搜索
ThemeRiver:VisualizingThemeChangesoverTimeSusanHavre,BethHetzler,andLucyNowellBattellePacificNorthwestDivisionRichland,Washington99352USA1+509+375-6948{susan.havre|beth.hetzler|lucy.nowell}@pnl.govAbstractThemeRiver™isaprototypesystemthatvisualizesthematicvariationsovertimewithinalargecollectionofdocuments.The“river”flowsfromlefttorightthroughtime,changingwidthtodepictchangesinthematicstrengthoftemporallyassociateddocuments.Colored“currents”flowingwithintherivernarroworwidentoindicatedecreasesorincreasesinthestrengthofanindividualtopicoragroupoftopicsintheassociateddocuments.Theriverisshownwithinthecontextofatimelineandacorrespondingtextualpresentationofexternalevents.Keywords:visualizationmetaphors,trendanalysis,timeline1.IntroductionInexploratoryinformationvisualization,onegoalistopresentinformationsothatuserscaneasilydiscernpatterns.Patternsrevealtrends,relationships,anoma-lies,andstructureinthedata,andmayhelpusersFigure1:ThemeRiver™usesarivermetaphortorepresentthemechangesovertime.ProceedingsoftheIEEESymposiumonInformationVisualization2000(InfoVis'00)0-7695-0804-9/00$10.00@2000IEEEconfirmknowledgeorhypotheses.Perhapsmoreimpor-tantly,theyalsoraiseunexpectedquestionsleadinguserstonewinsights.Thechallengeistocreatevisuali-zationsthatenableuserstofindpatternsquicklyandeasily.ThemeRiver,showninFigure1,isaprototypesystemdesignedtorevealtemporalpatternsintextcollections.InformationvisualizationsystemssuchasEnvision[13],BEAD[1],LyberWorld[3,4]andSPIRE[18]representeachdocumentorgroupofdocumentswithaglyphoricon,portrayingvariousdocumentattributes.Variousmethodshavebeenexploredforshowingchangeovertimeindocument-centricvisualizations.SeeSection3below.However,ausermaybelessinterestedindocumentsthemselvesthaninthemechangeswithinthewholecol-lectionovertime.Forexample,howdidShakespeare’sthemeschangeduringvariousperiodsofhislifeorinrelationtocontemporaryevents?Suchinformationisdifficult,ifnotimpossible,togleanfrommostvisuali-zations.Avisualizationthatfocusesonthemes,ratherthandocuments,couldbemoreusefulforsuchexplora-tion.ThemeRiverprovidesuserswithamacro-viewofthematicchangesinacorpusofdocumentsoveraserialdimension.Itisdesignedtofacilitatetheidentificationoftrends,patterns,andunexpectedoccurrenceornon-occurrenceofthemesortopics.Inourprototype,weusetimeastheserialdimension.Weprovidecontextualinformationthroughatimelineandmarkersforco-occurringeventsofinterest.Figure1showsasampleThemeRivervisualization.ThispaperdescribesthedesignofThemeRiver,walksthroughasampleinforma-tionexplorationsession,anddiscussesresultsofforma-tiveusabilitytesting.2.DesignOurmajordesigngoalwastoprovideavisualizationofthemechangeovertime.Considerusingahistogramtovisualizethesechanges.Inahistogram(suchastheoneshowninFigure2),eachbarrepresentsatimeslice,andcolorvariationsandsizewithinthebarrepresenttherelativestrengthofthemesspecifictothatslice.However,understandingthehistogramrequiresuserstoworkatintegratingthethemesacrosstimebecausethebarsareanchoredtoabaselineandthepositionofaparticularthemewithinthebarsmayvaryconsiderably.Likeahistogram,ThemeRiverusesvariationsinwidthtorepresentvariationsinstrengthordegreeofrepresentation.However,itconnectsthestrengthvaluesinadjacenttimesliceswithsmoothandcontinuouscurves.Thehorizontalflowoftheriverrepresentstheflowoftime.Coloredcurrentsthatrunhorizontallywithintheriverrepresentthemes.Eachverticalsectionoftherivercorrespondstoanorderedtimeslice.Thewidthofeachcurrentchangestoreflectthethematicstrengthforeachtimeslice.Forexample,inFigure1thetheme“soviet”increasesinrelativestrengthinJune1960asindicatedbythewideningoftheupperbrightorangecurrent.“Soviet”losesrelativestrengthinJulyandAugust;thusthesamecurrentnar-rowsinthenexttwotimeslices.“Soviet”thenincreasessignificantlyinrelativestrengthinSeptember;thecurrentwidensproportionately.Currentsmaintaintheirintegrityasasingleentityovertime.Ifathemeceasestooccurinthedocumentsforaperiodoftimeandthenrecurs,thecurrentlikewisedisappearsandthenreappears.Consistentcolorandrelativepositiontootherthemesmakethemecurrentseasytorecognize.InFigure1,thelowerpurplebanddepictsthechangesinrelativestrengthofthetheme“cane.”The“cane”currentoccursgrowsandshrinksovertime;“cane”occursmoststronglyinMarch1961.WebelievethatThemeRiver’scontinuouscurveshavemuchtodowithitsusability.TheGestaltSchoolofPsychology[8],foundedin1919inGermany,theorizedthatwithperception,“thewholeisgreaterthanthesumoftheparts.”Simplyput,duringtheperceptionprocesshumansdonotorganizeindividual,low-level,sensedelements,butsensemorecomplete“packages”thatrepresentobjectsorpatterns.Inhisrecentbook[6],Hoffmanpresentsacompellingdiscus-sionofhowourperceptualprocessesidentifycurvesandsilhouettes,recognizeparts,andgroupthemtogeth-erintoobjects.Numerousaspectsoftheimageinflu-enceourabilitytoperceivethesepartsandobjects,includingsimilarity,continuity,symmetry,proximity,andclosure.Forexample,itiseasiertoperceiveobjectsthatareboundedbycontinuouscurvesthanthosethatcontainabruptchanges[17].Theverticalproximityoftherivercurrentsmakesiteasyforu