AssignpeopleinmoviesPivotConnectPromotePipeAABBABGroupToggle directionConvertedgetohyperedgesupernodegroupandclassMerge DissolveedgeshyperedgesnodessupernodesMurejsTowardFlexibleAuthoringandReshapin ID: 857157
Download Pdf The PPT/PDF document "CHIVISVIS" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1 Assign people in movies Pivot Connect Pr
Assign people in movies Pivot Connect Promote CHIVISVIS Pipe A AB BAB Group Toggle direction Convertedge to hyperedge supernode, group,and class Merge / Dissolve edges hyperedges nodes supernodes Mure.js:TowardFlexibleAuthoringandReshapingofNetworksAlexBigelow*CarolinaNobreAlexanderLexMiriahMeyer§UniversityofUtah Figure1:Examplesofoperationsthatourframeworkismeanttosupport.Theseincludeusinggraphtopologyforpivotingselections,connectingnodes,assigningclassestoitems,convertingbetweenconstructs,groupingitems,promotingvaluestonodes,pipingattributestooralongedges,merginganddissolvingsupernodesandhyperedges,andtogglingedgedirection.IndexTerms:Human-centeredcomputingVisualizationVisu-alizationsystemsandtools;InformationsystemsDatamanage-mentsystemsDatabasedesignandmodelsGraph-baseddatabasemodels1INTRODUCTIONWheninterpretingdataasagraphforvisualization,ananalystrstassignssemanticmeaningtographconcepts.Forexample,theymaychoosetorepresentactorsandmoviesasnodes,androlesasedges.Alternatively,theymaywishtorepresentmoviesasedges,connect-ingactornodeswhentheycollaborate.Dataabstractionchoicessuchasthesearecritical,becausedifferentdataabstractionscanlimitorinspiredifferentanalysisquestions,approaches,perspectives,andvisualizations[2,5].However,currentnetworkmodelingframeworks,systems,anddatabasesnarrowlydenegraphabstractionconstructssuchasnodes,edges,node/edgeclasses,supernodes,hyperedges,etc.intermsofhowthedataisstoredinmemoryorondisk,ratherthansemantic,human-drivenabstractions.Consequently,theabilityofananalysttoiterateonadataabstractionbecomesfundamentallylimitedbytheimplementationdetailsofdatawranglingsoftware.Wepresentwork-in-progresstowardabroaderframeworkformodelingnetworkdatathatislessdependentonitsunderlyingstruc-tureandstorage.Ourgoalistousesemanticdataabstractioncon-structstoinformhowalgorithmswrangledata,insteadofallowingalgorithmicconcernstodeneandconstrainthesemantics.Wealsopresentmure.js,asoftwarelibrarythatrepresentsaninitialimplementationofthisframework,allowinguserstomapsemanticgraphconstructstoarbitrarydatastructuresasmetadata,thatcan,inturn,beusedtoselect,navigate,andreshapetheunder-lyingdata.Additionally,wediscussanearlysoftwareprototypeofavisualgraphwranglingsystembasedonthislibrary.2RELATEDWORKTheframeworkforsemanticnetworkmodelingthatwearedevel-opingextendsideasfromexistingnetworkmodelingframeworks,systems,anddatabasesthatstresstheroleoftheuserindeciding *e-mail:abigelow@cs.utah.edue-mail:cnobre@sci.utah.edue-mail:alex@sci.utah.edu§e-mail:miriah@cs.utah.eduwhatpartsofthedataarenodes,andwhatpartsofthedatainformconnections[4,6].Liketheseexistingefforts,weemphasizethefactthatadataabstractionisdesigned,ratherthannaturallyoccurring.Existingnetworkmodelingtoolstendtodeneconstructsintermsofspecicdatastoragestrategies,suchastheshapeofdataobjectsinmemory[3],asspecicinterpretationsofrelationaldatabaseconcepts[7],asspecicinterpretationsofattributevalues[6],orsomespeciccombinationoftheseinterpretations[4].Theseap-proacheshaveclearimplementationadvantages,inthatitiseasyandefcienttoimplementgeneral-purposealgorithms[3],efcientqueriesingraphdatabases[7],powerfulstrategiesforestablishingconnections[4,6],orefcientvisualizationsofmanynodes[1].However,rigid,data-basedconstructdenitionshavetwodistinctdisadvantages:theyresultinlimitedabstractions,andtheyarebrittletorefactoring.Whenconstructsaredenedbydatastructure,theyarelimitedbythatstructure.Forexample,NetworkX[3]requiresthatanentiregraphbedenedaprioritoconformtoaspecicsetofconstraints,suchasdirectedvsundirected.Theseconstraintsmakesensefromanalgorithmicperspective,butbecauseedgesmustfollowoneoftheseglobalpatterns,thereisnosupportforgraphsthatmixdirectedandundirectededges,orforotherconstructssuchashyperedges.Stillotherframeworks,suchasOrion[4],deneconnectionsassimplevalue-basedlinks,ratherthandistinctitems,makingitimpossibleforedgestohavetheirownattributes.Data-denedconstructscommonlyresultinworkarounds,suchascreatingintermediatenodesasaplacetostoreedgeattributes,orasaproxyforhyperedges.Theotherdisadvantageofdata-denedconstructsisthatiteratingonadataabstractionisdifcultorimpossibleinpractice.Relativelysimplesemanticchanges,suchasswappingwhatisinterpretedasanode,withwhatisinterpretedasanedge,canbeincrediblypowerfulinpractice[5],yetsupportforthiskindofgraphwranglingisunder-explored.Ourframeworkisdistinctfromexistingworkinthatnetworkmodelingtakesplaceatamoresemanticlevel.Itmapssemanticconstructstodatavaluesandstructures,ratherthandeningthemintermsofdatavaluesandstructures.Assuch,refactoringthemappingisfarlesscomplicated,anditenablesaricherlibraryofconstructs.3MAPPINGCONSTRUCTSTODATAGraphdatacanbechallengingtostore,especiallyasedgesoftenintroducecyclicalstructuresthatdonotmapwelltolesystemordatastructurehierarchies.Consequently,manystrategiesexistforrepresentingagraphinmemory,includingadjacencymatrices, node-linklists,andrelationaltables.Ourapproachistomapsemanticgraphconstructstoitemsindatastructuresastheyalreadyexist,ataninstancelevel,ratherthanattempttoforceallgraphdataintosomecanonicalstructure.Assuch,itemscouldberowsinaCSVleorrelationaltable,objectsinaJSONleorNoSQLhierarchy,orelementsinanXMLdocument.Atthedatalevel,theonlyrequirementisthatanitemmustatleastsupportattributes;itdoesnotmattertoourframeworkwhethertheyarenested,orwhereinaletheyexist.Onceitemsareinitiallymappedtosemanticconstructs,theyenableasetofoperationsthatallowconversionfromoneconstructtoanotherthisprocessinformsanddrivestheunderlyingdatawranglingoperations,makingitpossibletoconvertbetweendifferentstoragestrategies.Wehavehintedatoneclearlimitationofthisapproach:inat-temptingtoavoidasingle,canonicaldatastructure,wecannotrelyonthestrengthsofanyofthem,suchasthecompactnessofadjacencymatricesforhighly-connectedgraphs.T
2 hismeansthatwerisktheabilitytoscaletolar
hismeansthatwerisktheabilitytoscaletolargerdatasets.Wesuspectthattheremaybeafundamentaltrade-offindatawranglingtoolswheregreatersemanticexibilitymaycomeatthecostofscalability.3.1ConstructsOurframeworkcurrentlyidentiesnodes,edges,supernodes,hyper-edges,classes,andgroups.Thisisnotnecessarilytheonlysetofusefulconstructsthatexistweintendtorenethislist,withfeed-backfromthecommunity.Meta-iterationonconstructsthemselvesisanotheradvantageofourmorelightweight,semanticapproach.Thesimplestconstructmappingthatcanbemadeistoidentifyanitemasanodeanindependentiteminagraph.Anedgeisadependentitemthatlinksatleasttwonodes,evenifbothnodesarethesame(aself-edge).Anedgemayormaynotbedirected,independentofanyotheredgesinthegraph.Asupernodeisanodethatrepresentsasetofnodes;itallowstheusertosummarizeless-relevantnetworkcomplexity.Supernodesalsoprovideatargetforedgesthataremeanttolinktoasetofitemsasawhole,ratherthanitsindividualmembers.Ahyperedgeisanedgethatlinksmorethantwonodes,andmaybefullydirectional,partiallydirectional,orundirected.Nodesandedgesmaybefurtheridentiedwithclassesamecha-nismwherebytheuserorganizesitemsbytheirsemantic,real-worldmeaning,enablingitemcomparisonacrossacommonsetofat-tributes.Additionally,groupsallowausertospecifyinformalorad-hocitemrelatednessdistinctfromthetopologicalnotionofrelatednessrepresentedbyedges,supernodes,orhyperedges,andalsodistinctfromtheattribute-basedrelatednessimpliedbyclasses.3.2OperationsOnceinitialconstructsareinplace,theyenablearichsetofop-erationsthattheusercanperformtoreinterpretandreshapethedata.Similartoconstructsthemselves,availableoperationsarealsoopen-endedinourframework;theoperationslistedhere,alsoshowninFigure1,aremeanttobeillustrative,notcomprehensive.Wedescribeeachoperationintermsofuserselectionsinourprototypeinterface,thesemanifestasinterfaceselections,whereasinthemure.jslibrary,theyaremoresimilarinspirittoD3selections.Oneselectioncanyieldanotherbypivotingeitherto,from,oralongedges;orfromaset-likeconstruct(supernode,group,orclass)toitscontentsormembers.Itemscanbegrouped,savingtheselectionforlateruseorcon-versiontoadifferentconstruct;theycanalsobeassignedtoneworexistingclasses.Nodescanbemergedintoasupernodes;supern-odescanbedissolvedbackintoregularnodes;nodescanconnectedtoeachother,similarto(butnotnecessarilyimplementedas)arela-tionaldatabasejoin;orconvertedtohyperedges,mergingthenodes'existingedgesintoasingleitem.Similarly,selectededgesandhyper-edgescanbemergedanddissolved;orconvertedtonodessplittinganedgeintoseparateentities,andcreatinganintermediatenode.Selectedvaluesorattributescanbepromotedtodistinctnodes,optionallyconnectingtheiroriginalsourcenodeswithnewedges.Attributescanalsobepipedalongedges,derivingnewnodeoredgeattributesbasedonconnectedproperties.4IMPLEMENTATIONWeareimplementinganexampleofourframework,focusingontheaboveconstructsandoperationsasanopen-sourcelibrarycalledmure.js1.Similartoothergraphdatabaseapproaches,itisasemanticlayerontopofanexistingdatabaseinthiscase,PouchDBbut,unlikeexistingapproaches,thelayerisfarmorelightweight,andcon-structsdonotrelyspecicallyontheshapeorstructureofPouchDBdocuments.Instead,constructsaremappedtoitemsindocumentsoreventhedocumentsthemselvesthroughametadatalayer.Inconjunctionwiththelibrary,weareprototypingavisualtoolforsemanticgraphwrangling.Theinterfaceexposestheconstructsandoperationsmadeavailablebythelibrary,inawaythatnon-programmerswillbeabletotakeadvantageofthem.Additionally,itwillprovidebasicvisualizationcapabilitiestoassistininspectingthedatabeforeitisexportedfordeepervisualizationinotherprograms.5CONCLUSIONSANDFUTUREWORKAsmentionedabove,weplantocontinuereningthesetofcon-structsandoperations,andtocontinuedevelopinganddesigningmure.jsanditscorrespondingvisualinterface.Wealsointendtoevaluatetheexpressiveness,usefulness,and/ortheusabilityoftheresultingsystems.Inpresentingthisposter,wehopetosolicitfeedbackfromthecommunityonourcurrentdirections.Perhapsmostimportantly,weseekopportunitiestodiscusshowtobestevaluatethiskindofwork.Whatkindsofusabilitytestsaremostappropriate?Howandshoulddatawranglingtoolexpressivenessbevalidatedbeyondcasestudies?Howcanweevaluatetheusefulnessofawranglingtoolinawaythatcanprovideinsightwithrespecttothesetofconstructsandoperationsthatitsupports?REFERENCES[1]M.Bastian,S.Heymann,andM.Jacomy.Gephi:AnOpenSourceSoftwareforExploratingandManipulatingNetworks.ProceedingsoftheThirdInternationalICWSMConference,p.2,2009.[2]A.Bigelow,S.Drucker,D.Fisher,andM.Meyer.ReectionsonHowDesignersDesignwithData.InProceedingsofthe2014InternationalWorkingConferenceonAdvancedVisualInterfaces,AVI'14,pp.1724.ACM,NewYork,NY,USA,2014.doi:10.1145/2598153.2598175[3]A.A.Hagberg,D.A.Schult,andP.J.Swart.ExploringNetworkStructure,Dynamics,andFunctionusingNetworkX.InG.Varoquaux,T.Vaught,andJ.Millman,eds.,Proceedingsofthe7thPythoninScienceConference,pp.1115.Pasadena,CAUSA,2008.[4]J.HeerandA.Perer.Orion:Asystemformodeling,transforma-tionandvisualizationofmultidimensionalheterogeneousnetworks.InformationVisualization,13(2):111133,Apr.2014.doi:10.1177/1473871612462152[5]C.B.Nielsen,S.D.Jackman,I.Birol,andS.J.M.Jones.ABySS-Explorer:visualizinggenomesequenceassemblies.IEEEtransactionsonvisualizationandcomputergraphics,15(6):881888,Dec.2009.doi:10.1109/TVCG.2009.116[6]A.Srinivasan,H.Park,A.Endert,andR.C.Basole.Graphiti:InteractiveSpecicationofAttribute-BasedEdgesforNetworkModelingandVisu-alization.IEEETransactionsonVisualizationandComputerGraphics,24(1):226235,Jan.2018.doi:10.1109/TVCG.2017.2744843[7]J.Webber.AProgrammaticIntroductiontoNeo4j.InProceedingsofthe3rdAnnualConferenceonSystems,Programming,andApplications:SoftwareforHumanity,SPLASH'12,pp.217218.ACM,NewYork,NY,USA,2012.doi:10.1145/2384716.2384777 1Thelibrary,sourcecode,anddocumentationareavailableathttps://github.com/mure-apps/mure-librar