您的浏览器禁用了JavaScript(一种计算机语言,用以实现您与网页的交互),请解除该禁用,或者联系我们。[CSET]:AI研究统计:探究英、中文献资料中的人工智能研究成果 - 发现报告
当前位置:首页/行业研究/报告详情/

AI研究统计:探究英、中文献资料中的人工智能研究成果

信息技术2022-08-15-CSET娇***
AI研究统计:探究英、中文献资料中的人工智能研究成果

DataBrief Counting AlResearch ExploringAlResearchOutputinEnglish-andChinese-LanguageSources Author DanielChou EMERGINGTECHNOLOGYJuly2022 Introduction U.S.policymakerscareaboutresearchoutput.Trackingresearchoutputcan,amongotherthings,informassessmentsofanygivencountry'sinnovativenessorassistin andotheremergingtechnologies,U.S.policymakersareeagertounderstandandcompareresearchtrends.Butmeasuringresearchoutputisnotasstraightforwardasitmayseem.Therearedifferentwaystodefineoutputanddifferentdatasourcesthat canbereliedon,whichcanaffecttheoutcomesofassessments. Onewayofmeasuringoutputisbycountingthenumberofpublicationsinaresearchareaasanindicatorofinterestorfocusinthatresearcharea.Capturingresearch measuringoutput.2ThisdatabrieftakestheformerapproachofcountingpublicationquantitiesasawaytoexploretrendsinAlresearch.3 WhencountingAlpublicationoutput,itiscriticaltoconsiderwherethepublicationsarepublishedandthelanguageinwhichaworkispublished.IndevelopingCSET'smergedcorpusofscholarlyliterature,weintentionallyincorporatedChinaNational KnowledgeInfrastructure(CNKl;中国知网),akeyChinese-languagedatasource,in DigitalScienceDimensions,MicrosoftAcademicGraph,arXiv,andPapersWithCode TheinclusionofCNKImeansthatcountingAlresearchatCSETmayproducedifferentresults,comparedtoanalyseswithoutCNKIl.WhiletheremaybetimeswhenitisappropriatetoseparateCNKIandpredominatelyEnglish-languagepublications,the abilitytoanalyzethemjointlyoffersabroaderviewoftheresearchlandscapeandanin-depthexplorationofChinese-languageAlresearch.ThisdatabriefexplorestheimplicationsofincludingCNKIbypresentingAlresearchpublicationtrendsinCSET's mergedcorpusincludingandexcludingCNKl. CenterforSecurityandEmergingTechnologyI1 TableofContents Introduction DataandMethodology 8 GlobalAlResearchOutput. AlResearchCollaboration AlResearchinChina,.11 Conclusion...15 Authors.16 Acknowledgments...16 AppendixA:DefinitionsandBaseline.17 AppendixB:CollaborationTrendsandPowerLaw.18 AppendixC:ExtendedVersionsofFiguresandTables22 AppendixD:AlKeywords.32 Endnotes36 CenterforSecurityandEmergingTechnology12 DataandMethodology Tocountartificialintelligence(Al)researchoutput,weexaminepublicationsinCSET's DigitalScienceDimensions,MicrosoftAcademicGraph,arXiv,PapersWithCode,andtheChinaNationalKnowledgeInfrastructure.*TheinclusionofCNKIinourmergedcorpusresultsin43millionChinese-languagepublications(17percentof publications),whereaswithoutCNKIweobserve4.2millionChinese-languagepublications(2percentofpublications).5 Weuse"publication"torefertoalldocumenttypesinthemergedcorpus,whichincludesjournalarticles,conferencepapers,bookchapters,andmore.ToidentifyAl publicationsinthemergedcorpus,weuseahuman-curatedlistofbilingualAlkeywords,Publicationswithakeywordmatchintheirtitle,abstract,or,whenavailable, full-textweretaggedasAl-relevant.ThisapproachdiffersfrompreviousCSETresearchthatidentifiesAlresearchusingamodeltrainedtopredictAl-relevanceforEnglish-languagepublications.7Weoptedtousethesamekeyword-basedsearchto identifyAlpublicationsinbothChineseandEnglishtoavoidusingdifferentmethodologiesfordifferentlanguages. Whenanalyzingpublications,weextractreportedauthororganizationaffiliations,the assignapublicationtoacountry,weusethecountryoftheaffiliatedorganizations.If thesameorganizationislistedmultipletimesonapublication,thenweassignthepapertothatorganizationonlyonce.Likewise,ifmultipleauthor-affiliated thatcountryonlyonce.SeeAppendixAformoredetailsandAppendixDforlistof keywords. CenterforSecurityandEmergingTechnologyI3 GlobalAlResearchOutput WefirstcomparetrendsinAlresearchoutputbyanalyzingthenumberandshareofAlpublicationsbycontributingcountry,thelocationoforganizationsproducingAlresearch,andtheorganizationsfundingAlresearch. ChinaandtheUnitedStatesweretheleadersinAlresearchoutputin2020byrawpublicationcounts,asseeninFigure1A.ThisholdswhenexaminingCSET'smergedcorpuswithandwithoutCNKl,thoughthenumberofAlpublicationswithChinese-affiliatedauthorsskyrocketswiththeinclusionofCNKl,jumpingfrom61,999to 254,098-—anadditional192,099AlpublicationscapturedbyCNKIasseeninFigure 1B.8 Figure1A:TopContributingCountriestoAlPublicationsin202O,withoutCNKI NumberofAIPublications 20,000 40,000 60.000 80,000 China 61.99 UnitedStates 57.121 India 20.508 UnitedKingdom 14.947 Germary 11.805 Japan 9.062 Canada 8.927 Austratia 8.351 Korea 8.167 Italy 6.943 France 6.876 Spain 5,496 Russia 4.906 Iran 4.511 Brazil4.321 Figure1B:TopContributingCountriestoAIPublicationsin2020,withCNKI Alpublicationsinmergedcorpus(withoutCNKl)AlpublicationsinCNKI 50.000100,000150.000200,000250,000 China61.999192.099 UnitedStates57.121 India UnitedKngdom Germany Source:CSETmergedcorpus. CenterforSecurityandEmergingTechnologyI4 Next,wepresenttheshareofAlpublicationsfromtheUnitedStates,China,andtherestoftheworld(ROW)from20102020inCSET'smergedcorpus,firstexcludingCNKI(Figure2A)andthenincludingCNKI(Figure2B).9 Figure2A:CountryShareofAlPublications20102020,withoutCNKI UnitedStates 80.0 60.0 ROW 40.0 20.0 China Figure2B:CountryShareofAlPublications20102020,inclu