基于知识推理的自然语言数据探索关键技术研究.pdf

ABSTRACT

ABSTRACT

Withtheincreasinglyscaleofdataaccumulatedbyhumansociety,thecostofdata

analysisandprocessingarealsoincreasing.Whetherknowledgecanbeefficientlyob­

tainedfrommassivedatahasbecomeoneofthekeyfactorsaffectingtheefficiencyof

dataanalysis.Dataexplorationreferstotheprocessofacquiringknowledgefromsource

data;dataexplorationbasedonnaturallanguagecanprovideanaturallanguageinterface

fordata,lowerthecostfordataanalysis,andimprovetheefficiencyofdataanalysis.

Knowledgereasoningcanminethepotentialdatasemanticsinexistingknowledge,thus

complementingtheknowledgebaseandknowledgegraph.

Aimingattheproblemofincompletemetadatainnaturallanguagedataexploration

tasks,thisthesisdesignsacompositemetadata,andaccordinglydesignsanaturallanguage

dataexplorationmethodbasedonknowledgereasoningtechnology.Themainworkisas

follows:

(1)Anewcompositemetadataisproposed,andanewconstructionmethodofre­

lationaldatabasecompositemetadatabasedonknowledgereasoningtechnologyisde­

signed.Basedonthemulti­sourceknowledgesuchastheschemaoftherelationaldatabase,

thesimplemetadataoftherelationaldatabaseisfirstdefined.Afterbuildingknowledge

graphsbasedonsimplemetadata,aggregatethegraphstoformamorecompleteknowl­

edgegraph.Thentheknowledgereasoningisperformedtocompletetheknowledgegraph,

soastofullyminetherichdatasemanticsinthegraph.Atlast,thefinalgraphservesasthe

compositemetadata.Theexperimentalresultsshowthatthroughknowledgereasoning,

thefinalgraphincludesmoreknowledge,andthedatasemanticscontainedinthemetadata

ismorecomplete.Thismethodcandealwiththeproblemofincompletemetadata.

(2)Adeeplearningmodelbasedongraphcomputing,featureenhancementandsome

othermethodsisproposedtoaccomplishanaturallanguagedataexplorationtaskforr