Semantic technologies and linked data for the Italian PA: the case of data.cnr.it

0.94 MB·English

by  Aldo Gangemi

Semantic technologies and linked
data, with a case study
at the Consiglio Nazionale delle
Ricerche (CNR)
AldoGangemi
Web evolution and resource dereferencing
Webisevolvingfromawebofdocuments(oftencalledWeb1.0)toa
webofentities(called,withsubtledifferencesinmeaning,semantic
web,webofdata,Web3.0). Thisevolutionispassingalsothrough
theavailabilitytouserstoedititscontentsandgeneratecomplex
socialnetworksthroughsimpleinteractionparadigms(knownasso-
cialweborWeb2.0). Thisishappeningprimarilythankstoadeeper
exploitation of the Web architecture designed since the nineties,1
whichenablesthedereferencingandlinkingofwebresources(iden-
tifiedbymeansofaWebaddress),throughsimplecommunication
protocols(e.g. HTTP).Forexample,whenonewritestheaddress
(URI)http://www.cnr.it(thewebaddressoftheportalofConsiglio
Nazionale delle Ricerche, CNR) in a browser, the HTTP client of
the browser dereferences that address by communicating with a
serveratCNR,whichreturnsaHTMLpage,visualizedonitsturn
1http://www.w3.org/TR/webarch.
JLIS.it.Vol.4,n.1(Gennaio/January2013).
DOI:10.4403/jlis.it-5457
A.Gangemi,SemanticTechnologiesandLinkedData
bythebrowser. Otherwebpagescanbepresentinthevisualized
page, so creating a network of hypertextual links, which enables
thebrowsingexperience. Thisisbasicallythewebofdocuments.
Sometimesdereferencingisindirect,asinthecasewhenanaddress
representsacalltoadatabase,e.g. whenlookingforone’staxdata
intheAgenziadelleEntrate(theItalianfiscalauthority)website:
thisisstillthewebofdocuments,butthedocumentsaregenerated
outofaquerytoadatabase,whoseanswerisrenderedinHTML
byusingXMLstylesheets. ThecaseofWeb2.0isamoresophisti-
catedindirectdereferencing,whichalsoenablesdirectchangestoa
databaseperformedbyusers: applicationssuchasvoiceprotocols,
email,tagging,automaticloganalysis,opinionminingetc. converge
inrich,customizableanddynamicHTMLpages,asinthecaseof
Facebook.
Two difficult problems: identity and
semantic interoperability
Web1.0and2.0havetwolimitations,whichactuallyexistininfor-
mationsystemssincecenturiesago: identityandsemanticinterop-
erability. The identity issue arises e.g. in the following example.
AldoGangemihasdifferenthomepages(oneonhisinstitutesite,
ISTC-CNR,onefromthewikiofhislab,STLab,oneontheseman-
ticweb.org site etc. He is also registered on many other portals
providingservicestothecitizen,tomembersofassociations,confer-
encecommittees,commercialservicesetc. Moreover,hehasseveral
accountsofsocialwebapplications(e.g. Facebook, Gmail, Flickr,
iTunesetc.). Evenmore,AldoGangemiisadatumwithinpublicor
personaldatabases,likeGoogleScholar,DBLPetc.;thatdatumhas
identifiersthatareownedspecificallybythosedatabases,gathering
JLIS.it. Vol.4,n.1(Gennaio/January2013).Art.#5457 p.254
JLIS.it.Vol.4,n.1(Gennaio/January2013)
thenasortof”positional”identitywithinoneoftheirtables).Finally,
AldoGangemiiscitedinotherwebpages: articles,bibliographic
references,eventreports. Nowtheissueis: howcanweknowthat
(thephysicalorsocialperson)AldoGangemiistheentitydenoted
byhishomepages,registrations,accounts,databaseIDs,citations?
Intuitively, the issue is not limited to persons, but it impacts on
everythingthathasanidentity: places,organizations,products,ser-
vices,events,laws,ideas,concepts,fictionalthingsetc.Thesemantic
interoperabilityissue,besidespurelysystem-orientedproblems(e.g.
differentcomputationalplatforms),arisesfromtheidentityissue:
ifwecannotresolvetheidentityofsomethingacrossthedifferent
sourcesandsystemsthatrefertoit,itgetsreallydifficulttoaggre-
gate(i.e. assemble)andintegrate(i.e. appropriatelyconnect)the
informationaboutit. Thisisquitelimitingwhenconsideringthat
therelationsbetweensomethingandsomethingelsecanbesimilar
withindifferentsystems: therelationbetweenAldoGangemiand
theemailmessagesaddressedtohim,orbetweenhimandhisrecip-
ients,aresimilarinanyemailingsystem,butthosesystemsassign
different identities to the same persons, if any. In addition, each
systemworksonaproprietaryinfrastructure: differentlanguages,
formats, protocols etc. All this makes data integration between
differentsystemspartialinthebestcases.
Some traditional solutions
Inthelastyears,asortofcartelhasemergedbetweencommercial
serviceproviderssuchasFacebook,Googleetc.,inordertomake
socialnetworkdatainteroperable: thishoweverconcernsonlydata
exchangethatarecommerciallyinterestingforthosesystems,and
thirdpartyapplicationsthatcountonthem. Databasemanagement
systemsusecomplexprocedurestointegratetheirdatawhenitis
JLIS.it. Vol.4,n.1(Gennaio/January2013).Art.#5457 p.255
A.Gangemi,SemanticTechnologiesandLinkedData
required: schemaintegration,identityresolution,datawarehousing
etc. Eachprocessistypicallymadeadhoconapairofdatabases.
Partialsolutionsfordataintegrationalsocomefromdatamining
ornaturallanguageprocessingtechniques. Forexample,thereare
effectivestatisticalapproachesfornamedentityrecognitionandres-
olution,aswellasfordiscoveringsimilarityandindirectrelationsin
data. Documentannotationisanapproachthatcomesbackatleast
to the beginning of 20th century: a document, or part of it (para-
graphs,terms)areannotatedwithacategoryortagtakenfromsome
knowledge organization system: thesauri, classification schemes,
nomenclatures,controlledvocabularies,whichhavedevelopedin
mostscientific,library,andcommercialdisciplines. Exemplarcases
of similar large efforts include SnoMed, ICD, MeSH (medicine),
Gettythesaurus(culturalheritage),Agrovoc(agriculture)etc. Re-
cently,annotationproceduresareassistedeitherbycomputational
support for manual annotation, or by automatic annotation algo-
rithms(e.g. textclassification),withvariableprecision.
The web of data
In2006,TimBerners-Leeintroducedlinkeddata,asimpleandele-
gantmethod2torealizesomepracticaldataidentityintegrationand
interoperabilityontheWeb. Linkeddataareaimedatrealizinga
webofdata(orEntities,orThings,dependingontheinteresttodata
management,toentitylinking,ortosensorsandthingsinthephysi-
calworld). Linkeddataisoneofthetechnologiesforthesemantic
web(discussedinthenextsection),andconsistsoffourprinciples
andmanygoodpractices. Theprinciplesinclude:
1. usewebaddresses(URI)asnamesforentities/things;
2http://www.w3.org/DesignIssues/LinkedData.html.
JLIS.it. Vol.4,n.1(Gennaio/January2013).Art.#5457 p.256
JLIS.it.Vol.4,n.1(Gennaio/January2013)
2. useHTTPURIssothatpeoplecanlookupanddereference
thosenames;
3. whensomeonelooksupaURI,provideusefulinformation,
usingthestandards(RDF,RDFS,SPARQL,OWL,RIF);
4. include links to other URIs, in order to be able to discover
morethingsanddata.
Amonggoodpractices,it’susefultomentionthosethathavebest
supportedtheLinkingOpenData(LOD)bootstrap,whosestateof
playisvisualizedperiodicallyasacloud3:
• useopenlicensestoobtainhighlyreusabledata;
• usenon-proprietaryformats(e.g. CSVinsteadofExcel);
• useW3Copenstandards(typicallyRDF,4SPARQL,5OWL6)to
identifythings,sothatpeoplecanpointatyourstuff,newlinks
canbecreated,betterqueriesandmoreextendedreasoning
canbeperformed.
ThesepracticesalsofitrecommendationsfromtheOpenDatamove-
ment,andarecurrentlyadoptedinmanydifferentfields,including
Public Administration data7 and are used in the integration and
enrichmentofdata,forexamplefortheexpertfindingtask.8
TheLODCloudcontainslinkeddatafrommanydifferentdomains,
inparticularbiomedicine,cultural,multimedia,bibliographic,geo-
graphicetc. Anexampleofthepotentialoflinkeddataisshownin
3http://linkeddata.org.
4http://www.w3.org/RDF/.
5http://www.w3.org/TR/rdf-sparql-query/.
6http://www.w3.org/2004/OWL/.
7http://data.gov;http://data.gov.uk;http://dati.gov.it.
8http://data.cnr.it.
JLIS.it. Vol.4,n.1(Gennaio/January2013).Art.#5457 p.257
A.Gangemi,SemanticTechnologiesandLinkedData
figure1,agraphbuiltautomaticallybyanapplication(RelFinder9),
whichincrementallyvisualizestherelationsbetweenanytwoenti-
ties,providedthattheyhaveanidentityonthewebofdata. Inthe
figure1,graphbuildingstartsfromtheentities:
<http://dbpedia.org/wiki/Neo-positivism>
<http://dbpedia.org/wiki/Francis_Bacon>
Figure1:TheemergingrelationsbetweentwoentitiesacrosstheLinking
OpenDatagraph.
Semantic web standards
W3Copenstandards,primarilyRDF,10SPARQL11andOWL,12en-
ableelegantandhomogeneousrepresentationof,aswellasquerying
9http://www.visualdataweb.org/relfinder.php.
10http://www.w3.org/RDF/.
11http://www.w3.org/TR/rdf-sparql-query/.
12http://www.w3.org/2004/OWL/.
JLIS.it. Vol.4,n.1(Gennaio/January2013).Art.#5457 p.258
JLIS.it.Vol.4,n.1(Gennaio/January2013)
andreasoningon,thedatafrommosttraditionaldatastructuresand
datamodels.
RDFisbasedonarecursivedatastructure,calledtriple,madeofa
Subject,aPredicate,andanObject,analogouslytothemostabstract
grammatical structure of Western languages, SVO (Subject-Verb-
Object).
Listing1:SampleRDFtriples.
<http://www.cnr.it/ontology/cnr/individuo/
unitaDiPersonaleInterno/MATRICOLA1582>
<http://www.w3.org/2000/01/rdf-schema#label>
’’Aldo Gangemi’’
<http://www.cnr.it/ontology/cnr/individuo/
unitaDiPersonaleInterno/MATRICOLA1582>
<http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
<http://www.cnr.it/ontology/cnr/personale.owl#
UnitaDiPersonaleInterno>
<http://www.cnr.it/ontology/cnr/individuo/
unitaDiPersonaleInterno/MATRICOLA1582>
<http://purl.org/dc/terms/subject>
<http://dbpedia.org/resource/Category:Semantic_Web>
RDFtriplescanbequeriedviatheSPARQLlanguage.
Listing2:Queryontriplesin1
SELECT ?l
WHERE {
?x <http://purl.org/dc/terms/subject>
<http://dbpedia.org/resource/Category:Semantic_Web>.
?x <http://www.w3.org/2000/01/rdf-schema#label> ?l}
JLIS.it. Vol.4,n.1(Gennaio/January2013).Art.#5457 p.259
A.Gangemi,SemanticTechnologiesandLinkedData
Thequeryinlisting2ontheprecedingpagegetstheanswer:
l
’’Aldo Gangemi’’
EachtriplecontainsSubjectsandObjectsthathaveatype,whichis
onitsturnaClass,e.g.
<http://www.cnr.it/ontology/cnr/personale.owl#
UnitaDiPersonaleInterno>
<http://www.w3.org/1999/02/22-rdf-syntax-ns#type>
<http://www.w3.org/2002/07/owl#Class>
EachtriplecontainsaPredicate(orProperty),whichtogetherwith
classesformsthevocabulary(alsocalledschemaorontology)used
by a dataset. In cases where logical validation and reasoning is
required,avocabularyisdefinedintheOWL(OntologyWebLan-
guage) standard,13 a language that allows the use of automated
reasonerstoderivelogicalinferencesoutofdatastructures. Forex-
ample,anautomatedreasonerinferstheinversesofexistingtriples,
thesymmetrictriples,thetriplesholdingtransitively(whenappro-
priateruleshavebeendefinedforthevocabulary)etc.
With the expressive power of OWL and SPARQL on the web of
data,onecanmakecomplexquestionstoheterogeneousknowledge
sources,e.g. intheRomalLawdomain,thefollowingnaturallan-
guagequestionscanbeformalizedasqueries,buttermsneedtobe
mappedtoappropriateentitytypesinRDFandOWL.Inthiscase,
underlinedtermsaresupposedtobemappedasclasses,bold-faced
termsasproperties,andtermsinitalicsasspecificentitiesorvalues:
-which Roman Law sources contain maxims concerning stipulation, cite
Ulpian,andincludecommentariespublishedinthelast10years?
13http://www.w3.org/2004/OWL/.
JLIS.it. Vol.4,n.1(Gennaio/January2013).Art.#5457 p.260
JLIS.it.Vol.4,n.1(Gennaio/January2013)
-whichcasesappearedinCommonLawsystemscontaininterpretationsrel-
ativetocontractsanalogoustostipulatio?
Inordertoimprovevocabularyqualityandinferencecapabili-
ties,additionalaxiomsneedtobedefined(e.g. whattypeofentities
canbeanalogoustowhat,whatcanbecitedinwhatetc.). Therefore,
vocabularydesignrequiresacertainaccuracyandqualitycontrol,
which can be obtained by means of approaches oriented to user
requirements,andwiththereuseofstandardvocabulariesandon-
tologydesignpatterns,14knowntodescribethedomainofinterest,
and/orsolvingthemodelingproblemsemergingfromuserrequire-
ments.
Semantic applications
Availability of large open data can provide a good motivation to
develop next generation applications, which build on both exist-
ingandnovelsolutions,focusedonthesemanticparadigm: using
meaningofdataasawidespreadorganizationalschema.
Thelifecycleofasemanticapplicationistypicallythefollowing:
1. reengineering existing data, by producing datasets in RDF
triples(data)andOWL(vocabularies);
2. linkingbetweenentitiesinmultipledatasets,andproduction
ofnewtriples;
3. extractionofnewentitiesandtriplesbymeansofdatamining
andnaturallanguageprocessingtechniques,andproduction
ofnewtriples;
14http://www.ontologydesignpatterns.org.
JLIS.it. Vol.4,n.1(Gennaio/January2013).Art.#5457 p.261
A.Gangemi,SemanticTechnologiesandLinkedData
4. reasoning on the logical structures obtained from previous
steps,andpossibleproductionofnewtriples(materialization);
5. publishingofdatasetsonappropriateplatforms,forSPARQL
querying;
6. presentationofenricheddatatobeusedbywebusers: textual,
graphic,richsnippets,explorativeetc.
Thelifecyclereflectsamultipleinterpretationofthetermsemantics.
Insteps2and3,wereferprimarilytothelinguisticsemanticsthatis
implicitintheanalyzedtexts;relatedtechnologiesarethoseoftext
anddataanalysis,andaimsatrecognizingentities,names,terms,
relations,facts,topicsetc. Onlyoncewehaveextractedthem,we
canproducenewformaltriples. Insteps1,4,5,werefertological
(or formal) semantics of data and schemas; related technology is
basicallywhatwehavementionedinprevioussectionsas”semantic
web”(whichisamixofwebscienceandknowledgerepresentation).
Instep6.,werefertothesemanticsofuserinteraction.
Technologiesorientedtolinguisticsemanticsallowe.g. torecognize
entitiesintexts,andtoresolvetheiridentitywithrespecttoknown
datasets. Onceidentityhasbeenresolved, itispossibletoenrich
the dataset with known relations between that entity and other
entities. Forexample,giventhefollowingtextfromtheproceedings
ofEuropeanUnionParliament:
The sensitivities of Northern Ireland are too important for
anyill-informedbandwagoningontheInternationalFundfor
Ireland. Raytheon has been welcomed to Derry by no less
thanNobelPeacePrizewinners,JohnHume–oneofourown
colleagues,andDavidTrimble. Raytheonwillbefundedby
theIndustrialDevelopmentBoardinNorthernIreland. Not
oneeuronoroneIrishpoundfromtheInternationalFundfor
IrelandisgoingtoRaytheon.
JLIS.it. Vol.4,n.1(Gennaio/January2013).Art.#5457 p.262