《兰德-新闻报道中的坏角色-追踪国家行为者操纵新闻的行为（英）-2021.11-20正式版.docx》由会员分享，可在线阅读，更多相关《兰德-新闻报道中的坏角色-追踪国家行为者操纵新闻的行为（英）-2021.11-20正式版.docx（21页珍藏版）》请在课桌文档上搜索。
1、BADACTORSINNEWSREPORTINGTRACKINGNEWSMANIPULATIONBYSTATEACTORSChristianJohnsonWilliamMarcellinoheglobalspreadofthecoronavirusdisease2019(COVID-19)createdfertilegroundforattenptstoinfluenceanddestabilizedifferentpopulationsandcountries.Inresponsetothis.RANDCorporationresearchersconductedaproof-of-conc
2、eptstudyfordetectingtheseeffortsatscale.Marryingalarge-scalecollectionpipelineforglobalnewswithmachine-learninganddataanalysisworkflows,theRANDteamfoundthatbothRussiaandChinaappeartohaveemployedinformationmanipulationduringtheCOVID-19Pandemicinservicetotheirrespectiveglobalagendas.Thisreportisthesec
3、ondinaseriesoftworeports;thefirst(Matthews,Migacheva1andBrown,2021)focusedonqualitativeanddescriptiveanalysisofthesamedatareferredtointhisreport.Here,wedescribeOuranalyticworkflowsfordetectinganddocumentingstate-sponsoredmalignandsubversiveinformationefforts,andWereportquantitativeresultsthatSupport
4、thequalitativefindingsfromthefirstreport.IntroductionAspartofOuranalysis,wesearchedforbothdifferencesandsimilaritiesinthetopicsdiscussedbyRussian,Chinese,andWesternnewsmedia,andwefoundthatconspiracytheoriesandgeopoliticalposturingwererelativelycommoninRussianandChinesenewsarticlesComparedwithWestern
5、(U.S.andUK)articles.TheworkwedescribeherelaysthefoundationforarobustprotectivecapabilitythatdetectsandshedsIightonstateactorinformationmanipulationandmiscondctintheglobalarena.Disinformation,Propaganda,andTruthDecayTheworldisexperiencingacrisisrelatedtodisagreementsovertheestablishedtruth,aphenomeno
6、nthatRANDreferstoasTruthDecayashiftinpublicdiscourseawayfromfactsandanalysisthatiscausedbyfourinterrelateddrivers(RichandKavanagh,2018):1. anincreasingdisagreementaboutfactsandanalyticalinterpretationsoffactsanddata2. ablurringofthelinebetweenopinionandfact3. anincreasingrelativevolume,andresultingi
7、nfluence,ofopinionandpersonalexperienceoverfact4. adecliningtrustinformerlyrespectedsourcesoffactualinformation.RANDwww.rand.orgTruthDecayisaseriousthreattobothdomesticU.S.andinternationalsecurity,onethatisbeingexacerbatedbymaligneffortsfromavarietyofnationalbadactors.Theseill-intentionedeffortstomi
8、suseinformatiarelabeledmanywaysreadersmighthaveseentheseeffortslabeledasdisinformation,misinformation,fakenews,andinformationoperations.Forclarityandconsistencyssake,weusethedefinitionstakenfromRichandKavanagh,2018,intheremainderofthispaper.OurdefinitionofconspiracytheoriescomesfromDouglasetal.,2019
9、.(SeetheKeyInformationDefinitionsbox.)TRUTHE)E3 二：一二，KEYINFORMATIONDEFINITIONSDisinformationFalseormisleadinginformatiospreadintentionally,usuallytoachievesomeMisinformationpoliticaloreconomicobjective,infIuencepublicattitudes,orhidethetruth(asyonymforpropaganda)FalseormisleadinginformationthatisCon
10、spiracyspreadunintentionally.byerrorormistakeInformationthatattemptstoexplainthetheoriesultimatecausesofsignificantsocialFakenewsandpoliticaleventsandcircumstanceswithclaimsofsecretplotsbytwoormorepowerfulactorsNewspaperarticles,televisionsnewsshows,TopicDefinitionorotherInformationdisseminatedthrou
11、ghbroadcastorsocialmediathatareintentionallybasedonfasehoodsorthatintentionallyusemisleadingframingtoofferadistortednarrativeNewsManipulationfromBothChinaandRussiaWefoundthatduringtheCOVID-19pandemic,bothRussiaandChinaengagedinnewsmanipulationthatservedtheirgeopoliticalgoals.1AlthoughEnglish-languag
12、enewsmediafrombothnationsdidengageIntraditionalreportingonCOVID-19reportingoninfection,deathrates,andmedicalresponsesgloballyIheyalsoconducteddistinctmediaeffortsthatappeartobepolitiCallydrivennewsmanipulation.WefoundthatRussianmediaadvancedanti-U.S.conspiracytheoriesaboutthevirusandthatChinesemedia
13、advancedpro-ChinanewsthatlaunderedBeijingsreputationintermsofCOVID-19response.Additionally,wefoundthatearlyinthepandemic,RussianmediasupportedChinasefforttoburnishitsreputation.Intotal,threemainpillarsOfChineseandRussiannewsabouttheCOVID-19pandemicwereidentified.First,unsurprisingly,ChineseandRussia
14、nnewsagenciesreportedonstorieswithbroadinterestthatis,newstopicscoveredSimiIarIybyWesternnewsagencies.GoodexampiesofthispillararearticlesdescribingthecasenumbersanddeathsrelatedtoCOVID-19.ThesecondpillarofnewsstoriesconsistsOfarticlesthatperformgeopoliticalreputation-launderingonbehalfofRussiaandChi
15、na.ManyChinesenewsarticles,forearple,praiseChinashandlingofthepandemicandhighlightitsdonationsofaidtoforeigncountries.interestingly,RussiannewspraisesChinainasimilarway.RussiannewsalsoappearedtodownplaytheoriginalVID-19outbreakinWuhan.(Weconsidertheinteractionbetweenthesedifferentpillarslaterinthisr
16、eport.)Finally,RussianandChinesenewsagenciespromotedconspiracytheoriesregardingCOVID-19andthepublichealthmeasuresimplementedtocontainit.ExamplesofnewsinthispillararethesuggestionthatC0V!D-19isabioweaponorotherwiseengineeredinaIaboratoryortheideathatcontact-tracingeffortsarepartofaneffortbygovernment
17、andtechnologycompaniestotrackcitizens.TheSuccessofourproof-Of-Conceptstudysupportstheideathatexisting,off-the-shelfnaturallanguageprocessingmethodscanbeusedtomakesenseofnewsreportingbynation,ataglobalscale.Thesemethods,linkedtoascalableinfrastructurethatingestsnewsfromaroundtheworld,couldcreateaU.S.
18、-supportedcapabilitytodetectnewsmanipulationatthenation-statelevel.Inplaceofattemptstoidentifyindividualnewsstoriesorsourcesthatareunreliable,suchaCapabilitycouldmakemanipulationofthebroadernewslandscapepubliclyVisible.Automatedsummarizingofanation,snewsoutputatanaggregatelevelwouldquicklyuncoverama
19、nipulatkxieffortforexample,thespreadingOfaconspiracytheorythatcontact-tracingprogramsarepartofagovernmenttrackingeffort.(ThisisarealexamplethatRussiannewssourcesspreadandthatourmodeldetected). By news rranipulation. we mean that news articles were published to f urther the agenda of a state sponsor
20、rather than to inf Orm the public. These aices are therefore subject to pressures beyond the standard editorial control of a news agency.Wehaveseveralreasonsforchoosingtofocusouranalysisondataaggregatedatthenation-statelevel(asopposedto,forexample,theindividualnewsoutletlevel).Frst1weviewedthisstudy
21、asanextensionofpriorworklookingatnation-stateleveldisinformationefforts(Marcellino1Johnson,etal.,2020;Marcellino1Marcinek1etal.,2020).Thesepriorworkslookedatnation-stateactorsengagedinbroaddisinformationeffortstointerferewithelections,andwelookedSpecificallyatstatemanipulationofnewsmediaduringapande
22、mic.Second,keyfeaturesthatpresentthemselvesonlyatthenationallevelwereofinterest:Mostimportantly,theUnitedStatesandUnitedKingdomhaverobust,independentpresseswhileRussiaandChinaexertstatecontrolOvertheirnewsmedia.Aseparateandequallycompellinganalysiswouldexaminepotentialnewsdisinformationwithinnations
23、(forexampie,bypartisannewssourcesintheUnitedStates).Itislikelythatsuchananalysiswouldfindsignificantdifferencesbetweenindividualoutletsthatareworthexploring,especiallythroughthelensofpoliticalpolarizationintheUnitedStatespartisannewshaspreviousIybeenidentifiedasadriverofTruthDecay(RichandKavanagh,20
24、18).ApotentiallimitationofthisworkisthatWefoCUSedonbonEnglish-languagearticles.RussiaandChinaarenotmajorityEnglish-speaking,sowearecorparingnewsstoriesaimedatdomesticaudiences(U.S.andUK)withonesaimedatforeignaudiences(RussianandChinese),insofarasthenewsoutletsaretryingtoinfluenceEnglish-speakingpeop
25、le,however,wefeelthattheycanbeusefullycopared.Cross-Iinguisticcorparisonofdomesticallyorientedreportingisanotherpotentiallinetofutureresearch.GiventheeffectivenessOfcombiningexistingoff-the-shelfmethodsinourreport,apublicsystemformonitoringglobalnewsthatdetectsanddescribesglobalnewsthemesbynationisp
26、lausible.SuchasystemcouldhelpguardagainstTruthDecayeffortsfrommaliciousstateactors.Thesystemalsocouldanalyzeadditionalsourcesofdata,suchassocialmediaposts,tounderstandboththenarrativesbeingpushedandwhichonestakehold.Moreinsightcouldalsobegarneredbyperformingdeeperanalysisattheindividualnewsagencylev
27、el:DifferentonlinecommunitiesareIikelytoresponddifferentlytosimilarnewsstories,dependingonwhichsourcetheyoriginatefrom,forexample.MorediscussionofsuchanewsmonitoringsystemcanbefoundintheDiscussionsection.MethodologyIdentifyingdisinformationinalarge,complexdatasetisnotasimpletask.Theworddisinformatio
28、nisacatchalltermusedtorefertoanarrayofdifferentphenomena-fromfakenews,toopinionpiecesmasqueradingasjournalism,tolegitimatenewsstoriesthatheapinordinateattentiononcertaintopics(whileignoringothers).Asdescribedinthedefinitionsbox,disinformationisusedtorefertothedeliberatespreadingofmisleadingorincorre
29、ctinformation;misinformationreferstohonestbutincorrectknowledge.However,thelinebetweenthetwocanSometimesbeblurred;priorRANDwork(Marcellino1Johnson,etal.,2020)showedthatcoordinatedbotactivitywaslikelyuseddeliberatelyintherun-uptothe2020U.S.presidentialelectiontoamPlifyauthentictweetsandmakethemappear
30、morepoplarthantheyreallywere(Commonlycalledastroturfing)inanattempttocreateafalseimpressiOfgrassrootsspread.Ourgoal,therefore,wasnottodetectdisinformationperselbuttoidentifywhenandTheworddisinformationisacatchalltermusedtorefertoanarrayofdifferentphenomena一fromfakenewsJtoopinionpiecesmasqueradingasj
31、ournalism,tolegitimatenewsstoriesthatheapinordinateattentiononcertaintopics.TROTH E3 UxrhowRussianandChinesenewsmediaappeartobemanipulatedbyforcesoutsidethenormalnewscycleandeditorialprocesses.BecauseourdataSetfeaturedmanyarticlesfromavarietyofU.S.andUKmedia,wemakethekeyassumptionthatnewsworthystori
32、eswillbecoveredbytheseWesternoutlets;instancesinwhichRussianandChinesemediacoverstoriesthatarequalitativelydifferentfromthosecoveredbyWesternmediaareworthyofmoreScrutinytodeterminewhethertheycouldbepartofadisinformaticampaign.Computationaltechniqueshavepreviouslybeenusedbyresearcherstostudydissemina
33、tionoffakenews,ParticularlyonTwitter.Grinbergetal.,2019,demonstratedthatfakenewsinthelead-uptothe2016U.S.presidentialelectionwasseenandsharedprimarilybyarelativelysmallnumberOfTwitterusers,primarilyconsistingbothofhighlyconservativeandcyborgaccounts.2Usingasimilarmethodology,Lazeretal.,2020,foundtha
34、tthesameConclusionsessentiallyheldtrueforthespreadoffakenewsrelatedtoCOVID-19.MarceIIino1Johnson,etal.,2020,usedadifferentmethodologytodeterminethatbot-Iikeaccountslikelyplayedasignificantroleinspreadingfar-rightconspiracytheoriesanddisinformationleadinguptothe2020election.Inshort,theavailableresear
35、chSuggeststhatmuchofthedisinformationonsocialmediaisspreadbyarelativelysmallnumberofmalignusers.ThesestudieshavemostlyexaminedmetadataandderivedfeaturestodrawtheirconclusionsinsteadofstudyingthelanguageOfdisinformationitself.3Thispaperbuildsonexistingresearchtostudynotonlymetadataaboutnews,buttheact
36、ualcontentofthenewsitself.WehopedthatunderstandingthetopicalthemesbeingspreadviaClisinformationwouldleadtonewinsightsthatcannotbeseenSimplybylookingatuserengagementonsocialnetworks,suchasTwitter.ThefirstreportinthisseriesidentifiedseveralkeymarkersofdisinformationinRussianandCinesenews:conspiracythe
37、ories,geopoliticalposturing,andanti-U.S.messaging.Althoughwehopedthatadata-drivenapproachwouldreplicatethesefindings,wesoughttoperformOuranalysisasblindlyaspossible;thatis,wedidnotseektoconfirmorsuspicionsandSimplysearchthedatatofindConspiracytheoriesJnsteadjWeusedalgorithmStodetectthedominantthemes
38、inthedataandonlythenanalyzedthesethemestodeterminetheircontent.Ouroverallstrategy,asmentionedearlier,restedontheideathatanydisinformationpublishedbyRussianandChinesenewssourceswouldbedetectablebecauseitscontentwoulddiffermeaningfullyfromthecontentinU.S.andUKnewsarticles.Certainly,somedifferencesinco
39、ntentaretobeexpectedunderano-manipulationhypothesis:Forexample,RussiannewssourcesmightbemorelikelytocoverstoriesaboutEasternEuropethannewSfromtheUnitedStates,Sirrplybecauseofgeographicalproximity.However,wehypothesizedthatbyinspectingthesedifferencesclosely,wewouldbeabletouncoverpatternsassociatedwi
40、thmanipulation.Ultimately,anydifferencesbetweenWesternandnon-Westernnewsarticleswoldalsorequirehumananalysistodeterminewhetherthedifferenceswereinnocuousormalign.DataDescriptionWeusedNewsAPItocollectallEnglish-Ianguagearticlesfrom43newssources(nineofwhichareRussianJiveChinese,27U.S.,andtwoUK)forthep
41、eriodJanuary1,2020,throughAugust31,2020,thatfeaturedeithercoronavirusor,COVlDinthetet.4Thisresultedinatotalof247,315articles,theVastmajorityofthem(230,865)fromU.S./UKsources,withSmallernumbersfromRussian(14,309)andChinese(2,141)sources.(WeprovideamoredetailedbreakdownOfarticlespublishedbynewsoutleti
42、ntheAppendix.)Foroursearchperiod,theoverallfrequencyofpublishedarticleswitheithertermmentionedgrewrapidlythroughJanuaryandFebruary,reachingapeakinMarchandApril.ArticlefrequencybyCountryoforiginisshownovertimeinFigure1.AsimilarpatternwasSeeninpublishingIrequencyovertimeacrossU.S./UK,Russian,andChines
43、esources,althoughRussiannewssourcesappearedtopublishsomewhatlessIrequentlyinmidtolateFebruary.More-detailedanalysisofthisapparentRussianslowdownisdescribedlaterinthisreport.2 Acybogaccountisonethatmixesautomatedbotactivitywithrealhumantweets.3 DerivedfeaturesreferstosuchthingsasthepresenceoffaknewsU
44、RLsinaTwitterfeed.*NewsAPIisanapplicationprogramminginterfacethatallowsuserstoautomaticallyconnecttoandsearchalargedatabaseofnewsarticles,includingnewswireservices(animportantadvantageoversuchrivalsourcesasLexisNexis),RANDhasbuiltascalableinfrastrcturetoretrieve,store,query,andthenanalyzevylargenews
45、articledatasets.Thisscalablearchitectureisapowerfultoolthatallowsustogatheranenormousamountofnewsdataforanass.butitalsohasaconstraint:Wecancollectonlynewsarticlesfromsourcescoveredbytheservice,whichdoesnotincludesourcesthatarebehindpaywallsorotherwiserestrictedinaccess.Forourstudy,inpaicularlonlynin
46、eRussianandfiveChinesesourcesinEnglisharecoveredbyNewsAPI.FIGURE1ArticIeFrequencyoverTimein2020U.S./UKnewsRussiannews,ChinesenewsNOTE：Themovingsevendayaveragepublishingrateisoverlaidoneachsourceasasolidline.Notethatthey-axisislogarithmicallyscaled;wehaveaboutanorderofmagnitudefewerRussiannewsarticle
47、sthanU.S./UK.andaboutanorderofmagnitudefewerChinesearticlesthanRussian.AepedS3orvBecausetheCOVID-19pandemicwassuchanimpactfulworldwideevent,wewerenotsurprisedtofindthatnewsstoriesaboutmanyothertopics,suchasthosethatweredominantlyabouteconomicorpoliticalStories1WereaIsorepresentedinordatasetbecauseth
48、eyalsoreferencedthepandemicinsomeway.However,acursoryexaminationofrandomarticlesinourdatasetshowedthatthemajoritywerefocusedonadifferent(nonpandemic)topic,althoughthepandemicplayedasignificantroleinmanyofthesearticles.WedecidedtomodelhowthisassortmentofdifferentsubjectsvariedacrossRussian,Chinese,andWesternnewsmedia.IfwecoulddeterminethatcertaintopicswerebeingdiscussedquiteoftenbyRussianorChinesenewsbutrarelybyWesternoutlets,thatWouldsuggesttheneedforadditionalexaminationandmightevenbeindicativeofamalignefforttopushcertainnarratives.Naturallanguageprocessing,thebranchofmachinelearningthatdea
