GizeATimeWarpintheWebofDataValeriaFionda1MelisachewWudageChekol2and - PDF document

1 Gize:ATimeWarpintheWebofDataValeriaFiond
Gize:ATimeWarpintheWebofDataValeriaFionda1,MelisachewWudageChekol2,andGiuseppePirro31UniversityofCalabria,Italyfionda@mat.unical.it2UniversityofMannheim,Germanymel@informatik.uni-mannheim.de3InstituteforHighPerformanceComputingandNetworking,ICAR-CNR,Italypirro@icar.cnr.itAbstract.WeintroducetheGizeframeworkforqueryinghistoricalRDFdata.Gizebuildsupontwomainpillars:alightweightapproachtokeephistoricaldata,andanextensionofSPARQLcalledSPARQ{LTL,whichincorporatestemporallogicprimitivestoenablearichclassofqueries.OnestrikingpointofGizeisthatitsfeaturescanbereadilymadeavailableinexistingqueryprocessors.1IntroductionQueryinghistoricaldataisofutmostimportanceinmanycontexts,fromcitymonitoring,whereoneneedstotrackdierentaspects(e.g.,pollution,popula-tion)togenericexploratoryresearch,whereoneisinterestedinposingquerieslike\Retrieveplayersthatarenowmanagingsomeclubtheyplayedfor"or\Re-trievetheannotationofagenesincethediscoveryofaparticularinteraction".TheclassicalapproachforqueryingRDFdata(e.g.,viaSPARQLendpoints)onlyconsidersthelatestversion.Infact,queryingofhistoricalRDFdataposessomechallenges.Therstconcernstherepresentationandstoringofhistoricaldata.Someapproaches(e.g.,[4,7])allowtoretrievedatabyprovidingtimes-tamps.Otherresorttodedicatedindexingstructures(e.g.,[5])tospeed-upqueryprocessing.Thesecondproblemconcernswhattypeofqueryingprimitivetopro-vide.ThemostcommonapproachistodeviseSPARQLextensionsthatworkwithintervalsorSPARQLtranslationsintotemporallogic.Theadditionofad-hoccomponentseitherintermsofdatarepresentation,querylanguageorboth,hinderstheapplicabilityonexistingRDF(query)processinginfrastructures.TocopewiththeseissueswepresenttheGize4framework.Gizeisbuiltaroundtwomaincomponents.TherstisalightweightapproachtostoreRDFdata,whereeachversionofthedataisstoredinaseparatenamedgraph.ThesecondcomponentisapowerfulextensionofSPARQLcalledSPARQ{LTL.ThislanguageinheritsavarietyoftemporaloperatorsfromLinearTemporalLogics(LTL)[2].ToevaluateSPARQ{LTLonexistingSPARQLprocessorswedeviseatranslationtostandardSPARQLqueries. 4Gize( )istheEthiopianwordfortime. RelatedWork. ApproachesliketheDBpediaWaybackMachi

2 ne[4]allowtore-trievedataatacertaintimes
ne[4]allowtore-trievedataatacertaintimestamp(providedbytheuser).Otherapproaches(e.g.,[7])accesshistoricaldatavia(HTTP)contentnegotiation,typicallyusingtheMementoframework.Yetotherapproaches(e.g.,[5])introducetimestampsintoRDFtriplesalongwithad-hocindexingstrategies.InDescriptionLogics,someproposalsfocusontemporalconjunctivequeryanswering(e.g.,[1]);othereorts(e.g.,[6])havefocusedonthetranslationofSPARQLintoLTL.Sur-prisingly,thedesignofsolutionsforqueryinghistoricalRDFdataonexistingSPARQLprocessorsisstillinitsinfancy.GizellsthisgapbycontributinganextensionofSPARQL,calledSPARQ{LTL,thatallowsforarichclassoftemporalprimitivesborrowedfromLTL(e.g.,SINCE,NEXT,PREVIOUS)alongwithatranslationfromSPARQ{LTLqueriesintostandardSPARQLqueries.2TheGizeFrameworkRepresentingHistoricalRDFData.ThetenetofGizeistoenablequeryingofhistoricalRDFdataonexistingprocessors.Torepresentversions,Gizelever-agesthenotionofRDFquad.AnRDFquad(forsimplicity,weomitbnodes)isatupleoftheformhs;p;o;ci2II(I[L)I,whereI(IRIs)andL(literals)arecountablyinnitesets.Theforthelementofthequadrepresentsthatnamedgraphtowhichthetriplebelongs. Fig.1.AnexceprtofevolvingdatafromDBpedia.Fig.1showstheevolutionofsomedatatakenfromDBpedia.Eachofthe5versionsconsideredisrepresentedbyanamedgraph.FromthissmalldatasampleonemaynoticethatItalyhaschanged3coachesfromJuly2012(C.Prandelli)toJuly2016(G.Ventura).Interestingly,thelatestcoach(G.Ventura)willstartonJuly18th.OnemayalsonoticethatsomeplayerslikeA.Pirlowerepartoftheteamalongthewholeperiod,whilesomeotherlikeG.PazziniorM.Darmianiwereleftoutoradded,respectively.Thegurealsoshows(bottomrightcorner)updatestatisticsaboutPeopleinDBpedia.Eachpercentageiscomputedwrtthepreviousversion.Theavailabilityofhistoricaldataallowstoposequerieslike\Findallplayersthatplayedwiththehighestnumberofcoaches"or\FindplayersthatplayedsinceC.Prandelliwasthecoach".Mostofexistingapproacheseitherarenotabletoexpresssuchkindofqueriesorhavetoresorttoad-hocprocessinginfrastructures. TheSPARQ{LTLLanguage.ThesyntaxofSPARQ{LTLisshownbelowwhileTable1providesadescriptionofthetemporaloperators.LetqbeaSPARQ{LTLquery,H=fh1;h2;:::;hmgb

3 ethesetofversions,andhcbethecurrent(notn
ethesetofversions,andhcbethecurrent(notnecessarilythelatest)versionofthedata.QP::=QP1:QP2jfQP1gUNIONfQP2gjfQP1gMINUSfQP2gjQP1OPTIONALfQP2gjjGRAPHI[VQP1jQP1FILTERfRgjt=(I[V)(I[V)(I[L[V)jjXfQPgjWfQPgjFfQPgjGfQPgjfQP1gUfQP2gjjYfQPgjZfQPgjPfQPgjHfQPgjfQP1gSfQP2g Operator SPARQ{LTLSyntax Meaning Xq NEXT Evaluateqonversionhc+1 Fq EVENTUALLY Evaluateqonallversionshc;::::hm Gq ALWAYS Theevaluationofqmustbethesameonallversionshc;::::hm q1Uq2 UNTIL IfS2isthesolutionofq2inaversion,hk2fhc;::::hmgthen thereexistsasolutionS1ofq1onhcsuchthat S1iscompatiblewithS2inallversionsfhc;::::hkg Yq PREVIOUS Evaluateqonversionhc�1 Pq PAST Evaluateqonallversionsh1;::::hc Hq ALWAYSPAST Theevaluationofqmustbethesameonallversionsh1;::::hc q1Sq2 SINCE IfS2isthesolutionofq2inaversion,hk2fh1;::::hcgthen thereexistsasolutionS1ofq1onfhk;::::hcgsuchthat S1iscompatiblewithS2inallversionsfhk;::::hcg Table1.MeaningofthetemporaloperatorsinSPARQ{LTL.SPARQ{LTLallowstouseanadditionalsetofkeywordswhenwritingSPARQLqueries.SPARQ{LTLareevaluatedbytranslatingthetemporaloperatorsviaa(setof)pattern(s)evaluatedonnamedgraphsmaintainingdataversions.WegivesomeexamplesbyconsideringtheveversionofdatashowninFig.1(storedinseparatenamedgraphs).Inwhatfollows,dbp:INFTisashorthandfordbp:Italy_national_football_team.Example1.SelectplayerswhoareplayingintheItaliannationalfootballteamandplayedatleastunderadierentcoachthanthecurrentone. SPARQ{LTL TranslationintoSPARQL SELECT?pWHERE{dbp:INFTdbpo:name?p.dbp:INFTdbpo:coach?c1.PAST{dbp:INFTdbpo:name?p.dbp:INFTdbpo:coach?c2.FILTER(?c1!=?c2)}} SELECT?pWHERE{dbp:INFTdbpo:name?p.dbp:INFTdbpo:coach?c1.{GRAPH&#xhttp;&#x://g;&#xize.;&#xorg/;&#xv500;{dbp:INFTdbpo:name?p.dbp:INFTdbpo:coach?c2.FILTER(?c1!=?c2)}}UNION{GRAPH&#xhttp;&#x://g;&#xize.;&#xorg/;&#xv400;{dbp:INFTdbpo:name?p.dbp:INFTdbpo:coach?c2.FILTER(?c1!=?c2)}}UNION{GRAPH&#xhttp;&#x://g;&#xize.;&#xorg/;&#xv300;{dbp:INFTdbpo:name?p.dbp:INFTdbpo:coach?c2.FILTER(?c1!=?c2)}}UNION{GRAPH&#xhttp;&#x://g;&#xize.;&#xorg/;&#xv200;{dbp:INFTdbpo:name?p.dbp:INFTdbpo:coach?c2.FILTER(?c1!=?c2)}}UNION{GRAPH&#xhttp;&#x://g;&#xize.;&#xorg/;&#xv100;{dbp:INFTdbpo:name?

4 p.dbp:INFTdbpo:coach?c2.FILTER(?c1!=?c2)
p.dbp:INFTdbpo:coach?c2.FILTER(?c1!=?c2)}}} TheSPARQLqueryontherightisautomaticallygeneratedandcanbeeval-uatedonexistingprocessors.Notethatthetranslation(becauseofthesemanticsofPASTdescribedinTable1)requirestolookintoallversions. Example2.FindthenameofthecoachoftheItaliannationalfootballteamafterthesackingofCesarePrandelli. SPARQ{LTL TranslationintoSPARQL SELECT?nWHERE{PAST{dbp:INFTdbpo:coachdbp:CP.NEXT{dbp:INFTdbpo:coach?n.FILTER(?n!=dbp:CP)}}} SELECT?nWHERE{{GRAPH&#xhttp;&#x://g;&#xize.;&#xorg/;&#xv500;{dbp:INFTdbpo:coachdbp:CP.GRAPH&#xhttp;&#x://g;&#xize.;&#xorg/;&#x/v60;{dbp:INFTdbpo:coach?n.FILTER(?n!=dbp:CP)}}}UNION......UNION{GRAPH&#xhttp;&#x://g;&#xize.;&#xorg/;&#x/v10;{dbp:INFTdbpo:coachdbp:CP.GRAPH&#xhttp;&#x://g;&#xize.;&#xorg/;&#x/v20;{dbp:INFTdbpo:coach?n.FILTER(?n!=dbp:CP)}}}} Inthepreviousquery,dbp:CPisashorthandfordbp:Cesare_Prandelli.Asbefore,thetranslationofPASTmakesusageofUNIONqueriesovereachver-sionsvi;then,foreachvi,NEXTchecksinversionvi+1(viaaFILTER)thatthecoachchanged.3ConclusionsWehaveoutlinedGize,whichenablestoset-upaninfrastructureforquery-inghistoricalRDFdataonexistingSPARQLprocessors.Gizeadoptsasimpleapproachtostoredierentversionsofthedataandapowerfultemporalex-tensionofSPARQLcalledSPARQ{LTL.Asafuturework,weareconsideringapproacheslikeRDFHDT[3]toimprovethestoragespaceconsumption.References1.S.Borgwardt,M.Lippmann,andV.Thost.TemporalizingRewritableQueryLan-guagesOverKnowledgeBases.JWS,33:50{70,2015.2.G.D.DeGiacomoandM.Y.Vardi.LinearTemporalLogicandLinearDynamicLogiconFiniteTraces.InIJCAI,2013.3.J.D.Fernandez,M.A.Martnez-Prieto,C.Gutierrez,A.Polleres,andM.Arias.BinaryRDFRepresentationforPublicationandExchange.JWS,19:22{41,2013.4.J.D.Fernandez,P.Schneider,andJ.Umbrich.TheDBpediaWaybackMachine.InSEMANTICS,pages192{195,2015.5.S.Gao,J.Gu,andC.Zaniolo.RDF-TX:AFast,User-FriendlySystemforQueryingtheHistoryofRDFKnowledgeBases.InEDBT,pages269{280,2016.6.R.Mateescu,S.Meriot,andS.Rampacek.ExtendingSPARQLwithTemporalLogic.Report,2009.7.H.VandeSompel,R.Sanderson,M.L.Nelson,L.L.Balakireva,H.Shankar,andS.Ainsworth.AnHTTP-basedVersioningMechanismforLinkedData.201