recipedrupal7

HidetheStack:TowardUsableLinkedDataAba-SahDadzie1,,MatthewRowe2,andDanielaPetrelli31OAKGroup,Dept.
ofComputerScience,TheUniversityofSheeld,UK2KnowledgeMediaInstitute,TheOpenUniversity,MiltonKeynes,UK3Art&DesignResearchCentre,SheeldHallamUniversity,UKa.
dadzie@dcs.
shef.
ac.
uk,m.
c.
rowe@open.
ac.
uk,d.
petrelli@shu.
ac.
ukAbstract.
TheexplosioningrowthoftheWebofLinkedDatahaspro-vided,forthersttime,aplethoraofinformationindisparatelocations,yetboundtogetherbymachine-readable,semanticallytypedrelations.
UtilisationoftheWebofDatahasbeen,untilnow,restrictedtothemembersofthecommunity,eatingtheirowndogfood,sotospeak.
TotheregularwebuserbrowsingFacebookandwatchingYouTube,thisutilityisyettoberealised.
TheprimaryfactorinhibitinguptakeistheusabilityoftheWebofData,whereusersarerequiredtohavepriorknowledgeofelementsfromtheSemanticWebtechnologystack.
Ourso-lutiontothisproblemistohidethestack,allowingenduserstobrowsetheWebofData,exploretheinformationitcontains,discoverknowl-edge,anduseLinkedData.
Weproposeatemplate-basedvisualisationapproachwhereinformationattributedtoagivenresourceisrenderedaccordingtotherdf:typeoftheinstance.
Keywords:LinkedData;KnowledgeVisualisation;InformationVisual-isation;UsableInterfaces;Human-ComputerInteraction.
1IntroductionTheWebofLinkedDatanowconnectsawiderangeofpreviouslydisparateandisolatedinformationsources,allowingcomplex,bespokequeriestobeansweredthatwerepreviouslynotpossibleorhardtoderiveanswersfor.
Totech-savvyusers,andinparticular,researchersintheLinkedData(LD)community,con-sumptionofLDiseasygiventheirknow-howwritingSPARQL1queriesorbyapplyingafollow-your-noseprincipletosnioutfactsandconnectionsbetweenpiecesofinformation.
However,tothemainstreamwebuser–whowedeneasthefrequentuserwhobrowseswebsites,chatswithfriendson,e.
g.
,Facebook,buthasnorealknowledgeoftheintrinsicfunctionalityoftheWebtheybasetheirinteractionon–thereexistsagapbetweenexploitingtheWebofData(WoD)toanswerqueriesandthetechnologicalknow-howtodoso.
Theregularwebuserdoesnot(andshouldnotneedto)knowSPARQL,norRDF2(Re-sourceDescriptionFramework),whatanontologyorLinkedDatais,noranyotherelementwhichtheSemanticWeb(SW)encompasses.
Thisisaproblem.
Towhomcorrespondenceshouldbeaddressed.
1SPARQLquerylanguageforRDF:http://www.
w3.
org/TR/rdf-sparql-query2ResourceDescriptionFramework(RDF):http://www.
w3.
org/RDFG.
Antoniouetal.
(Eds.
):ESWC2011,PartI,LNCS6643,pp.
93–107,2011.
cSpringer-VerlagBerlinHeidelberg201194A.
-S.
Dadzie,M.
Rowe,andD.
PetrelliToreducethisgap,weproposetomakeLDusable,allowingenduserstoembracethepoweroftheWoDandbrowseanddiscoverconnectionsbetweenpiecesofinformationandfactsastheywouldontheWorldWideWeb(WWW–theReadableWeb).
Inbridgingthegapwewillputapowerfuldatabaseatthedisposalofendusers,onewhichiscommunitymaintainedandprovidesanswerstouniquequestions.
Inessenceweproposetohidethestackfromendusers,allowingthemtouseRDF,SPARQL,ontologies,andallthoseotherelementsthatmakeupthelayercake[4],withoutbeingawaretheyaredoingso.
Wedenethisthesisastheinvisiblestackthatisexpressedbytheresearchquestion:HowcanwemakeLinkedDatausabletoreal,endusersFromexploringthisquestionweproposeatemplate-basedapproachtovisu-alisinglinkeddata,that,startingfromtheunderlyingdatastructureassociatedwithagivenresource,presentsinformationinalegibleandcoherentform.
Therdf:typeofaresourceprovidestheprimaryindicatorastowhichtemplate(s)toemploy,bydereferencingtheURI,returningtheinstancedescription,andtai-loringthisinformationintoalegibleform,thatcaterstotheuser'scontext,i.
e.
,currenttaskandendgoal.
Ourapproachreducestheinformationloadontheuser–anendemicproblemconcerningLDvisualisation,giventhescaleoftheWoD–supportingeasierinterpretationandacoherentviewofdata.
Wehavestructuredthispaperasfollows:section2describesthechallengesimposedonthevisualisationofLD,basedonthecurrentstateoftheWoD.
Section3presentsrelatedworkandhowsuchchallengeshavebeenaddressedthusfar.
Section4containsourcentralcontribution:ourapproachtovisualis-inglinkeddataviatemplates.
Wedenescenarioswhichmotivateandprovidereferencesforclarityinexplainingourapproach.
Thesectionconcludesbydis-cussingaformativeevaluation.
Section6concludesthepaperwiththendingsandlessonsdrawnfromourwork,andplansforfuturework.
2TowardUsableLinkedDataLinkedDatainitsrawformconsistsof(oftenverylarge)setsofRDFstatements.
RDFwasdesignedtosupportreadingandinterpretationofdataontheWebbymachines.
Thisfocusoftenresultsindatathatisnotalwayseasilyinterpretedbyhumans,especiallyoutsidetheSWcommunity.
Take,e.
g.
,Fig.
1B,whichde-scribesapublicationintheData.
dcs3linkeddataset;theURI(UniformResourceIndicator)http://data.
dcs.
shef.
ac.
uk/paper/4169thatreferencesitisfairlycryptic–theonlyinformationdirectlyderivedfromitisthatitisapaperbe-longingtotheinstitutionrepresentedbytheURIhttp://data.
dcs.
shef.
ac.
uk(henceforthabbreviatedas'data.
dcs:').
ThecompleteRDFdescription,basedonitsbibtex4citation,ishowevereasilyinterpretedbyhumans;thisincludesitsbib:title,publicationyear(bib:hasYear),thebook/collectionthatcontainsit(bib:hasBookTitle),anditsauthors(foaf:makers).
3TheData.
dcsαLDdatasetmaybebrowsedfrom:http://data.
dcs.
shef.
ac.
uk4SeetheBibTeXresourcepagesat:http://www.
bibtex.
orgHidetheStack:TowardUsableLinkedData95Aba-SahDadzie.
.
.
96152418f9a4f2d4512f07fa06efd308f34cbb6bImprovingSupportforWeb-basedVisualAnalysisofSocialGraphs.
2009IEEEVisWeekWorkshopMatthewRowe.
.
.
bd2cda94c756832460fd7c8f6de5c3d2525bbdbaOrganisations,InformationandKnowledgeGroup.
.
.
.
.
.
.
.
.
.
.
.
ABCDFig.
1.
ExtractsfromData.
dcs,highlightinglinksbetweenselectedresources.
AandCdescribetwofoaf:Personresources;Dtheirswrc:affiliation;Babib:Entryresource(apaper),http://data.
dcs.
shef.
ac.
uk/paper/4169,commontoboth.
Figs.
1AandCdescribetwofoaf:Personresources(withorangeborders):data.
dcs:person/Matthew-Roweanddata.
dcs:person/Aba-Sah-Dadzie,res-pectively,indirectlylinkedthroughco-authorshipofthepaperinB(bib:Entry,green).
Ddescribesasecondindirectrelation,usingabrokenlink–theircommonswrc:affiliation,data.
dcs:group/oak(blue).
Fig.
1eectivelycommunicatestheinter-relationshipsbecausesmallextractsfromData.
dcshavebeencollectedincloseproximityandspecicregionshigh-lightedandcolourcoded.
This,supplementedwithbi-directionalarrowsbetweenresources,resultsinsimplevisualencodingthatallowstheusertogainveryquicklyanoverviewandunderstandingoftherelationshipswithinthedata.
Overlayingthevisualencodingonthetextallowstheusertodelvefurtherintothedatatoretrievemoredetailforregionsofinterest(ROIs),e.
g.
,browsingfromtheresourcedata.
dcs:person/Matthew-Rowetoobtainahuman-readablede-scriptionoftheobjecttheymade:data.
dcs:paper/4169.
Asdatasetsizeincreases,however,humanabilitytoidentifysuchrelation-shipsandretaintheminmemorydecreasessignicantly.
Thisposesachallengefortheverylargeamountsofdatageneratedintoday'sinformation-richsoci-ety,withdatasetscontaininguptomillionsofentities.
Data.
dcsbycomparisonistiny,containingonly8000statements.
Howevereventhisposessignicantcognitivechallengesformanualanalysis,duetothedicultyobtainingagoodmentaloverviewoflargeamountsofcomplex,highlyinterlinkeddata[8,9,10].
Furtherdicultiesarisewhenexploringanewenvironmentaboutwhichauserhaslittleinformation,beyondthatitcontainsanswerstothequestionstheywishtoask.
WhileanSWexpertwouldbecomfortablewithorexpecttostartbrowsingLDfromaspecicURI,themainstreamenduserwillnothavethedomainortechnicalknowledgetodoso.
InkeepingwithtypicalWebusage,96A.
-S.
Dadzie,M.
Rowe,andD.
Petrelliausersuchasthoseinourscenarios(seesection4.
1)mayhaveobtainedastartaddressfromaierorviaanaturallanguagequeryinawebsearchengine.
FromthispointtheywillstarttoexploretheWoD;whetherthisistheReadableortheSemanticWebshouldbetransparenttothemainstreamuser.
EndusersshouldbegivenusabletoolstoexploretheWoD,linkingtorelevantLDorotherdatainthewiderWeb,andaccesstosimplemethodsforexportingthedatatoalternativereadableformats.
Forthetech-savvyuseritisalsousefultoallowextractionoftheunderlyingRDFdatausingformalsyntaxsuchasSPARQL.
Forsuchinteractiontooccur,wehaveidentiedkeyusabilitychallengeswhichcurrentlyexistinusingLD:2.
1ChallengestoLinkedDataUseExplorationstartingpoint:wheretostart;existingLDbrowsersassumetheenduserwillstartbrowsingfromaspecic,validURI.
Howcanavisualisa-tionstartingpointbepresentedtousersinsuchawaythatitismeaningfulCombatinginformationoverload:presentingenduserswithalltheproper-tiesofagivenresource,alongwiththerelationsthroughwhichtheresourceislinkedtootherentities,leadstoinformationsaturationandadenseinfor-mationspace.
HowcanwepresentthisinformationinamorelegibleformReturningsomethinguseful:RDFisthestaplerecipeforresourcedescrip-tions,returninginformationusingthisknowledgerepresentationformatin-hibitscomprehension.
HowcanRDF,andtheinformationcontainedwithininstancedescriptions,berepresentedinamorelegible,manageableformEnablinginteraction:endusersarefamiliarwiththemakeupoftheWebanditsbrowsablenature.
IsitpossibletoreplicatesuchfamiliaritywhichusersexperiencewhenbrowsingtheWWWontheWoD3RelatedWorkTheSWcommunityhastodatelargelyfocusedontheuseoftext-basedrepre-sentationsoflinkeddata,often(explicitlyortransparently)throughtheuseofSPARQLendpoints.
Thisisduetotwomainreasons:(1)theinfancyoftheLDinitiative;(2)thefocusonprototypestoservetheneedsoftheSWcommunityandotherspecialiseddomains,togenerateandanalyseLDandrelatedSWdata.
ThemostwidelyreferencedLDbrowsersincludeSig.
ma[20],Marbles[3]andURIBurner5.
Simile6providesaccesstoasetoftoolsandAPIs(ApplicationProgrammingInterfaces)forpresentingandinteractingwithRDFdata.
Haystack[15]wasoneoftherstSWbrowsersdeveloped,tolowerthebarriertoconsumptionofSWresources.
Itusesstylesheetsandviews/lensesdenedinRDFtoaggregatedistributedinformationandcustomiseitspresentationforspecicusersandtasks.
Corlosquetetal.
,[6]describetheextensionofDrupal75URIBurner:http://linkeddata.
uriburner.
com6Simile:http://simile.
mit.
edu7DrupalContentManagementSystem:http://drupal.
orgHidetheStack:TowardUsableLinkedData97tocreateenrichedwebsites,whosemodelsaredenedusingstandardontologies,inordertoembedRDFintotheunderlyingcontent.
TheirapproachconsequentlyenablestheretrievalalsoofrelevantLDonthey,viaSPARQLendpoints.
TextualpresentationofRDFdata,whilefamiliarandusefultoSWexpertusers,resultsinhighcognitiveloadfornon-technicalandnon-domainexperts,andevenforexperts,asdataamountandinterconnectivityincrease.
Visualisa-tionisacknowledgedtoenhanceknowledgediscoveryandanalyticalabilitywhileloweringcognitiveload,byharnessingpowerfulhumanperception[10,11,17]toenableintuitiveconstructionofanunderstandingofthestructureoflargeamountsofcomplex,interactingdata.
Visualisationaordsanumberofeverydaymetaphors,byencodingdataattributesinto,e.
g.
,graphs,mapsandtimelines,providingmoreunderstandableuserinterfaces(UIs)overmachine-friendlyRDF.
TheLDcommunityis,notsurprisingly,examiningvisualsolutionsforbothtech-nicalandmainstreamuse.
TheseoftenvisualisetheRDFgraphstructure,e.
g.
,[7,9,13,18];suchgraphsareveryusefultotechnicalexperts,whoserequirementsincludeinspectiontoidentifyerrors,retrieveselecteddataandrelations,anddevelopspecialisedapplications.
Awell-knownRDFvisualisationtoolisIsaViz[13],whichoverlaysRDFgraphswithstylesheetsbasedonFresnellenses[14].
AsmallnumberofLDvisualisationbrowsersincludeRelFinder[7],whichautomateslinkdiscoverybetweenuser-speciedresources.
Tabulator[5]inter-pretsLDbymappingtostandardontologiessuchasFOAF8(FriendofaFriend),withoutputtoanestedhierarchyandvisualisationsincludingmapandcalendarviews.
RDFScape[18]isaCytoscape9plug-intoenhanceanalysisinBioinfor-matics,visualisingtheresultsofontology-basedinferencewithnode-linkgraphs.
However,thereisadearthinapplicationstargetedatmainstream,non-technicaluse.
Hirschetal.
,[9]helptollthisgap,byvisualisingthe(semantic)knowledgecontentinFreebase10andWikipedia11usingnode-linkgraphs.
Theyuseicons,colourandrelativenodeandedgesizetoencodedataattributes,anddraw(labelled)linksbetweenclustersofrelatedsemanticinformation.
LESS[2]supportsthecreationof(shareable)web-basedtemplatestoaggregateanddis-playLDtomainstreamusersthroughthefamiliarpresentationmethodsofthereadableWeb.
DBPediaMobile[3]exploitsageographicalmetaphortolinkandpublishinformationaboutresourcesintheuser'svicinity.
ItfurtherlowersthebarriertointeractionwithLDbyenablingthepublicationofnewinformationaboutresourcesinusers'physicallocationtotheLDcloud.
Ourreviewofthestateoftheart,withrespecttothechallengesoutlinedinsection2.
1,highlightstheworktobedonetomakeLDusabletonon-techsavvyendusers.
Inparticularwenotetheneedforsolutionswhichallowmoreexible,open-endedknowledgediscovery.
RelFinderandDBPediaMobilecomeclosesttofulllingthis,byrevealing,usingdierentmechanisms,relationsbe-tweendistinctentities.
Additionallywenotethatlowuserfamiliaritywithlinked8FOAFOntologySpecication:http://xmlns.
com/foaf/spec9Cytoscape:http://www.
cytoscape.
org10Freebase:http://www.
freebase.
com11Wikipedia:http://en.
wikipedia.
org98A.
-S.
Dadzie,M.
Rowe,andD.
Petrellidatasets–neithertheintricaciesofthedata,noritsentiresubjectscope–hinderstheformulationofinitialquestions.
Insuchcasesaviablemeansofexplorationisrequired,muchasWebbrowsingfunctions,toallowintuitiveinformationdiscov-ery.
Inturn,achallengeofsuchdiscoverymechanismsisinformationoverload.
VisualisationisessentialinprovidinganentrypointintotheWoDandcom-batingthecognitiveloadassociatedwiththeuseoftheselarge,distributed,highlyinter-connecteddatasets.
Howevervisualisationhasitsownlimits.
Insection4weexplorethesynergiesbetweenhigherleveldataoverviewsandde-tailed,interactiveanalysisofROIs.
Whatconstitutesausefuldetailviewhowevervariesdependingontheenduserandtheirtask;whilewefocusonthedesignofUIsthatsupportmainstreamenduserswearecarefulnottoignoretheexpertuser.
Thispaperaimsalsotoaddressthislastchallenge–graphicaloverviews,forinstance,allowuserstoobtainahighlevelunderstandingofdatastructure.
Themainstreamendusercanusethesetoidentifyastartingpointfortheirexploration,whiletheexpertwillquicklyrecognisevalidpatternsandclustersinadditiontoanomaliesinthedatastructure.
4Template-BasedVisualisationofLinkedDataDierencesinusers,theirtasksandenvironmentsmeannoonesolutioncanclaimtoexhaustivelymeetallrequirementsforusingLD.
However,template-basedapproachesholdsignicantpotentialforhelpingtobridgethesechallenges.
Templatesareauseful,exibletoolthatmaybeusedtodenehowtoformatRDFdataintoahuman-readablerepresentation[2,6,14,15,16],andsynthesiserelatedbutdistributed,heterogeneousinformation[3,9,18,20],improvingman-agementofitsknowledgecontent.
Inthissectionwediscussourdesignrationaleforatemplate-basedvisualisationapproachtopresentinglinkeddata,asolutionthattacklesthechallengesidentiedinsection2.
1.
Toprovidesucientcontexttoillustrateourwork,wedetailtwoscenarios:4.
1ScenariosofUsea.
AprimaryschoolteacherinSheeldispreparingatechnologyprojectforyear6pupilsthatexaminesthefutureoftheWeb.
TohelpherpickatopicthatwillengageherstudentsshewishestospeaktoresearchersinthelocaluniversityexploringleadingedgeWebtechnology.
ShegoestotheUniversityofSheeldwebsiteandnavigatestotheDepartmentofComputerScience.
Shendsalinktotheresearchareas.
.
.
b.
Annehasjustcompletedher'A'levelsandreceivedagranttostudyComputerScience(CS)attheUniversityofSheeld.
Herparentsknowthatshewishestoworkasalectureraftershegraduates.
Theywanttoreassurethemselvesthatshewillbeexposedtoawiderangeoftopicsandbeabletointeractwithacademicsattheforefrontoftheirelds.
ThereforetheygotheUniversitywebsiteandnavigatetotheDepartmentofComputerScience.
.
.
Intheremainderofthepaperwewillcompletebothstories,illustrating,withData.
dcs,howoursolutionhelpstheseactorsretrievetheinformationtheyseek.
HidetheStack:TowardUsableLinkedData994.
2DesignRationaleandImplementationOursolutiontovisualisinglinkeddataistohidethestack;bythiswerefertoutilisingSWtechnologiesinherentinLD,butwiththeseelementstransparenttotheuser.
Inshort,thisbypassesthecomplexityassociatedwithRDFandSPARQLbyrenderinginformationreturnedusingsuchstandardsinahuman-legibleform.
Theproposedsolutionmakesuseofcustomtemplatesloadedbasedontherdf:typeoftheresourcebeingviewedbytheuser.
Twoviewsareprovided:agraphviewthatdrawstherelationsbetweenresources,displayinginformationdenedusingobjectproperties;andaco-ordinateddetailviewthatpresentsattributesandinformationdenedusingdatatypeproperties.
Oursolutionallowsthecollapsingofdetailintocompoundnodes,revealingdetailonlyforthefocus.
Moreadvancedoptionsarealsoavailabletothetech-savvyuser,foradvancedqueryingthatlinksdirectlytothewiderWoDusingpublicSPARQLendpoints.
Werstdescribeourdesign,andhowwetranslatedthisintothepresenta-tionofinformationusingtemplates.
Wethensummarisetheresultsofafocusgroupstudydesignedtocollectuseropinionsaboutthecapabilityenabledforexploratoryknowledgediscoveryanddirectedsearch.
TheTemplatesTherststepinourtemplatedesignistoidentifywhichresourcetypesinagivendatasetmaybeofinteresttotheenduser.
Atthesametimeasolutionisrequiredthatdoesnotassumemorethancursoryknowledgeofdatastructureorcontent.
WellspeciedLDshould,asfarasispossible,re-useand/orextendstandardontologiesandvocabularieswhendescribingdatacontent.
Coreconceptsfromsuchontologiesandvocabulariesarethereforeanidealrstpointfromwhichtocreatereusabletemplates,adaptabletousercontextandtasks.
Weillustratethisformetadatathatdescribespeopleandinformationaboutpeople,startingfromtheorganisationstheyworkfor.
Examplesofwidelyappli-cableontologies,fromwhichweselectrelevantclassesandpropertiesorrelationstodenetemplates,include:FOAFtodescribepeopleandtheirrelationswithotherpeopleandorganisations;SWRC(SemanticWebforResearchCommuni-ties12),relevanttotheusecaseswediscuss.
Further,SWRCiseasilygeneralisedtootherorganisationalstructures;BibTEXtodescribepublications;PRV,theProvenanceVocabulary13,tosupportvericationofdatacontentandquality.
Theseexamplesbothprovideasolutionandraiseanotherissue–Organiza-tionandPerson,forinstance,aredenedinbothFOAFandSWRC.
Fortheinitialversionsofthetemplateswechosetomodelconceptsusingthemostwidelyreferencedontologyorvocabulary,toincreasereusability.
Soourtemplatesde-faultto,e.
g.
,FOAFforOrganizationandPerson,butusebothmemberfromFOAFandaffiliationfromSWRC,tomodeldirectedrelationshipsbetweenanOrganizationandaPerson,respectively.
Documents(ofwhichPublicationsareasubset)aremodelledusingBibTEXratherthanFOAForSWRC,with12SWRCOntologySpecication:http://ontoware.
org/swrc13PRVCoreOntologySpecication:http://purl.
org/net/provenance/ns#100A.
-S.
Dadzie,M.
Rowe,andD.
PetrellitherelationbetweenaPersonandabibtex:EntrymodelledusingtheFOAFpropertymaker.
Inordertoapplythetemplatesalsotoontologiesthatredenecommonlyreferencedconceptsweplantoincludeselectorsthatallowcascadingtocaterforredundancy,inadditiontoequivalentclassesacrossontologies,andsubclassesdenedinasingleontologyorbyextensioninanewontology.
TosupportextensibilityandreusabilitywedeneatemplateforeachkeyresourcerstusingFresnellensSPARQLselectors(seetop,Fig.
5).
WethentranslatethesetoasetofpresentationformatsinaJavaprototype,usingthedesignideasexpressedinFig.
2.
ThreetemplatesthatprovidecustomisedviewsoverRDFdataareillustratedinthispaper,usingtheData.
dcslinkeddataset,fortheresourcetypes:Group(Organisation):Person:Publication:.
(a)Name:Organisations,InformationandKnowledgeGroupWeb:http://oak.
dcs.
shef.
ac.
ukMembers:FabioCiravegnaMatthewRoweName:MatthewRoweWeb:http://www.
dcs.
shef.
ac.
uk/~mrowePapers:InterlinkingDistributedSocialGraphs,MRoweLinkedDataontheWebWorkshop,WWW2009GettingtoMe-ExportingSemanticSocialNetworkInformationfromFacebook,MRoweandFCiravegna.
SocialDataontheWebWorkshop,2008Title:InterlinkingDistributedSocialGraphsAuthors:MatthewRowePublishedat:LinkedDataontheWebWorkshopYear:2009Title:GettingtoMe-ExportingSemanticSocialNetworkInformationfromFacebookAuthors:MatthewRoweFabioCiravegnaPublishedat:SocialDataontheWebWorkshopYear:2008Name:FabioCiravegnaWeb:http://www.
dcs.
shef.
ac.
uk/~fabioPapers:GettingtoMe-ExportingSemanticSocialNetworkInformationfromFacebook,MRoweandFCiravegna.
SocialDataontheWebWorkshop,2008swrc:afliationswrc:afliationfoaf:madefoaf:madefoaf:made(b)Fig.
2.
Adesignsketch(2a)illustratestheuseoficonstoencodegraphnodes,basedonrdf:type.
Linksshowprimaryrelationships,e.
g.
foaf:Personfoaf:madebib:Entry.
Interactionwiththegraph,e.
g.
onNodeClick,revealsagroupdetailtemplate(left);aPerson(lower,right);aPublication(top,right).
2billustratespotentialimplementationasasub-graphofnodesexpandedtoshowdetail,withdirected,labellededges.
Returningtothescenarios,ourdesignaimstosupportespeciallythenon-expertuser(outsidetheSWandCSdomains).
Thistypeofuserwilltypicallyhave,atbest,abroadideaofwheretondtheinformationtheyseek.
Theyaremostlikelytoexhibitexploratoryinformationseekingbehaviour,movingontodirectedsearchoncetheyhaveamorecompleteunderstandingofwhatknowledgeisavailable.
ThescenarioshaveeachactoratthestartpageoftheCSdepartment,wheretheyarepresentedwith:traditional,text-basedbrowsingorvisualisation-basedbrowsingofthedepartmentalstructure.
Theychoosethevisualisationoverreadingtheresearchpagesonthewebsite,becausethesiteex-plainsthatthelatterdisplaystherelationshipsbetweengroupsandresearchers.
HidetheStack:TowardUsableLinkedData101Fromanimplementationpointofview,thatthevisualisationisbasedonData.
dcsis(andshouldbe)transparenttotheuser.
WecannotassumethatourtargetenduserwillhaveaspecicURItobrowsefrom(therootandinitialfocusoftheirvisualgraph).
Inthiscasetwooptionsareavailable:(1)presentingtheuserwithalistofpotentialstartingpoints,e.
g.
,forData.
dcs,researchgroupsortheirheads–thisishoweveronlyfeasibleforarestricteddataset,whichisseldomthecaseforLD;(2)randomselection,restrictedtokeyresourcetypes.
Wecurrentlyusethesecondoption,rootingthe(directed)graphwiththerstsubjectreadthatmatchesoneofthepre-speciedresourcetypesofinterest–intheData.
dcsexamplethiswouldbeagroup/aliation,apersonorapublication.
TheGraphViewSimplydrawingthegraphshowingtherelationswithinData.
dcsresultsinarepresentationsuchasthatinFig.
3a.
Toreducethepotentialforcognitiveoverloadthegraphdisplaysonlythersttwolevelsfromtherootatstart-up,leavingtheusertounfoldthegraph,uptoitsleaves.
Thishoweverquicklyresultsinadensewebofnodes,aproblemcommontonode-linkgraphs[8].
ThisisseenevenwithreducedocclusioninFig.
3a,byabbreviatingresourceURIs,andwithonlypartofthegraphdrawn,illustratingexpansionfromthethirdtothefourth(outofeleven)levels.
Presentingthistousers,whethernon-expertorexpert,onlyhighlightsthedensityandcomplexityoftheinformationtheywishtoexplore;amoreusablesolutionisnecessary.
Weresolvethisissuerstbylteringthedataset,basedonkey(RDF)re-sourcetypes,identiedthroughtheuseofstandardtemplates,similartotheapproachinFresnel[14]:inthiscase,(research)group(Organisation),Personandpaper(Publication).
The8000statementsinData.
dcsgenerate3000distinctgraphnodes;theltersreducethenodecountto300.
Thedetailforeachinstanceoftheseresourcesiscollapsedintowhatweshallrefertoasacom-poundnode.
Theresultinggraph,inFig.
3b,providesahighleveloverview,withsignicantreductioninocclusion,thatdisplayskeyinformationinthedata.
However,thissolutionpresentsyetanotherissue:theltershidewhatisuseful,detailedinformationfromtheuser.
Asdiscussedalsoin[8,17]thereistheneedtosupporttheexaminationofdetailinselectedROIs.
Weresolvethisbymaintainingthevisualgraphasthecentreoftheuser'sexplorationactivity,andcoupleitwithaviewthatformatsforhumanconsumptionthedetailforeachcompoundnode(seeFigs.
4and5).
ThisallowstheusertomaintainthecontextofsurroundinginformationevenwhileexploringROIsintheco-ordinateddetailwindow(whichwedescribeinthefollowingsub-section).
Colourcodingandnodesizearethetwomainvisualencodingmethodsforkeypropertiesinthegraph.
Organisationnodesandoutlinkshaveablueborder,andPersonandPublication,orangeandgreen,respectively.
Thefocusnodehasaredll,andthebordersofitsimmediateneighbourschangetoredonMouseHover.
Keywordquerymatchesarelledinpink.
Nodesizemaybeweightedbythenumberofoutlinks;thisisespeciallyusefulwhenchildnodesarefoldedintoaparent,orhiddenusingalter.
In,e.
g.
,Fig4,groupmemberswithrelatively102A.
-S.
Dadzie,M.
Rowe,andD.
Petrellifocusnode(a)focusnode(b)Fig.
3.
3ashowsthelayoutfortherstthreelevelsofthecompletegraphforData.
dcs,asitisbeingexpandedtothenextlevel.
3bshowstheequivalentgraphfortherstfourlevelsthatcollapsesdetailintocompoundnodesusingacustomlter.
highpublicationcountstandout.
Thesmallestgroupnodehasthesmallestsize(comparingheightandnodefontsize;widthisdeterminedbylabellength).
Theactorsinourscenariosbothchoosetoexplorebyresearchgroup.
Anne'sparentsaresimplybrowsingtogetagoodfeelforthevarietyanddepthoftheresearchcarriedout.
Theschoolteacherhasamorecomplexgoal,toiden-tifywhichgroupismostlikelytobeabletoprovideherwithinformationforhertechnologyproject.
Wewillconcentratefornowonthelatter.
MovingfromFig.
3bsheelectstodisplayonlyOrganisationandPersonnodes(lters–top,right,Fig4),toremovetheclutterofthelargenumberofPublications,asthesearecurrentlyofsecondaryinterest.
Shethenexpandsthelteredgraphtodis-playalllevels,showninthelargerpaneinFig4.
Theoverallstructureofthedepartmentiseasilydiscerned.
Distinctclustersdierentiateeachgroup,whilelinksbetweenthemduetoresearchersinmultiplegroupsspanthespacebetweenclusters.
Aquickscanofgroupnamesreturns:(1)ComputationalSystemsBiology(CompBio),(2)MachineLearning(ML),(3)NaturalLanguageProcessing(NLP),(4)Organisations,InformationandKnowledge(OAK),(5)SpeechandHearing(SpandH)and(6)VericationandTesting(VT).
Theteachereliminatesallbutone:OAK.
Graphvisualisationisabletogiveahighleveloverviewofsuchlarge,interlinkeddatasets.
However,thegraphsarenotaseectivefordetailedknowl-edgeexploration.
Wedescribenexttherationalebehindthetemplatedesignusedtoprovidehuman-readabledetail,whichtheteacherwillusetodelveintothegroup'sresearch,todetermineifitisrelevanttoherproject.
HidetheStack:TowardUsableLinkedData103ElizabethAmparoCanoBasave,MatthewRowe.
MetasocialWikiAba-SahDadzie,ElizabethAmparoCanoBasave,MatthewRowe,MatthewRowe.
ApplyingSemanticSocialGraphstoDisambiguateMatthewRowe,JonathanButters.
AssessingTrustMatthewRowe,JoseIria.
LearningtoClassifyIdentityWebRefereMatthewRowe.
InterlinkingDistributedSocialGraphsMatthewRowe.
TheInteroperabilityofLightweightSemanticsforOrganisations,InformationandKnowledgehttp://data.
dcs.
shef.
ac.
uk/person/Matthew-RoweMatthewRowehttp://www.
dcs.
shef.
ac.
uk/~mroweNameWebAfliationPublicationsNodeWeightingTreeDepth>Aba-SahDadzieElizabethAmparoCanoBasaveAlfonsoSosaOrganisations,InformationandKnowledgeGrouphttp://oak.
dcs.
shef.
ac.
ukMembershttp://data.
dcs.
shef.
ac.
uk/group/oakNameWebfocusnodeWebNameFig.
4.
Filteringoutpublicationshighlightslinksacrossgroupsinthedepartmentalstructure.
Thefocusnode,http://data.
dcs.
shef.
ac.
uk/person/Matthew-Rowe,ishighlightedinred,anditsdetailshowninthetemplatewindowontheright.
ThegrouptemplateforthisPersonispopulatedandsuperimposedonthegraph.
TheDetailViewTheteacherbrowsestheinformationabouttheOAKGroupbyclickingonnodesformembers(inthegraph).
Thisupdatesthecoupleddetailtemplatewindow,allowinghertoexamineeachPersonresourceinmoredetail.
Alternatively,shecouldselecttheOAKResearchGroupnode,topopulateanddisplaytheOrganisationdetailtemplate(overlay,Fig.
4),andswitchtobrowsethedetailforitsmembersviathethumbnailslinkingtoeachPersontemplate.
Themember"MatthewRowe"listsasetofpublicationsthatmayberele-vant(rightpane,Fig.
4);theteacherselectsone,"ImprovingSupportforWeb-basedVisualAnalysisofSocialGraphs",toexamineingreaterdetail–thisisthehuman-readablelabelfortheURIdata.
dcs:paper/4169.
Fig.
5showstheSPARQLquerytemplate(basedontheFresneltemplateforthisresource)usedtoretrievethefullviewforapublication,andtheresultofapplyingthistem-platetodata.
dcs:paper/4169.
ThisisthesamenodehighlightedinFig.
3;thedierenceinabilitytointerpretwhatthenoderepresentscanbeclearlyseen.
Theteacherreducesthedepthofthegraphtotwolevels,removesthePublica-tionlter,andzoomsintothedetailforthethreeco-authors(bottom,Fig.
5).
Thisallowshertoscan(inthegraph)thetitlesofotherpublicationswrittenbytheauthors,whilemaintainingthefocusonthepublicationofinterestinthedetailtemplatewindow(right).
Armedwiththisinformationshegoestothewebpagesofeachoftheco-authors,viathePersondetailview.
Shewillalsobrowsethegroup'swebpages,toextractmoreinformationabouttheirprojects.
104A.
-S.
Dadzie,M.
Rowe,andD.
PetrelliSPARQLquerytemplateforthefullpublicationviewformockURIPREFIXfoaf:PREFIXbib:SELECTDISTINCTpublicationTitleyearbookTitlepersonUriauthorimageUriWHERE{bib:titlepublicationTitle;bib:hasYearyear;bib:hasBookTitlebookTitle;foaf:makerpersonUri.
personUrifoaf:nameauthor;foaf:imgimageUri}ORDERBYDESC(year)publicationTitleImprovingSupportforWeb-basedVisualAnalysisAba-SahDadzieElizabethAmparoCanoBasaveMatthewRoweIEEEVisWeekWorkshop2009http://data.
dcs.
shef.
ac.
uk/paper/4169Publishedat:Year:Authors:Title:Fig.
5.
ThedetailviewforthePublicationhttp://data.
dcs.
shef.
ac.
uk/paper/4169isshownontheright,whilethegraphinthecentreiszoomedintoshowother(Publication)linksfromthePersonresourcesofinterest,itsco-authors(ormakers)5FormativeEvaluationToevaluateourapproachafocusgroupstudywascarriedoutto:(1)reviewtherequirementsforapplicationsforconsumingLD;(2)determineiftheco-ordinatedviewsallowbothmainstreamusersandtechnicalanddomainexpertstoobtainagoodunderstandingofdatacontentandstructure;(3)identifywherethedesignrequiredrevisionpriortoformal,summative,usabilityevaluation.
Aspartofaconferencetutorial14tostudytheimportanceofHuman-ComputerInteraction(HCI)anduser-centredinterfacedesignwhenbuildingendusertoolsonSWtechnology,14participantsfromtheSWcommunityandrelatedresearchareasgavefeedbackontheuseoftheprototypeforinformationexplorationandretrievaltasks.
Theparticipants,whowouldbeconsideredtobetech-savvy,as-sumedtheroleofexpertreviewers(see[12,19])inspectingtheinterfacefromthepointofviewofthetargetenduser.
Atthisstageinthedesignweconsider,inadditiontothemainstreamuser,thetechnicaluserbuildingSWtools.
Suchusersneedsupportalsotoidentify:(1)incompletedata;(2)errors,e.
g.
,redun-dancyandincorrectlinkingduringautomaticLDgeneration;(3)keypropertiesofthedataandoptimalwaysforpresentingtheseandtheirinter-relationships.
Theparticipantsweregivenanoverviewoftheprototypeanditsdesignratio-nale,andchallengestoLDexplorationanduse.
Theythencarriedoutapractice14'EssentialHCIfortheSemanticWeb'@ESWC2010:http://www.
eswc2010.
orgHidetheStack:TowardUsableLinkedData105taskonasmalldatasetbasedonatelevisionseries,toextractstructuralin-formation.
Thiswasfollowedbyamorecomplexexplorationtask,toretrieveinformationfromData.
dcson,amongothers,collaborationacrossgroups,andresearcherswithmultiplealiationsandrelativelyhighpublicationcount.
Infocusgroupsofupto4,theparticipantsthenreviewedtheprototypeaspartofaparticipatorydesignactivity.
TheyassessedhowwelltheUIsupporteduserrequirements,basedonthreemaincriteria:(1)eectiveness,i.
e.
,howsuccessfultheywereindiscoveringtheinformationrequired;(2)eciency,i.
e.
,timetocompletetasks;(3)usersatisfaction,guidedbyaquestionnaire.
5.
1ResultsBecausetheevaluationwasformativethefocuswasoncollectingqualitativeinformationtovalidatetherequirementsthatfedintotheUIdesign,anddeter-minehoweectivelythishadbeentranslatedtosupportingLDconsumption.
Informationwasobtainedbyobservingtheparticipants,supplementedbyade-briefinwhichtheoutcomesofthefocusgroupstudywerediscussed.
Thisphasegeneratedanumberofpost-itnotesthatwereclusteredandanalysed.
Theparticipantsfound,overall,thegraphstobequiteexpressiveandeectiveingivingasenseofthedatadistribution.
Howeverthosenotaccustomedtographmanipulationhaddicultycontrollingthedisplay,duersttothelargegraphsize,andalsothe(spring)layoutalgorithmthataltersthelayouttore-centreonfocuschange.
Commentssuchas:"eventuallyyougotabigpictureofthedata"and"Ilikedthedirectmanipulationbutthegraphshouldstayput[whenIclick]",showbothfrustrationandappreciation.
Anotherkeyfeaturerecognisedwastheeectivenessindisplayingdetailforkeyresources:"inaneatandconciseway".
TheprototypewasseentohavepotentialforexploringanddebuggingLD.
Thecapabilityforkeywordsearchwasmuchappreciated.
Thiswasreinforcedbyrequestsformoreselectiveltering,bothnaveandexpert(e.
g.
,viaSPARQL).
Thoughthelattercouldbemotivatedbytheparticipants'technicalexpertise,itismorelikelytoindicatetheneedfortoolsforboth(usable)browsingandsearchtobeintegratedintothesameinterface.
Suggestionsforimprovinginteractionwiththeknowledgecontentincludedexplicithighlightingofsearchpathsinthegraph;browsingthroughpreviousactions(history);additionaloptionsfornavigation,e.
g.
,portingbetweendisconnectedsub-graphs.
6ConclusionsWehavepresentedasolutiontomakingLDusablebymainstreamendusers.
Thisseekstohidethestackfromsuchusers,enablinguseoftherichnetworkofinformationonoer,butwithoutrequiringknowledgeofelementsintheSWtechnologystacktoprocesstheinformation.
Forinstance,thetemplatesinourapproachare,inessence,SPARQLqueries,theboundvariableswithinwhicharetiedtoregionsintheinformationpresentationview.
TheuserneednotknowthatSPARQLisutilised,northattheresourcedescriptionisreturnedasRDF.
106A.
-S.
Dadzie,M.
Rowe,andD.
PetrelliWehighlightedfourkeychallengesforinteractingwithLDthatweaddressasfollows:theselectionofanexplorationstartingpointcanbeachievedvialistsofpotentialstartingpoints,allowingtheusertochoosetheresourcetofocuson;orbyrandomlyselectingaresourcefromwhichtostartbrowsing.
Forcombatinginformationoverloadweutilisetwotypesofviews,basedontemplates:agraphandadetailview.
Theformerprovidesanoverviewofthenetworkstructuresurroundingaresourceofinterest.
Ourdefaultdisplaysre-sourcesuptotwosteps(levelsinthegraph)fromthefocusontheWoD.
Thisprovidesaclearpicturethatdescribestherelationsbetweenexternalresourcesandpossiblepathsandtransitionsthroughthespace,whilepreventinginfor-mationoverload.
Thisisconsistentwithrecentworkin[7],whererelationsarerevealedbetweenDBPediaconceptsinagraph.
Oursecondviewdisplaysthepropertiesofthefocusresource,similartoworkby[2],byrenderingtheRDFresponsefromdereferencingtheURIintoalegibleform.
Bymarryingthetwoviewsweareabletoseparateoutthelogicalcomponentsofrelationsandprop-erties,whereweusethelattertoshowdatatypepropertiesoftheresource,andtheformertoshowthecontextandroleoftheresourcewithinthewiderWoD.
Theuseoftemplateshasalsoaddressedthechallengeofreturningsomethinguseful,byprovidinginformationinaformatthatendusersareabletounderstandandinterpret.
Colourcodingresourcesinthegraphviewbasedonrdf:type,forexample,enablestheusertoobserve(ratherthanhavingtoreadandinterpret)theconnectionsandrelationsbetweendierenttypesofinstances.
WeaddressthenalchallengeofenablinginteractionbymimickingthebrowsablenatureoftheWWWviaclickableregions,whichshiftsthefocustotheselectedresource.
Weplantocarryoutfurtherusabilityevaluationwithalargersampleandrangeofendusers,usingspecicproblemsolvingtaskstotesttheutilityandusabilityofourapproach.
Webelievethatthespecicationofsuchtaskswillprovidethecommunitywithbenchmarkexperimentsforvalidatingtheeective-nessofmethodsforvisualisingLD.
Ourtemplate-basedvisualisationapproachutilisesSPARQLqueriestoretrieveinformationfromaresource'sinstancede-scriptionandpresentthisinalegiblemanner.
Wehavedetailedthreestatictemplateswhichdemonstratethisfunctionality;ourfutureworkwillenabletheconstructionoftemplatesbyendusers,inessencecreatingSPARQLqueriesinanimplicitfashion.
Providingenduserswithmorecontrolwillalsobeexplored,toallowfortherestrictionofselectedpropertyvalueswhenexploringtheWoD,andlteringthroughonlycontentrelevanttotheircurrentactivity.
Finally,weplantoincorporateexistingworkintheeldofontologymapping,toenabletemplatestobeloadedforinstancesofclassesthataredenedasbeingequiv-alenttoothersforwhichtemplatesexist,e.
g.
,loadinganexistingtemplateforaninstanceofanotherclassthatisdenedtobeequivalenttofoaf:Person.
Acknowledgements.
A.
-S.
DadzieisfundedbytheEuropeanCommission(EC)7thFrameworkProgramme(FP)projectsSmartProductsandWeKnowIt(grants231204,215453).
M.
RoweisfundedbyWeGov(ECFP7248512).
A.
-S.
DadzieandD.
PetrelliwerepreviouslyfundedbyX-Media(ECFP6026978).
HidetheStack:TowardUsableLinkedData107References1.
Aroyo,L.
,etal.
(eds.
):ESWC2010.
LNCS,vol.
6089.
Springer,Heidelberg(2010)2.
Auer,S.
,Doehring,R.
,Dietzold,S.
:LESS-template-basedsyndicationandpre-sentationoflinkeddata.
In:Aroyo,L.
,etal.
(eds.
)[1],pp.
211–2243.
Becker,C.
,Bizer,C.
:ExploringthegeospatialsemanticwebwithDBpediaMobile.
JournalofWebSemantics7(4),278–286(2009)4.
Berners-Lee,T.
:Semanticweb–XML2000(2000),http://www.
w3.
org/2000/Talks/1206-xml2k-tbl5.
Berners-Lee,T.
,Hollenbach,J.
,Lu,K.
,Presbrey,J.
,Prud'ommeaux,E.
,Schraefel,mc.
:Tabulatorredux:Browsingandwritinglinkeddata.
In:LinkedDataontheWebWorkshopatWWW'08(2008)6.
Corlosquet,S.
,Delbru,R.
,Clark,T.
,Polleres,A.
,Decker,S.
:ProduceandconsumelinkeddatawithDrupal!
In:Bernstein,A.
,Karger,D.
R.
,Heath,T.
,Feigenbaum,L.
,Maynard,D.
,Motta,E.
,Thirunarayan,K.
(eds.
)ISWC2009.
LNCS,vol.
5823,pp.
763–778.
Springer,Heidelberg(2009)7.
Heim,P.
,Lohmann,S.
,Stegemann,T.
:Interactiverelationshipdiscoveryviathesemanticweb.
In:Aroyo,L.
,etal.
(eds.
)[1],pp.
303–3178.
Herman,I.
,Melanon,G.
,Marshall,M.
:Graphvisualizationandnavigationininformationvisualization:Asurvey.
IEEETransactionsonVisualizationandCom-puterGraphics6(1),24–43(2000)9.
Hirsch,C.
,Hosking,J.
,Grundy,J.
:Interactivevisualizationtoolsforexploringthesemanticgraphoflargeknowledgespaces.
In:WorkshoponVisualInterfacestotheSocialandtheSemanticWebatIUI2009(2009)10.
Keller,T.
,Tergan,S.
-O.
:Visualizingknowledgeandinformation:Anintroduc-tion.
In:Tergan,S.
-O.
,Keller,T.
(eds.
)KnowledgeandInformationVisualization:SearchingforSynergies,pp.
1–23(2005)11.
Kerren,A.
,Stasko,J.
T.
,Fekete,J.
-D.
,North,C.
(eds.
):InformationVisualization:Human-CenteredIssuesandPerspectives.
Springer,Heidelberg(2008)12.
North,C.
:Towardmeasuringvisualizationinsight.
IEEEComputerGraphicsandApplications26(3),6–9(2006)13.
Pietriga,E.
:Semanticwebdatavisualizationwithgraphstylesheets.
In:SoftVis'06:Proc.
2006ACMSymposiumonSoftwareVisualization,pp.
177–178(2006)14.
Pietriga,E.
,Bizer,C.
,Karger,D.
R.
,Lee,R.
:Fresnel:Abrowser-independentpresentationvocabularyforRDF.
In:Cruz,I.
,Decker,S.
,Allemang,D.
,Preist,C.
,Schwabe,D.
,Mika,P.
,Uschold,M.
,Aroyo,L.
M.
(eds.
)ISWC2006.
LNCS,vol.
4273,pp.
158–171.
Springer,Heidelberg(2006)15.
Quan,D.
A.
,Karger,R.
:Howtomakeasemanticwebbrowser.
In:WWW'04:Proc.
13thInternationalConferenceonWorldWideWeb,pp.
255–265(2004)16.
Rutledge,L.
,vanOssenbruggen,J.
,Hardman,L.
:MakingRDFpresentable:inte-gratedglobalandlocalsemanticwebbrowsing.
In:WWW'05:Proc.
14thInter-nationalConferenceonWorldWideWeb,pp.
199–206(2005)17.
Shneiderman,B.
:Theeyeshaveit:ataskbydatatypetaxonomyforinformationvisualizations.
In:Proc.
IEEESymposiumonVisualLanguages,pp.
336–343(1996)18.
Splendiani,A.
:RDFScape:Semanticwebmeetssystemsbiology.
BMCBioinfor-matics9(suppl.
4),S6(2008)19.
Tory,M.
,Moller,T.
:Evaluatingvisualizations:doexpertreviewsworkIEEEComputerGraphicsandApplications25(5),8–11(2005)20.
Tummarello,G.
,Cyganiak,R.
,Catasta,M.
,Danielczyk,S.
,Delbru,R.
,Decker,S.
:Sig.
ma:Liveviewsonthewebofdata.
JournalofWebSemantics8(4),355–364(2010)