectionandResultsMergingInProceedingsofthethInternationalConferenceonInformationandKnowledgeManagementCIKMSinghaletalASinghal,CBuckley,andMMitra,Pivoteddocumentlengthnormalization,ProceedingsofACMSIGIRSinghalASinghal,ModernInformationRetrieval:ABriefOverviewAmitSinghalInIEEEDataEngineeringBulletin,pages,SongCroftFSongandWBCroftAgenerallanguagemodelforinformationretrievalInProceedingsofEighthInternationalConferenceonInformationandKnowledgeManagementCIKMSparckJonesKSparckJones,AstatisticalinterpretationoftermspecificityanditsapplicationinretrievalJournalofDocumentation,,and,,ReferencescontSparckJonesetalKSparckJones,SWalker,andSERobertson,Aprobabilisticmodelofinformationretrieval:developmentandcomparativeexperimentspartandpartInformationProcessingandManagement,:andSparckJonesetalKSparckJones,SRobertson,DHiemstra,HZaragoza,LanguageModelingandRelevance,InWBCroftandJLaffertyeds,LanguageModelingforInformationRetrieval,KluwerAcademicPubSrikanthSrihariMSrikanth,RKSrihariExploitingSyntacticStructureofQueriesinaLanguageModelingApproachtoIRinProceedingsofConferenceonInformationandKnowledgeManagementCIKM'SrikanthMSrikanthExploitingqueryfeaturesinlanguagemodelingapproachforinformationretrievalPhDdissertation,StateUniversityofNewYorkatBuffalo,TanetalBinTan,XuehuaShen,andChengXiangZhai,,Mininglongtermsearchhistorytoimprovesearchaccuracy,ProceedingsofACMKDDTaoetalTaoTao,XuanhuiWang,QiaozhuMei,andChengXiangZhai,Languagemodelinformationretrievalwithdocumentexpansion,ProceedingsofHLTNAACLTaoZhaiTaoTaoandChengXiangZhai,RegularizedestimationofmixturemodelsforrobustpseudorelevancefeedbackProceedingsofACMSIGIRTurtleCroftHTurtleandWBCroft,EvaluationofaninferencenetworkbasedretrievalmodelACMTransactionsonInformationSystems,:vanRijsbergenCJvanRijsbergenAnonclassicallogicforinformationretrievalTheComputerJournal,WittenetalIHWitten,AMo#at,andTCBellManagingGigabytesCompressingandIndexingDocumentsandImagesAcademicPress,SanDiego,ndedition,WongYaoSKMWongandYYYao,AprobabilitydistributionmodelforinformationretrievalInformationProcessingandManagement,:WongYaoSKMWongandYYYaoOnmodelinginformationretrievalwithprobabilisticinferenceACMTransactionsonInformationSystems,:XuCroftJXuandWBCroftClusterbasedlanguagemodelsfordistributedretrievalInProceedingsoftheACMSIGIR,pages,XuetalJXu,RWeischedel,andCNguyenEvaluatingaprobabilisticmodelforcrosslingualinformationretrievalInProceedingsoftheACMSIGIR,pagesReferencescontZaragozaetalHugoZaragoza,DHiemstraandMTipping,BayesianextensiontothelanguagemodelforadhocinformationretrievalInProceedingsofSIGIR:ZhaiLaffertyaCZhaiandJLaffertyAstudyofsmoothingmethodsforlanguagemodelsappliedtoadhocinformationretrievalInProceedingsoftheACMSIGIR,pagesZhaiLaffertybCZhaiandJLaffertyModelbasedfeedbackinthelanguagemodelingapproachtoinformationretrieval,InProceedingsoftheTenthInternationalConferenceonInformationandKnowledgeManagementCIKMZhaiLaffertyCZhaiandJLaffertyTwostagelanguagemodelsforinformationretrievalInProceedingsoftheACMSIGIR,pagesZhaietalCZhai,WCohen,andJLafferty,BeyondIndependentRelevance:MethodsandEvaluationMetricsforSubtopicRetrieval,InProceedingsofACMSIGIRZhaiLaffertyCZhaiandJLafferty,Ariskminimizationframeworkforinformationretrieval,InformationProcessingandManagement,,Jan,pagesZhaiCZhai,LanguageModelingandRiskMinimizationinTextRetrieval,PhDthesis,CarnegieMellonUniversity,ZhaiLaffertyCZhaiandJLafferty,Ariskminimizationframeworkforinformationretrieval,InformationProcessingandManagement,,Jan,pagesZhangetalYZhang,JCallan,andThomasPMinka,NoveltyandredundancydetectioninadaptivefilteringInProceedingsofSIGIR,RoadmapThislecture:systematicreviewoflanguagemodelsforIRNextlecture:formalretrievalframeworksWhatWeHaveAchievedSoFarFrameworkandjustificationforusingLMsforIRSeveraleffectivemodelsaredevelopedBasicLMwithDirichletpriorsmoothingisareasonablebaselineBasicLMwithinformativepriorsoftenimprovesperformanceTranslationmodelhandlespolysemysynonymsRelevancemodelincorporatesLMsintotheclassicprobabilisticIRmodelKLdivergencemodeltiesfeedbackwithquerymodelestimationMixturemodelscanmodelredundancyandsubtopicsCompletelyautomatictuningofparametersispossibleLMscanbeappliedtovirtuallyanyretrievaltaskwithgreatpotentialformodelingcomplexIRproblemsCanLMsconsistentlyconvincinglyoutperformtraditionalmethodswithoutsacrificingefficiencyCanwedomuchbetterbygoingbeyondunigramLMsHowcanwelearneffectivelyfrompastrelevancejudgmentsHowcanwebreakthedocumentunitinaprincipledwayHowcanweexploituserinformationandsearchcontexttoimprovesearchWhatrolecanLMsplaywhencombiningtextwithrelationaldataHowcanweexploitLMstodevelopmodelsforcomplexretrievaltasksHowcanwedevelopaneffectiveunifiedretrievalmodelforWebsearchPw|dwPw|dwloglikelihoodimumLikelihoodEstimatorNewton’sMethodLeaveoneoutwwPwn|dwnwnabcabcabcddabccdddabdababababcddecdewordbyauthorwordbyauthorabcabcabcddabecbefacffbefaffabefcdcdbgefsSupposewekeepsamplingandgetmorewordsWhichauthorislikelyto“write”morenewwordsNow,supposeweleave“e”out?mustbebig!moresmoothingdoesn’thavetobebigTheamountofsmoothingiscloselyrelatedtotheunderlyingvocabularysizeQueryQq?qmNimumLikelihoodEstimatorExpectationimizationEMalgorithmPw|ddPw|dNdN?Stagepw|dpw|Upw|dNpw|UStageEstimatedinstageAverageprecisionDB’squerytypes,topics*IndicatessignificantdifferenceCompletelyautomatictuningofparametersISPOSSIBLE!BasictranslationmodelTranslationmodelRegulardocLMHowlikelydocDbelongstoclusterCOnlyeffectivewheninterpolatedwiththebasicLMscoresLikelihoodofQgivenCClassicProbModelQuerylikelihood“LanguageModel”ReldocmodelNonReldocmodel“Relquery”modelPD|Q,RPD|Q,RPQ|D,Rq,d,q,d,q,d,q,d,q,d,q,d,q,d,q,d,q,d,q,d,ParameterEstimationInitialretrieval:queryasreldocvsdocasrelqueryPQ|D,RismoreaccurateFeedback:PD|Q,RcanbeimprovedforthecurrentqueryandfuturedocPQ|D,Rcanalsobeimproved,butforcurrentdocandfuturequeryDocbasedfeedbackQuerybasedfeedbackBothmethodsoutperformthebasicLMsignificantlyOriginalformulainLavrankoCroftqueryentropyignoredforrankingQueryQDocumentDResultsFeedbackDocsFd,d,?,dnGenerativemodelNofeedbackFullfeedbackwwFd,?,dnimumLikelihoodPw|Pw|CPsourceBackgroundwordsTopicwordsNoiseinfeedbackdocumentstheaweto?textmining?KnownBackgroundpw|C?textminingassociationword?Unknownquerytopicpw|F“Textmining”ObservedDocsSuppose,weknowtheidentityofeachwordMLEstimatorIdentity“hidden”variable:zibackground,topicthepaperpresentsatextminingalgorithmthepaperziSupposetheparametersareallknown,what’sareasonableguessofzidependsonwhydependsonpw|Candpw|FhowEstepInitially,setpw|Ftosomerandomvalue,theniterate?MstepAssumeExpectationStep:AugmentingdatabyguessinghiddenvariablesimizationStepWiththe“augmenteddata”,estimateparametersusingimumlikelihoodTrectopic:“airportsecurity”MixturemodelapproachWebdatabaseTopdocsTranslationmodels,Relevancemodels,andFeedbackbasedquerymodelshaveallbeenshowntoimproveperformancesignificantlyoverthesimpleLMsParametertuningisnecessaryinmanycases,butseeTaoZhaifor“parameterfree”pseudofeedbackEnglishChineseEnglishChinesewordMethod:Method:TranslationmodelEstimatewithabilinguallexiconOrParallelcorporaEstimatewithparallelcorporaTitleAbstractBodyPartBodyPart?DDDDDkWanttocombinedifferentpartsofadocumentwithappropriateweightsAnchortextcanbetreatedasa“part”ofadocumentApplicabletoXMLretrieval“partselection”probServesasweightforDjCanbetrainedusingEMSelectDjandgenerateaquerywordusingDjRefinementofthismodelleadstospecificretrievalformulasSimplemodelsoftenendupinterpolatingmanyunigramlanguagemodelsbasedondifferentsourcesofevidence,eg,shorttermsearchhistoryShenetalorlongtermsearchhistoryTanetalContextindependentQueryLM:ContextsensitiveQueryLM:imumLikelihoodestimatorEMalgorithmReferenceLMLMforDMeasureofredundancyChengXiangZhai翟成祥DepartmentofComputerScienceGraduateSchoolofLibraryInformationScienceInstituteforGenomicBiology,StatisticsUniversityofIllinois,UrbanaChampaignczhaiSourceTransmitterencoderDestinationReceiverdecoderNoisyChannelPXPY|XXYX’PX|YWhenXistext,pXisalanguagemodelBayesRuleManyExamples:Speechrecognition:XWordsequenceYSpeechsignalMachinetranslation:XEnglishsentenceYChinesesentenceOCRErrorCorrection:XCorrectwordYErroneouswordInformationRetrieval:XDocumentYQuerySummarization:XSummaryYDocumentUnigramLanguageModelpw|?textminingassocationclustering?food?Topic:Textmining?foodnutritionhealthydiet?Topic:HealthDocumentdTextminingpaperFoodnutritionpaperSamplingGiven,pd|variesaccordingtodUnigramLanguageModelpw|Documenttex

您觉得这篇文章: 不错0 一般0

hg3088皇冠版权及免责声明:

  • ①凡本网注明来源:“hg3088皇冠”或“hg3088”的所有作品,版权均属于hg3088皇冠或hg3088,未经hg3088皇冠或hg3088协议授权不得转载、链接、转帖或以其他方式发表,已经hg3088皇冠或hg3088授权的媒体、网站,在使用时必须注明来源“hg3088皇冠”或“hg3088”。
  • ②如本网转载稿涉及版权等问题,请作者在来电或来函与hg3088皇冠联系,我们将及时处理解决。联系方式:365jilin@365jilin.com