Page 122 - 2023年第54卷第9期
P. 122
guagemodelprompts[C]??Proceedingsofthe2022CHIConferenceonHumanFactorsinComputingSystems.
2022.
[21] ARORAS,NARAYANA,CHENM F,etal.AskMeAnything:Asimplestrategyforpromptinglanguagemod
els[EB?OL].(2022 - 11 - 20)[2023 - 01 - 23].http:??arxiv.org?abs?2210.02441.
[22] 王一钒,李博,史话,等.古汉语实体关系联合抽取的标注方法[J].数据分析与知识发现,2021,5(9):
63 - 74.
[23] SUNY,WANGS,LIY,etal.Ernie:Enhancedrepresentationthroughknowledgeintegration[EB?OL].(2019 -
04 - 19 )[2022 - 12 - 17].http:??arxiv.org?abs?1904.09223.
[24] RADFORDA,WUJ,CHILDR,etal.Languagemodelsareunsupervisedmultitasklearners[J].OpenAIblog,
2019,1(8):9.
[25] BROWNT,MANNB,RYDERN,etal.Languagemodelsarefew - shotlearners[J].Advancesinneuralinfor
mationprocessingsystems,2020,33:1877 - 1901.
[26] HUANGQ,LIAOD,XINGZ,etal.SEFactualKnowledgeinFrozenGiantCodeModel:AStudyonFQNand
itsRetrieval [EB?OL].(2022 - 12 - 16)[2023 - 02 - 27].http:??arxiv.org?abs?2212.08221.
[27] GAOT,FISCHA,CHEND.Makingpre - trainedlanguagemodelsbetterfew - shotlearners[EB?OL].(2021 - 06 -
02 )[2022 - 12 - 04].http:??arxiv.org?abs?2012.15723.
[28] MISHRAS,KHASHABID,BARALC,etal.ReframingInstructionalPromptstoGPTk’sLanguage[EB?OL].
(2022 - 03 - 15)[2022 - 12 - 16].http:??arxiv.org?abs?2109.07830.
[29] WANGX,WEIJ,SCHUURMANSD,etal.Self - consistencyimproveschainofthoughtreasoninginlanguage
models [EB?OL].(2023 - 03 - 07)[2023 - 03 - 15].http:??arxiv.org?abs?2203.11171.
[30] HUANGC,LIC,YAOS,etal.AHybridschemeforParsingCantoneseTextBasedonPyCantonesePlusand
PyLTP[C]??2022EuropeanConferenceonNaturalLanguageProcessingandInformationRetrieval(ECNLPIR).
IEEE ,2022.
[31] SCHEAFFERRL,MENDENHALLIIIW,OTTR L,etal.Elementarysurveysampling[M].CengageLearning,
2011.
ApplicationofAIchainintegratingcrowdintelligencestrategyin
floodcontrolknowledgeinference
1
1,2
3
1
1
YANGYangrui,ZHUYaping,CHENSisi,LIUXuemei ,LIHuimin
(1.SchoolofInformationEngineering,NorthChinaUniversityofWaterResourcesandElectricPower,Zhengzhou 450000,China;
2.CollaborativeInnovationCenterforEfficientUtilizationofWaterResources,Zhengzhou 450000,China;
3.SchoolofWaterConservancy,NorthChinaUniversityofWaterResourcesandElectricPower,Zhengzhou 450000,China)
Abstract:Floodcontrolknowledge(entitiesandrelationships)isanessentialcomponentofthefloodcontroland
rescuebusinessknowledgegraph.Thecomplexrelationshipsamongfloodcontrolentitiesaredistributedinunstruc
turedtext,andthescarcityandlowqualityofavailabletextsposechallengestoknowledgeextractioninthisfield.
Inresponse ,anovelapproachisproposedtouselargelanguagemodelstocompletefloodcontrolandrescueknowl
edgeinference(FCRKI).BasedonLLMs,threesub - modulesaredesigned:floodcontrolentityextractor,flood
controlentityknowledgeparser ,andfloodcontrolinter - entityrelationshipdecider.A seriesofeffectivetask
promptsaredesignedandconnectedtoform anAIchain.Thefloodcontrolknowledgeinferencetaskisgradually
completedthroughreal - timeinteractionbetweenthepromptsandLLMsintheAIchain.Furthermore ,aswarmin
telligencestrategyisdesignedtoenhancethereliabilityoffloodcontrolinter - entityrelationshipinference.Compa
ringFCRKIwithexistingmethods,experimentalresultsshowthatFCRKIhasahigheraccuracyininferringrelation
shipsbetweenfloodcontrolentities,verifyingtheeffectivenessoftheAIchainandswarmintelligencestrategy.This
newknowledgeextractionparadigmprovidesanovelsolutionforintelligentprocessingofhydraulicengineeringtexts.
Keywords:floodcontrolknowledgeinference;floodcontrolknowledgegraph;crowdintelligencestrategy;AIchain
(责任编辑:于福亮)
— 1 1 2 —
3