DOI: 10.4018/IJACDT.2019010101

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

Copyright©2019,IGIGlobal.CopyingordistributinginprintorelectronicformswithoutwrittenpermissionofIGIGlobalisprohibited.

1

Data-Driven Maps of Art HistoryDoron Goldfarb, Vienna University of Technology, Vienna, Austria

Dieter Merkl, Vienna University of Technology, Vienna, Austria

ABSTRACT

Many approaches to explain conceptions of developments in the arts to both peers and a moregeneralaudienceincludeddiagramsrepresentingnetworksofrelatedentitiessuchasgenealogiesofimportanthistoricalactorsorstyles.Whilesuchvisualizationsweretraditionallycreatedbyhand,therecentemergenceofextensivedigitalrepositoriesofarthistoryinformationenablenewmeansofpresentation.Thisworkseekstoexplorethepotentialofopenlyavailabledatasourcestocreatebottom-up,datadrivenversionsofsuchnetworkmapsofarthistory.Ithighlightscommonalitiesanddifferencesbetweenviewsderivedfrominstitutionalandcrowd-sourceddatarepositoriesandcomparesthemwithidentifiedhistoricalexamples.Theresultssuggestthattheavailabledatacanbeusedtocreatelargescaleviewsonspecificdevelopmentsinarthistory,potentiallyservingasaidfornavigatingvastonlinecollectionsofdigitizedartworksbutalsoasmeansofreflectionontheoriginofthedatasourcesthemselves.

KeywoRDSArt Styles, Biographies, Core-Periphery, Data Analytics, DBpedia, Genealogy, Network Visualization, ULAN, Wikidata, Wikipedia

INTRoDUCTIoN

Especiallyaftertheinceptionofarthistoryasanacademicdisciplineinthelate18thcentury,diagramshavebeenusedasvisualdevicestoconveyviewsonrelateddevelopments.Theywereincludedinpublicationsandtextbooksandservedascommunicationalmeanstosupportscholarlypositionsandaseducationaltoolsprovidingbird’seyeviewsonlarge-scalehistoricaldevelopments.Famousexamplessuchasthe“DiagramofStylisticEvolutionfrom1890until1935”from(Barr,1986/1937)soughttopresentvisualaggregationsofmultitudesofindividualprocessesasmacroscopic,law-likeviewsonartisticevolution.Asdiscussedin(Schmidt-Burkhardt,2005),suchchartswereoftenpraisedfortheireducationalvaluebutalsochallengedforlackingobjectivityduetoomissionsorspecialfociintroducedbytheirauthors.

Inrecentyears,theincreasingavailabilityofbothdigitizedandborn-digitalarthistoryresourceswentinparallelwiththeemergenceofnewparadigmsofdata-drivenanalysiswhichcanbesubsumedundertheumbrellaterm“dataanalytics”.Theemergingfieldofdigitalarthistoryincreasinglyusesalgorithmicmethodstoanalyzetheaccumulatingbodyofhistoricalmaterial,whilerecenttrendsinmuseologyproposedigitalmeansofpresentingartcollectionstoanincreasinglytech-savvyaudience,

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

2

seekingtocontextualizeexhibitsusingmultimediaguidesandinteractiveinstallationsforlocalanddedicated online presentations for remote visitors. Information visualization plays an importantroleinbothscenarios,supportingresearchersintheinterpretationoflargedatacollectionsandthecommunicationof their findings,but alsoprovidingvirtualvisitorswithnavigationalguides forfindingtheirwaythroughvastonlineartcollections.

Theoriginalaimof thisworkwas toexplore thepotentialofhistoricalsocialnetworkdata,representedthroughinterlinkedartistbiographies,forprovidingbird’seyeoverviewsonarthistoricaldevelopmentsbeyondtherelativelylimitedscopeofexisting,oftenmanuallycreatedmapsofarthistory.Thiswasdrivenby the expectation that large-enoughamountsof recorded tiesbetweenartistsand/orotherimportantpersonsfromtheartworldwouldaggregateintolarge-scalenetworksspanningmultiple centurieswhosevisualizationcould support the contextualizationof artworksinvirtualpresentations.Theanalysisofexistingbiographicaldatasources,however,revealedthattheywerenotfreeofcultural/institutionalbiaseswhichappearedinformofdifferingcompositionsofnationalitiesand/orrolesofcoveredpersons.Thisprovidedanadditionalandrelevantresearchperspectivewhichdirectlylinkedtoexistingdiscussionsaboutmodesofinclusionandexclusionintheformationofthecanonofarthistory,globalaspectsofwhichwereoftensubsumedunderthenotionofthecoreandtheperipheryoftheartworld(Joyeux-Prunel,2014).

Theaimsoftheworkpresentedinthisarticleweretherefore(1)todemonstratethepossibilityofcreatingdata-drivenmapsofarthistoryand(2)toshowtheiruseforrevealingspecificbiasespresentindifferentdatacollections.Thispaperunifiestheauthors’previousrelatedwork(Goldfarb,Arends,Froschauer,&Merkl,2013)(Goldfarb,Arends,Froschauer,Weingartner,&Merkl,2014)underacommondataprocessingframeworkanddiscusseslarge-scalenetworkvisualizationsofarthistorybiographiesgeneratedwithcontentfromtheGettyUnionArtistNamesandWikipedia,comparingthemwitheachotherandwithexistingscholarlyexamples.Differentfilteringapproachesareusedtohighlightdataspecificaspects,includingmeanstounravelchronologicalstructureembeddedinhighlyinterlinkedsetsofhistoricalentitiesandtorevealhiddeninteractionsbetweensubgroupsofthem.Theresultsshowthatdata-drivenmapsofarthistorysuccessfullyextendtheirmanuallycreatedscholarlycounterpartsbyputtingtheminlargerhistoricalcontextandatthesametimeserveastooltoreflectuponthenatureoftheuseddatasourcesthemselves.

DATA SoURCeS AND TooLS

TheGettyUnionListofArtistNames(ULAN)isanestablisheddatasourcecuratedbydomainexpertsascontrolledvocabularyforpersonnamesrelevantforarthistory.Asdescribedin(Baca&Gill,2015),ithasrecentlybeenmadeavailableasLinkedData,reflectingthecurrenttrendforculturalinstitutionstoopentheirdatabasesforpublicaccessandre-use.Besidescontainingrecordsforabout200,000personsandinstitutions,theformerwithbasicinformationsuchasnationality,role,birthanddeathdates,etc.,theULANalsoprovidesassociativelinksbetweenasubsetofitsrecords,representing identifiedprofessional, familialandother relationship types.Due to its institutionalbackground,thepresenceofpersonsintheULANcanbeconsideredasevidencefortheirimportanceforthedomainofarthistory.

Asoneofthepremierexamplesforfreelyaccessible,crowd-curatedcontentontheInternet,Wikipediainturnunitesawiderangeofcontributionsfrombothdomainexpertsandlaypersons,mainlyrelyingonmutualreviewsforensuringthequalityofprovidedcontent.Beinganencyclopedia,its articles are usually dedicated to distinct entities such as persons, places or concepts and aremutuallyconnectedviahyperlinks.ThisallowstotreatinterlinkedbiographiesonWikipedialikeasocialnetworkandtocompareitwithmoreexplicitnetworkstructuressuchasfoundintheULAN.SeveralprojectsandinitiativesmakeWikipediadataaccessibleforquantitativeuse:Preprocessedandyearlyupdatedpage-linkdatasetsextractedfromvariousWikipedialanguageversionsareprovidedbyDBpedia,anapproachtoextractstructuredinformationfromWikipages(Aueretal.,2007).The

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

3

WikidataplatforminturnfeaturesstructuredrecordswithcommunitycontributedfactsaboutentitiespresentinWikipediaandincludesmanypersonrecordsinterlinkedwiththeirrespectiveWikipediabiographiesinmultiplelanguages(Vrandečić&Krötzsch,2014).DuetocontinuouseffortsoftheWikipediacommunity toattachauthoritycontrol information toWikipediaarticles,described in(Klein&Kyrios,2013),manyofthesepersonrecordscontainULANmappingswhichcanbeusedasreferencetoidentifyWikipediabiographiesrelevanttoarthistory.

In thiswork,Wikidata,DBpedia and theULANwereused toderive three relateddatasets,including (1) the network of ULAN biographies connected via associative ULAN links, (2) thenetwork of multilingual Wikipedia biographies mapped to the ULAN via a dedicated Wikidataproperty(ID: P245)andinterlinkedviamultilingualWikipediahyperlinksderivedfromDBpediaand(3)thebipartitenetworkofDBpedialinksbetweenmultilingualarticlesaboutart/architecturestylesandWikipediabiographies,allidentifiedviaWikidataentitiestaggedasartmovement(ID: Q968159),architecturalstyle(ID: Q32880)orhuman(ID: Q5),thelatternotlimitedtothosemappedtotheULAN.Datasets(1)and(2)wereusedforthepersonnetworkexperimentdescribedbelow,followedbytheperson-stylenetworkexperimentusingdataset(3).

AllthepresentednetworkvisualizationsarebasedontheForceAtlas2algorithmdescribedin(Jacomy,Venturini,Heymann,&Bastian,2014)asimplementedintheGephiplatform.Thislayoutalgorithmturnedouttobemostsuitedforvisualizingnetworkswithintrinsicchronologicalfeatures.TheRstatisticsplatformwasusedfordataprocessingandanalysis.

PeRSoN NeTwoRKS IN THe ULAN AND IN wIKIPeDIA

ULAN Person NetworkTheULANdatasetofsummer2015connected21,942personrecordsvia50,076uniquebutmirrored–e.g.allstudent_oflinkshadcomplementaryteacher_oflinks–associativelinks,correspondingto25,038uniquepairsofinterlinkedULANrecords.Having4,083connectedcomponents(CC)intotal,thenetworkfeaturedagiantconnectedcomponent(GCC)consistingof10,444(47.58%)personsinterlinkedvia29,956(59.82%)links.Sincethenextlargestcomponentonlyfeatured39persons,theGCCwasthefocusoftheanalysisanditsvisualizationisshowninFigure1.Thepersonsfeaturedinthechartwerefoundtofollowaroughchronologicalsuccessionfromlefttoright,withRenaissanceandBaroqueartistsdominatingtheleftside,late19thand20thcenturypersonstheright.Thecoloralphabetproposedin(Green-Armytage,2010)wasusedforcolor-codingthepersonnodesbythetop26(plusonefor“other”)preferrednationalitiesassignedintheULAN.Thisevenmoreunderscoredthechronologicalnatureofthenetworkbyrevealingclear-cutnationalclustersofpersons,especiallyvisiblefor the three largestnationalitiesrepresentingabout50.5%of thepersons in theGCC—Italian(14.33%),French(22.71%)andAmerican(13.47%).Theirapparentchronologicalsuccessionreflectedshiftingculturalcentersoftheartworldthroughtime.Additionaldevelopmentsappearedaboveandbelowthismainstrand,suchchainsoflooselyconnectedgroupsofDutch(6.26%)andFlemish(3.16%)personsontop,manyofthemfromtheDutchGoldenAge,andasimilarlyloosegroupofBritish(8.56%)personsatthebottom.Anotherlargegroupofpersons,Germans(8.35%),wasspreadacrossthewholeplotwithoutformingacontiguoussection.Manydenselyconnected,clique-likesubgroupsofpersonswerefoundtorepresentfamilies.

ThedenselydocumentedrelationshipsforItalian,FrenchandAmericanpersonsstoodincontrasttotherather“skinny”networksrecordedforothernationalities,whichsuggestedthatthesegroupsofpeoplewerethefocusofthecollectionspresentintheinstitutionscontributingtotheULAN,anissuetobetakenintoaccountwhenconsideringthesedataforcreatingalarge-scaleviewonarthistoricaldevelopmentsovercenturies.FundamentalarthistoricalepochssuchasItalianRenaissance,DutchGoldenAge,FrenchAcademism,etc.,neverthelessbecameclearlyvisibleinthevisualizationoftheULANnetwork,whichwasfoundtobeininterestingagreementwiththe“ChartofArtHistory”producedbyNewtonin1941forthetextbook”EuropeanPaintingandSculpture”(Newton,1956/1941,

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

4

p.237).QuiteliketheULANnetwork,themanuallycreatedchartfeaturedthecenturiesbetween1255to1900tobedominatedbyItalianandFrenchartistswhichoutnumberedtherepresentativesofall theotherfeaturedEuropeanschools.Overall, thevisualizationoftheULANGCCshowedthatbiographicalnetworkdatacouldinprinciplebeusedtoprovidelarge-scaleviewsonhistoricaldevelopmentsandtheULAN’sapparentfocusonspecificnationalitiesprovidedastrongincentivetoexploreadditionalsources,suchasWikipedia,forcomparison.

Person Networks in wikipediaAsofMarch2016, 47,463out of 55,122Wikidataperson recordswithULANmappingshad acorrespondingWikipediaarticleinatleastonelanguageand40,943ofthemwereconnectedvia283,368uniqueWikipediahyperlinksasidentifiedinoneormoreofthe50largestDBpedialanguageversions.IncontrasttotheULAN,Wikipediahyperlinksbetweenpersonbiographieswerenotalwaysmirrored, the identified283,368hyperlinks represented237,139uniquepairsofULANmappedbiographies.Althoughstillverysparse,themultilingualnetworkofWikipediahyperlinksthushadhigherdensitythantheULANassociativelinknetwork(5.79vs.1.14averagelinksperperson,Linkdensity0.000283vs0.000104)anditsGCCcontained99.08%ofthe40,943involvedpersons.12,239(55,78%)oftheULANpersonrecordsinterconnectedviaassociativelinksintheULANwerefoundtohavecorrespondingWikipediaarticlesmutually interlinked inat leastoneof the50observedWikipedialanguageversionsand80.72%oftheexistingULANassociativelinksbetweenthemwerealsorepresentedasWikipediahyperlinks.

TheoverlapbetweentheULANandtheWikipedianetworksuggestedthatanintrinsicchronologyshould also exist in the latter, but an initial visualization exposed no visible chronology there,promptingforamorethoroughexploration.InordertoanalyzethetemporaldistributionofWikipediahyperlinksbetweenthemappedULANpersons,thedistributionsofthesourceandtargetbirthdates

Figure 1. Visualization of the ULAN network by nationality, colored by nationality

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

5

ofthelinkedpairswereanalyzedandplottedona2DhistogramshowninFigure2,revealingvarioustemporalcharacteristicsofthestructureoftheWikipediahyperlinknetwork.Generallyincreasinginnumbersacrosstimeandinparallelwiththeoveralldistributionofpersonsbirthdatespeakingataround1850,about85%ofallthehyperlinksconnectedpersonsbornnotmorethan75yearsapart.Moreover,abouttwothirdsoftheWikipediahyperlinkspointedfrombiographiesofyoungerpersonstothoseofolderpersons.Onereasonforthistemporalasymmetrywerearticlesaboutoutstandingartistsandotherimportantpersonswhichwerefarmoreoftenreferencedbybiographiesoflatergenerationsthanviceversa.Somebiographies,e.g.articlesaboutimportantteachers,neverthelessfeaturedrelativelylonglistsoflinkstotheirpupils,whichinturncontributedtothestillextensivenumberofhyperlinkspointingintothefuture.Finally,therewasanapparentdiscontinuitybetweengroupsofpersonsfromGreekandRomanantiquity,visibleastwodistinctblobsat thediagonalcenteredaroundtheyears500BCand0AD,andthe”main”groupofpersonsfromthelatemiddleagesuntiltoday.Separatedbyaclearlyvisiblegapduringthemiddleages,theironlyconnectionvialinksoflongtemporaldistancereflectedtherenewedinterestinthecultureandtheartsofantiquityattheonsetofthemodernage.

Concerningtheidentifiedtemporalcharacteristics,themaindifferencebetweentheULANandWikipedianetworkswerethattheformerconsistedofsymmetriclinksmainlybetweencontemporaries(95.55%bornwithin75,98%within100yearsofeachother),whichresultedinpersonsfromantiquitytobedisconnectedfromtheULANGCC.Thissuggestedthatusingacutoffof75yearstoremove“longduration” links fromtheWikipedianetworkcoulduncover the temporalsuccessionof thefeaturedbiographiesviatheforce-basedlayout,althoughresultinginasimilar“lostconnection”toantiquity.ThiswasconfirmedinFigure3whichshowsthevisualizationoftheGCCofthefilterednetworkcombinedfromall50DBpedialanguageversions,featuringonlypersonsbornafter1000AD.Usingthesamecolor-codingfornationsasusedinFigure1fortheULANGCCrevealedasimilarchronologicalsuccessionofdistinctarthistoricalperiods.

SincetheselectionofWikipediabiographieswasbasedonpersonsfeaturedintheULAN,itwasnotsurprisingtofindasimilarsuccessionofItalian(green),French(red)andAmerican(blue)clustersinthefilteredWikipedianetwork,whichneverthelessalsocontaineddenseclustersofothernationalities.ManyDutch/Flemishpersons(light/darkorange)forexamplewereconcentratedina

Figure 2. 2D histogram of birth dates of interlinked Wikipedia biographies

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

6

“DutchGoldenAg”cluster,whileBritishpersons(lightblue)appearedinaparalleldevelopmenttothecentralFrench“strand”.ClustersofothernationalitiesinturnoverlappedintheupperpartoftheFiguretoahighdegree.

While the visualization of the combined Wikipedia biography network showed a relativelybalancedpresenceof thedifferentnationalities,mostof the50 individual contributing languageversions had a clear tendency to focus on culturally related content. In order to highlight thisobservation,Figure4showseightindividualnetworksrepresentingthesevenlargestcontributinglanguageversions(English,German,French,Italian,Spanish,RussianandDutch)andtheJapaneseversionaslargestnon-Europeanlanguage,allcreatedusingthesamelayoutasusedinFigure3forbettercomparison.TheseparatedviewsemphasizedthateachWikipediaversionfeaturedespeciallydensepersonclustersforthosenationalitiesrelatedtoitslanguage,whichwasoftenduetouniquepersonbiographiesnotpresentelsewhereandsometimesrevealedremarkabledevelopments.ThedistinctJapanesestrand(turquoise)shownatthebottomoftheJapaneseWikipedianetworkvisualizationforexamplehadonlyminimalconnectionstootherculturesuntilthelate19thcenturyMeijiera,whoseopeningtowardstheWestbecamedramaticallyvisibleintheplot.

Regardlessoftheemphasisputontheirowncultures,theobservedlanguageversionsalsofeaturedmanysimilarities.Mostofthemsharedbiographiesaboutthemostfamousrepresentativesofarthistorywhosemutualtiesformedarecurring”skeleton”foundinallthedifferentlanguageversionsandalsointheULANassociativelinknetwork,spanningfromtheRenaissance(Giotto,DaVinci,Raphael,Dürer,..)andtheBaroque(Caravaggio,Rembrandt,Rubens,..)acrossthe18thand19thcenturies(Goya,David,Delacroix,Courbet,Monet, ..) tomodern(Cézanne,Matisse,Kandinsky,Picasso,Dalí,..),postwar(Pollock,Warhol,..)andcontemporaryart(Koons,Hirst,Banksy,..).Interestingvisualanalogieswerefoundintwodifferenthistoricaldiagrams:SimilartothevisualizationsofthenationalclustersinthefilteredWikipedianetworks,aconcurrentpresenceofinterconnectedculturalstrandswasfoundin(Strass,1828/1804),aremarkablemanuallycreatedvisualizationoftheinterplaybetweendifferentworldculturesacrosshistoryasparallel,mergingandbifurcatingrivers.AlfredH.Barr’scuratorial“torpedo”diagramsfrom(Barr,1941),sketchingthecompositionoftheidealMoMAcollection,representedanotherstrikinganalogy:ThefilteredWikipediabiographynetworkvisualizationsshowedsimilarsnapshotsoftheinterplayofchangingcentersofculturalhegemony

Figure 3. Pruned network of Wikipedia biographies identified via the ULAN in 50 language versions, colored by nationality

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

7

acrosstime,butthistimenotasdeterminedbyasinglearthistorianbutviathecollectionofULANbiographiesfilteredbylanguage-specificcoveragesofdifferentWikipediaversions.

Network of Aggregated Links Between NationalitiesThevisualizationsofthenetworkofULANbiographiesinWikipediahighlightedthechronologicalinteractionsbetweengroupsofpersonsofdifferentnationalities.Thehighnumberofhomogeneousrelationshipsbetweenpersonsbelongingtooneofthefewtopcountingnationalities(the“core”),however,oftenmaskedtheinteractionsbetweenpersonsoflessprominentorigin(the“periphery”).Oneattempt tohighlight interactions inorwith theperipherywas tomove from the individual-basedmicro-leveltoamacro-levelnetworkofinteractingnationalities,derivedfromaggregatingnation-to-nationinteractionsfromtheindividuallevel.Thismacro-levelnetworkconsistedofnodesrepresentingnationalities,allpairsofwhichconnectedvialinksweightedbytheabsolutecountsofindividual-levellinksbetweenpersonsoftherespectivenationalities.Theobservedcounts,however,werestillhighlyskewedtowardsthecorenationsandinteractionsintheperipherythusassignedverylowoverallweights,callingforarelativemeasureforweightingtheaggregatednation-to-nationtiestakingthesizeofeachnationalityintoaccount.

Anexistingapproachcouldbeidentifiedforthispurpose,basedon“standardizedresiduals”asdescribedin(Agresti,2007,p.38)inthecontextofPearson’sChi-Squaredtestfortheindependence

Figure 4. Pruned biography networks as present in eight individual Wikipedia languages, colored by nationality

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

8

oftwocategoricalvariables.The34,710possiblepairsof195sourceand178targetnationalitiesoftheULANmappedandinterconnectedWikipediabiographieswerethusarrangedinacontingencytableholdingthehyperlinkcountsbetweeneachnationpairasobservedinthecombinedWikipedialanguagenetwork.Inordertoreduceeffectsfromnoise,countsbelowfiveweresettozerobeforecalculatingthestandardizedresidualsasdescribedin(Agresti,2007).Theresultswerefilteredforaminimumofthreepositivestandardizeddeviationsfromexpectation,yieldinganetworkof371interactionsbetween128distinctnationalities.Figure5showsavisualizationofthisnetworkwherethethicknessofeachtierepresentsthevalueofitsstandardizedresidual,nowrevealinginterestingconnectionsinandwiththeperiphery.Therewasanapparent,clusteredstructureofdifferentnationalitygroupseachhavinghighlyinterconnectedmembers.

The clustering could be confirmed by applying the Louvain community detection methoddescribedin(Blondel,Guillaume,Lambiotte,&Lefebvre,2008)whichidentified14highlyseparatedgroupswithahighmodularityvalue,asdefinedin(M.E.J.Newman,2006),of0.736.TheidentifiedgroupsareshownindifferentcolorsinFigure5anditwasinterestingtoobservethatinmanycases,theywerebasedonformercolonialties,especiallyvisiblefortheFrench(bottomcenter,purple),Spanish(lowerright,blue)andBritish(topcenter,black)spheresofinfluence.Othergroupswerebasedonchangingstateorganizationsthroughouthistory,suchasthegroupofformerYugoslaviancountries (lowercenter, lightblue), theRussian/Soviet sphereof influence (lowercenter,green)ortheAustrian/GermangroupincludingformermembersoftheAustro-Hungarianempire(lowerleftcenter,orange).Someoftheclustersalsolinkedancientandmiddleage“nationalities”totheirmoderncounterparts,suchasthosecontainingItaliansandGreeks(topleft,red)orIraniansandIndians(topcenter,lightgreen).

Althoughclearlydistinctfromeachother,thedifferentencounteredclustersrevealedvaryinglevelsofhomogeneity.WhiletheclusterforScandinaviannations(center,grey)clearlyreflectedthetightrelationshipsbetweentheinvolvedcountries,theclusterofAsianandNorthAmerican(topright,cyan)

Figure 5. Residual network of aggregated nation to nation links

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

9

waslesshomogeneous,suggestingthatthecommunitydetectionalgorithmwasnotabletodetectalltheclusterspotentiallypresentintheobserveddataset.Overall,theclustersofnationalitiesrevealedinterestingassociationswithanetworkofWikipedialanguagesstudiedin(Samoilenko,Karimi,Edler,Kunegis,&Strohmaier,2016).TheauthorsfoundculturalclustersofWikipedialanguageeditionsandusedcontemporarydatasourcesaboutcountryfeaturessuchasreligionandspokenlanguagestoexplaintheirformationtobemainlybasedonbilingualityandsharedreligion.Althoughbasedondifferenttypesofentities,theclusteringofcontemporaryandhistorical“nations”observedinthisworkappearedtobebasedonrelativelysimilarsocio-culturalfactors,raisinginterestingquestionsforfuturework.Consideringacomparablequantitativeanalysisforhistoricalentitiesthatexistedonlyforcertaindurationsandsignificantlychangedtheirgeographicalandculturalextentovertime,however,wouldrequiresignificanteffortforacquiringorevencreatingusabledatasourcesinthisregard.

BIPARTITe NeTwoRK oF STyLeS AND PeRSoNS LINKeD IN wIKIPeDIA

ThebottomupviewsoninteractionsbetweenpersonsofdifferentculturaloriginderivedfromtheULANandWikipediaprovidedinterestinginsightintothehigh-levelstructureofdevelopmentsinarthistoryandshowedanalogiestohistoricalscholarlydiagrams.Manyofthelatter,suchastheinitiallymentionedonefrom(Barr,1986/1937),however,mainlyconnecteddifferentstyleswitheachotheranddidnotfeaturemanyindividuals.Suchformalistviewsonstylisticinfluenceshavebeendrivenbypositionssuchasthe“arthistorywithoutnames”,assumingthatstyleexistsbeyondtheindividualgeniusandisrepresentativeforaspecificcultureataspecifictime.Computationalapproachestosuchconceptionsofstylisticinfluenceswouldcallforadvancedimageanalysismethodsallowingtodirectlyextractpatternsofstyleandtheirmutualinteractionsfromtheartworksthemselves.Whilesuchmethodswillcertainlyinfluencearthistoryresearchinfuture,itisneverthelessofinteresttoanalyzetheexistinghistoriographyaboutartstylesandperiodsforitshigh-levelstructureandexplorebird’seyevisualizationsofdevelopmentsrepresentedthere.

Theidentifiedcharacteristicsofthelarge-scale,multi-lingualWikipediabiographynetworkthusmotivatedtocombineitwithadditionalinformationaboutartandarchitectureperiodsandstyles,whichwasrealizedasabipartitenetworkconsistingonlyoflinksbetweenbiographiesandmulti-lingualWikipediaarticlesreferencedfrom426distinctWikidataentriessuchas“Renaissance”or”Cubism”.Theunderlyingassumptionwasthatalargeenoughdatasetofsuchlinkswouldyieldacontiguousstructureofstylesinterlinkedviabiographies,featuringachronologicalsuccessionasfoundforthebiographicalnetworksalone.TheextractionofWikipediahyperlinksbetweenthe426identifiedstylearticlesandanyWikipediapersonbiographyfromthe50largestDBpedialanguageversionsresultedinacombineddatasetof145,079hyperlinksconnecting406oftheidentifiedstyleswith69,908persons.Onlyaboutaquarterofthelatterwereamongstthe40,000+ULANbiographiesmappedtoWikipedia,suggestingthestyle-basedidentificationtobeanalternativewaytofindart-historicallyrelevantpersonsintheencyclopedia.

Sincea firstvisualizationof thebipartitenetworkdidnot showanyapparentchronology, atemporalfilteringapproachliketheoneusedforthepersonnetworkswasconsidered.Asitturnedout,especiallythetemporaldistributionsofpersonslinkedtobroadstylisticperiodssuchasBaroqueorRenaissanceoftencoveredaverywiderangeacrosscenturiesandwerenotlimitedtoonlythepersonsaliveduringtheiractualduration.Thissuggestedtofilterthedatabyremoving,foreachstyle,thelinksfrompersonswhowerenotaliveduringitsactualhistoricalperiod,whichwascomplicatedbymissingbeginningandendingdatesinmanyoftheWikidatarecordsaboutstylesandrequiredtoderivethedurationsviathetemporaldistributionofinterlinkedpersonsinstead.Asimpleapproachwastoextract“peaks”inthetemporaldistributionsofpersonslinkedtoeachstyleandtochoosetheareaaroundthelargestpeakascandidateforitsactualduration.Unfortunately,thiswasveryimpreciseandfailedespeciallyfor“older”periodswhichweremoreoftenreferencedbyrecentbiographiesthanbythosefromtheirowntime,callingforanalternativeapproachtakingthegenerallyskewedtemporal

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

10

distributionofpersonsintoaccount.Thissuggestedtocreateacontingencytabledividingstyle-personlinkcountsintorowsbystyleandcolumnsbyhalf-centuriesrepresentingtheroundedaverageofthebirthanddeathyears(mid-life-year)ofinterlinkedpersonsandtocalculatestandardizedresidualsasdescribedin(Agresti,2007).Foreachstyle,thehalf-centurywithhighestpositiveresidualanditsneighborhoodaboveacertainthresholdweredeterminedtobethestyle’sduration.

Table 1 illustrates this procedure for a dummy example. Four styles have differenttemporaldistributionsofpersonsinterlinkedwiththeminWikipedia.Foreachstyle,theleftmostvalue in eachcell represents theobservedcountof interlinkedpersonshavingtheir“mid-life-year”intherespectivehalf-century.Thecentervalueineachcellrepresentsthe expected count as derived from the marginal distributions (shown at the right andat the bottom of the Table), while the right value represents the resulting standardizedresidualderivedfromthedeviationbetweenobservedandexpectedcounts.Assumingathresholdofzero,theresidualsmarkedwithanasteriskshowthepeakinghalf-centuriesandtheirabove-thresholdneighborsdeterminingtheextracteddurationoftherespectivestyle.Thesecondrowshowsanexampleforastylehavingmorereferencesfromlaterbornpersonsthancontemporaries.Theresidual-basedapproachdeterminesthelowercountingcontemporaries tobedeviatingmoresignificantly fromexpectationand thus favors the“earlier”peaktorepresentthestyle’sduration.

After removing 31 styles whose temporal distribution never exceeded two person links perhalf-century,theresidual-basedapproachwasevaluatedagainstthecountbasedoneusingmanualannotationswithbeginningandenddatesfortheremainingsetof375styles.Linksbothfallingwithintheannotatedstyledurationandthedeterminedstyledurationwerecountedas“truepositives”,linksonlyfallingintotheformeras“falsenegatives”,thoseonlyintothelatteras”falsepositives”,thoseoutsidebothrangesas”truenegatives”.BothF1aswellasMCCmeasuresrankedtheresidual-basedapproachtoyieldbettercombinationsofprecisionandrecallandsuggestedaquitelowthresholdof0.33fordeterminingstandardizedresidualpeaksandtheirneighborhoods.Forthe375styleperiods,themanualannotationsandtheextractedbeginningandenddatesatleastoverlappedfor353ofthem,while22weremismatched.

Theextractedstyledurationswereusedtofilterthenetworktoareducedsetof93,197hyperlinksconnectingthe375styleswith55,499personsandasshowninFigure6,theprunednetworknowunfoldedintoanapproximatelychronologicalstructure.Asuccessionofstylesbecameclearlyvisible,startingwith(Pre-)Romanesqueartandarchitecture,continuingwithGothic,Renaissance,Mannerist,Baroquestylesuntilabouttheeraof(Neo)Classicism,wherethesuccessiondividedintorelativelydistinctstrands,thetoppartmainlyrepresentingarchitectural,thelowerpartfineartstyles.ThesetwostrandswereneverthelesscontinuouslyinterlinkedviastylessuchasArtNouveau,ArtDecoandBauhaus,whichappearedasbridgesbetweenthenowseparateddisciplines.TheregionbeneathBauhausalsocontainedtheothermodernartstylesfeaturedinBarr’s“DiagramofStylisticEvolutionfrom1890until1935”from(Barr,1986/1937),showingthatthedata-drivenversionderivedfromWikipediasuccessfullyputtheminalargerarthistoricalcontextandalsoincludedstyleswhichwere

Table 1. Determining stylistic periods via standardized residuals

1700 1750 1800 1850 1900 1950

Style 1 0/3.46/-2.35 0/5.35/-3 5/3.46/*1.04 10/5.35/*2.61 20/12.91/*2.9 5/9.45/-2 40

Style 2 10/3.38/*4.53 7/5.22/*1.01 5/3.38/*1.11 0/5.22/-2.95 12/12.59/-0.24 5/9.21/-1.91 39

Style 3 1/1.91/-0.75 10/2.94/*4.86 0/1.91/-1.59 5/2.94/1.42 6/7.1/-0.55 0/5.2/-2.87 22

Style 4 0/2.25/-1.76 0/3.48/-2.25 1/2.25/-0.98 2/3.48/-0.96 3/8.39/-2.54 20/6.14/*7.17 26

11 17 11 17 41 30 127

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

11

notrepresentedinthescholarlyversion.Therightmostpartwasoccupiedbystylesrepresentingmorerecentpost-WWIIdevelopmentsinthearts.

UsingthepalettefromFigure1tocolorthepersonnodesbynationalityagainrevealedaseparatestrandofJapanesestylesatthebottomwhichmergedwiththemainstrandataround1900,butalsohighlightedotherstylestobemorehomogeneousregardingtheoriginofinterlinkedpersons,suchas“ArtsandCraftsMovement”,“Futurism”,“Modernisme”,“SchoolofParis”,“AbstractExpressionism”and“SocialistRealism”.EspeciallythelatterwasseparatedfromthemainbodyandcontainedmanyartistsnotfeaturedintheULAN,underliningthespecificfocusofthebiographiescollectedthere.Similarlydetachedandnotevendirectlyrelatedtoarthistory,“20th-centuryclassicalmusic”wasalsomarkedasartmovementinWikidataandthusincludedinthevisualization.Itsconnectionwiththemainbody reflects themutual influencebetweendifferentartisticdisciplinesand invites forfurtherexploration.

Bipartite ProjectionsThebipartiteperson-stylenetwork couldbe reduced to twounipartitenetworks, one connectingpairsofstylesvialinksweightedbythenumberofcommonlylinkedpersons,theotheroneviceversa,viabipartitenetworkprojectionasdescribedin(M.Newman,2010,p.123ff.).Especiallytheresultingstyle-stylenetworkwasofinteresttobestudiedinmoredetailbutasencounteredforthenationalitynetwork,itfeaturedafewstronglyinterconnectedstyleswhichmaskedpotentiallyrelevantinteractionsbetweenlessprominentstyles.Thiscalledforanotherapplicationofastandardized-residual-basedfilteringtechniqueandtheFDSMmethoddescribedin(Zweig&Kaufmann,2011)wasasuitableapproachforbipartiteprojections.Therequired10,000randombipartitenetworkscould be efficiently calculated using functions from the R library BiRewire, demonstrating thatavailablesoftwarecomponentsfromthedomainofBioinformaticscouldbestraight-forwardlyre-usedinculturalanalyticssettings.

The unipartite network of styles weighted by the FDSM-based standardized residuals wassubsequently filteredbypruningall style-style linkshavingstandardized residualweightsbelowthreeanditsresultingvisualizationisshowninFigure7.Clustersofdenselyinterconnectedstyles

Figure 6. Pruned bipartite network of art and architecture styles linked via biographies, the latter colored by nationality

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

12

wereidentifiedviatheLouvainclusteringmethodandareshownindifferentcolorsintheFigure,theygenerallyappearedtofollowachronologicalorder:Fromlefttoright,theclustersstartedwithagroupofRomanesque/Gothic/Renaissanceartsandarchitecturestyles(green)whichblendedintoagroupofBaroque/Classiciststyles(purple)whoseendataround1800againmarkedtheseparationof tworelativelydistinctstrandofgroupsof19th(red)and20th/21stcentury(yellow)architecturestylesatthebottomandgroupsof19th(cyan),early20thmodernist(orange)andpostwar/21stcentury(blue)artstylesatthetop.Incontrasttotheimmediatevisualizationoftheperson-stylenetworkinFigure6,therepresentationoftheprojectedandstatisticallyfilteredstyle-stylenetworksuggestedthatmodernismrepresentedaperiodwhereartandarchitectureapproximatedeachotheraftertheirlongphaseofseparation,justtobefollowedbymorerecentdevelopmentsthroughoutwhichtheyappearedtodriftapartagain.

ReLATeD woRK

Asfarasdata-basedvisualizationsofarthistoricaldevelopmentsareconcerned,amaindistinctioncanbedrawnbetweenformalistandsocial-context-orientedconceptionsofarthistoricalprocesses.Approachestowardstheformerincludetheclusteringofartworksbyimagefeaturessuchasdominantcolor,saturationandbrightness,discussedin(Manovich,2015),whilemorerecentworksuchasdescribedin(Elgammal,Mazzone,Liu,Kim,&Elhoseiny,2018)appliedcontemporarymachine-learningmethodstoextractstylisticrelationshipsbetweenartworksbasedonWölfflin’sseminalworkonstylisticpatterns.Potentiallylocatedbetweenformalistandsocial-contextapproaches,visualizationsofmetadatacollectionsoftencontainaspectsfromboth.VisualizingitemsfromtheTatecollectionmetadata,theworkpresentedin(BoydDavis&Kräutli,2014)forexamplediscussedthebenefitsofusingvisualmeanstonavigatelargemuseumcollectionsacrossdifferentdimensions.Regardingsocialcontext,(Schichetal.,2014)presentedvisualizedpatternsofmigrationviapersonsfeaturedintheULAN,providinganinterestingviewonarthistoricaldevelopmentsfromthisperspective.Morerecentworkfromthesamefirstauthorandotherspresentedin(Schich,Huemer,Adamczyk,Manovich,&Liu,2017)usednetworkvisualizationtoshowbuyer-sellerinteractionsatdifferentgeographiclocationsderivedfromtheGettyProvenanceIndex.VisualizationsoftheULANassociativelinknetworkfocusingonegonetworksaroundsinglepersonsselectedfromtheULANwereusedinearlyworkinthecontextofculturalwebportalssuchasCultureSampo,describedin(Kurki&Hyvönen,2009)andforconstructingvirtualmuseumlandscapessuchasdescribedin(Goldfarb,Arends,Froschauer,&Merkl,2011).BiographicalnetworksonWikipedianotspecificallyfocusedonarthistoryhavebeendiscussedandcomparedacrossdifferentlanguageversionsinworkssuchaspresentedin(Aragon,Laniado,Kaltenbrunner,&Volkovich,2012)andin(Eometal.,2015).Whilethelatterdidnotusevisualizationsofthenetworkfortheircomparison,theformerpresented

Figure 7. Projection of pruned person-style network between art and architecture styles

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

13

aviewonthenetworkofWikipediabiographiessimultaneouslypresentinmultiplelanguages,thushighlightingtheiragreementandnottheirdifferences.Moreover,aspectsregardingthechronologicalstructureofthenetworkwerenotconsidered.Theworkpresentedin(Samoilenkoetal.,2016)studiedanetworkofWikipedialanguageversionsbasedoncommonlyfeaturedtopics,revealingsignificantclustersofrelatedlanguageversionsappearingtofollowsocio-culturalcommonalities,forwhichtheauthorsidentifiedbilingualityandsharedreligiontobethedrivingfactors.

CoNCLUSIoN

ThisworkpresentedvisualizationsofbiographicalnetworksofimportantpersonsfromarthistoryaspresentintheGettyULANandininterlinkedbiographiesindifferentWikipedialanguageversions.Their mutual comparison with each other and with existing scholarly examples revealed bothsimilaritiesbasedonthecommoncanonofarthistoryaswellasdifferencesduetopresentinstitutionalandculturalbiases,thelatterespeciallybecomingvisibleinthevaryingcoverageofnationalitiesin different Wikipedia language versions. The analysis of the aggregated network of interlinkednationalities revealedhigh level interactionsbetweencultureswhichappeared tobeclusteredbysocio-historicalboundariessuchascolonialismandthechangingextentofhistoricalempires.Theextractionandvisualizationof abipartitenetworkofbiographies andart andarchitecture stylesintroducedanadditionalperspectiveonarthistoryonWikipediaandenabledtogobeyondtheULANassourceforidentifyingrelevantpersons.ThisrevealedinterestingdifferencesincoveragesuchastheexclusivepresenceofmanyUSSRartistsintheRussianWikipedia,forwhichnocorrespondingULANrecordscouldbeidentified.Overall,theresultssuggestedthatlargeenoughnumbersofinterlinkedbiographiesanddescriptionsofstylesandperiodscanbeusedtoconstructlarge-scaleviewsonarthistoricaldevelopments,buttheavailabledatasourcesmustbehandledwithcareduetoidentifiedbiases.Initiallyintendedasnavigationalaidsforvirtualartcollections,suchviewscanthusalsobeusedastooltorevealtheselectivecoverageofindividualculturalheritagedatasets.Asshownfortheanalysisofthemulti-lingualWikipedianetworks,theculturallyspecificfociofindividualdatasetscanbeleveragedforthegoodbyunitingtheminaconsolidateddatasetrepresentingmorethanthemeresumofitsparts,potentiallyleadingtomoreglobalviewsonculturaldevelopments.

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

14

ReFeReNCeS

Agresti, A. (2007). An Introduction to Categorical Data Analysis. Hoboken, NJ: Wiley-Interscience.doi:10.1002/0470114754

Aragon,P.,Laniado,D.,Kaltenbrunner,A.,&Volkovich,Y.(2012).BiographicalSocialNetworksonWikipedia:ACross-culturalStudyofLinksThatMadeHistory.InProceedings of the Eighth Annual International Symposium on Wikis and Open Collaboration (WIKISYM).NewYork:ACM.doi:10.1145/2462932.2462958

Auer,S.,Bizer,C.,Kobilarov,G.,Lehmann,J.,Cyganiak,R.,&Ives,Z.(2007).DBpedia:ANucleusforaWebofOpenData.InK.Abereret al.(Ed.),Proceedings of the 6th International Semantic Web Conference(Vol.4825).Busan:Springer.doi:10.1007/978-3-540-76298-0_52

Baca, M., & Gill, M. (2015). Encoding Multilingual Knowledge Systems in the Digital Age: The GettyVocabularies.Knowledge Organization,42(4),232–243.doi:10.5771/0943-7444-2015-4-232

Barr,A.(1941).AdvisoryCommitteeReportonMuseumCollections.NewYork:TheMuseumofModernArtArchive,P9.https://api.moma.org/Images/Size3/Images/MA70_CCCR.jpg

Barr,A.(1986).Cubism and Abstract Art.Cambridge,MA:BelknapPressofHarvardUniversityPress.(Originalworkpublished1936)

Blondel,V.D.,Guillaume,J.-L.,Lambiotte,R.,&Lefebvre,E.(2008).Fastunfoldingofcommunitiesinlargenetworks.Journal of Statistical Mechanics,2008(10),P10008.doi:10.1088/1742-5468/2008/10/P10008

BoydDavis,S.,&Kräutli,F.(2014).Scholarly Chronographics: Can a Timeline be Useful in Historiography?Paperpresentedatthe10thEuropeanSocialScienceHistoryConference,Vienna,Austria.

Elgammal,A.,Mazzone,M.,Liu,B.,Kim,D.,&Elhoseiny,M.(2018).TheShapeofArtHistoryintheEyesoftheMachine.In32.AAAI Conference on Artificial Intelligence.NewOrleans,LA.

Eom,Y.-H.,Aragon,P.,Laniado,D.,Kaltenbrunner,A.,Vigna,S.,&Shepelyansky,D.L.(2015).InteractionsofCulturesandTopPeopleofWikipediafromRankingof24LanguageEditions.PLoS One,10(3),e0114825.doi:10.1371/journal.pone.0114825PMID:25738291

Goldfarb,D.,Arends,M.,Froschauer,J.,&Merkl,D.(2011).Revisiting3dInformationLandscapesfortheDisplayofArtHistoricalWebContent.InProceedings of the 8th International Conference on Advances in Computer Entertainment Technology.ACM.doi:10.1145/2071423.2071480

Goldfarb,D.,Arends,M.,Froschauer,J.,&Merkl,D.(2013).ComparingArtHistoricalNetworks.Leonardo,46(3),279.doi:10.1162/LEON_a_00575

Goldfarb,D.,Arends,M.,Froschauer,J.,Weingartner,M.,&Merkl,D.(2014).CollectivizingtheBarrModel.Leonardo,47(3),270.doi:10.1162/LEON_a_00773

Green-Armytage, P. (2010). A Colour Alphabet and the Limits of Colour Coding. JAIC - Journal of the International Colour Association, 5.

Jacomy, M., Venturini, T., Heymann, S., & Bastian, M. (2014). ForceAtlas2, a Continuous Graph LayoutAlgorithm for Handy Network Visualization Designed for the Gephi Software. PLoS One, 9(6), e98679.doi:10.1371/journal.pone.0098679PMID:24914678

Joyeux-Prunel,B.(2014).TheUsesandAbusesofPeripheriesinArtHistory.Artl@s.Bulletin,3(1).

Klein,M.,&Kyrios,A.(2013).VIAFbotandtheIntegrationofLibraryDataonWikipedia.Code4Lib Journal,2.

Kurki, J.,&Hyvönen,E. (2009).AuthorityControlofPeopleandOrganizationson theSemanticWeb. InProceedings of the International Conferences on Digital Libraries and the Semantic Web.

Manovich,L.(2015).DataScienceandDigitalArtHistory.Intl. Journal for Digital Art History,1.https://doi.org/10.11588/dah.2015.1.21631

Newman, M. (2010). Networks: An Introduction. Oxford: Oxford University Press. doi:10.1093/acprof:oso/9780199206650.001.0001

International Journal of Art, Culture and Design TechnologiesVolume 8 • Issue 1 • January-June 2019

15

Doron Goldfarb received his MS in Computer Science from the Vienna University of Technology in 2008. Since then, he has been involved in different national and European scale research projects in the context of digital cultural heritage and virtual research infrastructures.

Dieter Merkl is Associate Professor of Applied Computer Science at the Institute of Information Systems Engineering, Vienna University of Technology, Austria. His main research interests are in the areas of information retrieval, data mining, and interaction design. He has published more than 140 scientific papers in these areas.

Newman,M.E.J.(2006,June).ModularityandCommunityStructureinNetworks.Proceedings of the National Academy of Sciences of the United States of America, 103(23), 8577–8582. doi:10.1073/pnas.0601602103PMID:16723398

Newton,E.(1956).European Painting and Sculpture(5thed.).Harmondsworth,Middlesex:Penguin.(Originalworkpublished1941)

Samoilenko, A., Karimi, F., Edler, D., Kunegis, J., & Strohmaier, M. (2016). Linguistic Neighbourhoods:ExplainingCulturalBordersonWikipediathroughMultilingualCo-EditingActivity.EPJ Data Science,5(1).https://doi.org/10.1140/epjds/s13688-016-0070-8

Schich,M.,Huemer,C.,Adamczyk,P.,Manovich,L.,&Liu,Y.-Y.(2017).NetworkDimensionsintheGettyProvenanceIndex.arXiv preprint arXiv:1706.02804.

Schich,M.,Song,C.,Ahn,Y.-Y.,Mirsky,A.,Martino,M.,Barabási,A.-L.,&Helbing,D.(2014).ANetworkFrameworkofCulturalHistory.Science,345(6196),558–562.doi:10.1126/science.1240064PMID:25082701

Schmidt-Burkhardt,A.(2005).Stammbäume der Kunst: Zur Genealogie der Avantgarde.Berlin:AkademieVerlag.

Strass,F. (1828).Der Strom der Zeiten oder bildliche Darstellung der Weltgeschichte von den ältesten bis auf die neuesten Zeiten. Leipzig: Köchly. (Original work published 1804), http://mdz-nbn-resolving.de/urn:nbn:de:bvb:12-bsb00003029-3

Vrandečić,D.,&Krötzsch,M.(2014).Wikidata:AFreeCollaborativeKnowledgebase.Communications of the ACM,57(10),78–85.doi:10.1145/2629489

Zweig,K.A.,&Kaufmann,M.(2011,April).ASystematicApproachtotheOne-ModeProjectionofBipartiteGraphs.Social Network Analysis and Mining,1(3),187–218.doi:10.1007/s13278-011-0021-0