bestmap:)context.aware)skos)...
TRANSCRIPT
BestMap: Context-‐Aware SKOS Vocabulary Mappings in OWL 2
Rinke Hoekstra
Overview
• Use Case – Access to court proceedings
• Vocabulary Mapping
• Requirements
• BestMap
• Discussion
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 1
BestPortal
• BEST Project – “BATNA Establishment using SemanTc Web Technology”
– Best AlternaTve to a NegoTated Agreement
• Improve access to court proceedings – Netherlands Council of the Judiciary hYp://www.rechtspraak.nl
– 50 thousand verdicts
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 2
“Does my case stand a chance in court?”
• Full text search is not enough – Laymen – Lawyers
• Lawyers have their own language: legalese – Bridge the gap between common sense and legal knowledge
• Knowledge-‐based soluTon too expensive – Modelling effort – Quality assurance – Legal theory: defini&ons
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 3
BestPortal: Requirements
• Translate layman descripTon to legal terms
• Search using fingerprints of legal terms
• Context in which layman concepts co-‐occur in a case determines the applicability of a legal concept
• A mapping is not the definiTon of a concept 24 October 2009 Rinke Hoekstra -‐ OWLED 2009 4
Layman Vocabulary
Layman Case Description
Case
LegalVocabulary
Legal Case Description
Case Law Database
Mapping
Vocabularies
• Cultural Heritage – Museums, libraries
– Huge repositories of (rich) informaTon – Annotated using many different vocabularies (knowledge organizaTon systems)
• Concept-‐based informa&on retrieval – Europeana portal (hYp://www.europeana.eu)
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 5
Why Vocabulary Mapping
• Integrated access? Vocabulary mapping! – Format (XML to RDF)
• e.g. via XSLT one-‐way transformaTon – Structure (VCard:Given + VCard:Family to foaf:name)
• e.g. via SPARQL++, one-‐way transformaTon – Concepts (foaf:Person to lkif:Person)
• Simple Knowledge OrganizaTon System (SKOS) – Liking exisTng KOS’s to the SemanTc Web – Every skos:Concept is an OWL individual – Lightweight seman&c relaTons: broader, narrower, and related. – Lightweight mapping relaTons between skos:ConceptSchemes.
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 6
Europeana Datacloud
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 7
Mapping in SKOS
skos:mappingRelation
skos:closeMatch(symmetric)
skos:exactMatch(symmetric, transitive)
skos:broadMatchskos:narrowMatch
skos:relatedMatch(symmetric)
skos:narrower skos:broader
skos:semanticRelation
skos:related
inverse disjoint
disjoint
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 8
Informa=on retrieval perspec=ve vs. lightweight seman=cs
• No many-‐to-‐many mappings – Mapping only between pairs of concepts
– Required for re-‐indexing and search across collecTons (Isaac et al. 2007)
• … fundamental issue – SKOS concepts and relaTons are ‘intensional’ – What does a mapping then mean?
• Implicit assumpTon of extensionality
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 9
Extensional View
• SKOS relaTons – “Resources annotated by some concept should be retrievable via its broader concept.”
• SKOS mappings – “Resources annotated by some concept should be retrievable via the concepts it is mapped to.”
• … only means to assess quality
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 10
BestMap: Requirements
• Extensional perspecTve – Concepts as annota&ons on resources
• CompaTbility – Integrated with SKOS
• Hierarchic mappings – Exploit skos:broader and skos:narrower
• Many-‐to-‐many mappings – Granularity – Context determines whether a mapping holds
• Flexible and Lightweight – A mapping is not the definiTon of a concept
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 11
ConnecTng to SKOS (1)
• RelaTon between :Resource and skos:Concept
• Direct and indirect annotaTons
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 12
the location of a crime, etc. However, in our case, the school is most relevant asthe owner of an animal (the horse) that caused some damage (the kick). If theperson being kicked is an employee, then it might also be a case of liability forsubordinates.
Lastly, specifying mappings at this level of specificity is not very flexible,and we should therefore exploit the skos:narrower and skos:broader hierarchy toallow for similar cases to be covered by a single mapping. Section 5 introducesa framework, or design pattern if you will, for specifying context-aware map-pings between two (or more) vocabularies, and specifies means to extend thisframework by including more contextual information using OWL 2 role chains.
5 BestMap: Context-Aware Mappings
As has been made clear in the previous sections, mapping relations often rely onintensional semantics, where the applicability of mappings is often existentiallydefined. In our approach, we exploit the power of OWL 2 DL to define restrictionson the members of sets of annotated resources as intensional descriptions. TheBestMap ontology makes explicit the medium – a resource – through whicha mapping can be specified.11 To do this, we introduce the property :aboutand its inverse that allow us to annotate resources with concepts from a SKOSvocabulary:
:about ≡ inv(:describes)
The :about property relates a :Resource (its domain) with some skos:Concept (itsrange). The reason for introducing a new relation is that other options, such aspunning of SKOS concept individuals, or reusing existing vocabulary are limitedin the sense that they cannot be subject to OWL 2 class restrictions.
As we are dealing with SKOS vocabularies, it seems useful to integrateSKOS’s semantic relations with this approach. There are two issues here. First,simply defining a sub property role chain that reuses :about is not allowed inOWL 2 as it introduces a cycle [3]. We therefore need to distinguish betweendirect annotations, and indirect annotations:
:d about � :about
:d describes � :describes
:d describes ≡ inv(:d about)
Secondly, SKOS makes a distinction between quite a variety of semantic rela-tions. We define sub properties of :about and :describes for each of these, takinginto account the direction of the relation. For instance:
:d about o skos:broaderTransitive � :bt about
:d about o skos:narrowerTransitive � :nt about
11 The BestMap ontology is available from http://www.best-project.nl/owl/bestmap.owl.
the location of a crime, etc. However, in our case, the school is most relevant asthe owner of an animal (the horse) that caused some damage (the kick). If theperson being kicked is an employee, then it might also be a case of liability forsubordinates.
Lastly, specifying mappings at this level of specificity is not very flexible,and we should therefore exploit the skos:narrower and skos:broader hierarchy toallow for similar cases to be covered by a single mapping. Section 5 introducesa framework, or design pattern if you will, for specifying context-aware map-pings between two (or more) vocabularies, and specifies means to extend thisframework by including more contextual information using OWL 2 role chains.
5 BestMap: Context-Aware Mappings
As has been made clear in the previous sections, mapping relations often rely onintensional semantics, where the applicability of mappings is often existentiallydefined. In our approach, we exploit the power of OWL 2 DL to define restrictionson the members of sets of annotated resources as intensional descriptions. TheBestMap ontology makes explicit the medium – a resource – through whicha mapping can be specified.11 To do this, we introduce the property :aboutand its inverse that allow us to annotate resources with concepts from a SKOSvocabulary:
:about ≡ inv(:describes)
The :about property relates a :Resource (its domain) with some skos:Concept (itsrange). The reason for introducing a new relation is that other options, such aspunning of SKOS concept individuals, or reusing existing vocabulary are limitedin the sense that they cannot be subject to OWL 2 class restrictions.
As we are dealing with SKOS vocabularies, it seems useful to integrateSKOS’s semantic relations with this approach. There are two issues here. First,simply defining a sub property role chain that reuses :about is not allowed inOWL 2 as it introduces a cycle [3]. We therefore need to distinguish betweendirect annotations, and indirect annotations:
:d about � :about
:d describes � :describes
:d describes ≡ inv(:d about)
Secondly, SKOS makes a distinction between quite a variety of semantic rela-tions. We define sub properties of :about and :describes for each of these, takinginto account the direction of the relation. For instance:
:d about o skos:broaderTransitive � :bt about
:d about o skos:narrowerTransitive � :nt about
11 The BestMap ontology is available from http://www.best-project.nl/owl/bestmap.owl.
ConnecTng to SKOS (2)
• TransiTve broader/narrower
• Similar for other SKOS relaTons
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 13
ljn:AE3482 lv:horse
lv:animal
:d_about
skos:broaderTransitive
:bt_about
lv:animal
skos:narrowerTransitive:nt_about
the location of a crime, etc. However, in our case, the school is most relevant asthe owner of an animal (the horse) that caused some damage (the kick). If theperson being kicked is an employee, then it might also be a case of liability forsubordinates.
Lastly, specifying mappings at this level of specificity is not very flexible,and we should therefore exploit the skos:narrower and skos:broader hierarchy toallow for similar cases to be covered by a single mapping. Section 5 introducesa framework, or design pattern if you will, for specifying context-aware map-pings between two (or more) vocabularies, and specifies means to extend thisframework by including more contextual information using OWL 2 role chains.
5 BestMap: Context-Aware Mappings
As has been made clear in the previous sections, mapping relations often rely onintensional semantics, where the applicability of mappings is often existentiallydefined. In our approach, we exploit the power of OWL 2 DL to define restrictionson the members of sets of annotated resources as intensional descriptions. TheBestMap ontology makes explicit the medium – a resource – through whicha mapping can be specified.11 To do this, we introduce the property :aboutand its inverse that allow us to annotate resources with concepts from a SKOSvocabulary:
:about ≡ inv(:describes)
The :about property relates a :Resource (its domain) with some skos:Concept (itsrange). The reason for introducing a new relation is that other options, such aspunning of SKOS concept individuals, or reusing existing vocabulary are limitedin the sense that they cannot be subject to OWL 2 class restrictions.
As we are dealing with SKOS vocabularies, it seems useful to integrateSKOS’s semantic relations with this approach. There are two issues here. First,simply defining a sub property role chain that reuses :about is not allowed inOWL 2 as it introduces a cycle [3]. We therefore need to distinguish betweendirect annotations, and indirect annotations:
:d about � :about
:d describes � :describes
:d describes ≡ inv(:d about)
Secondly, SKOS makes a distinction between quite a variety of semantic rela-tions. We define sub properties of :about and :describes for each of these, takinginto account the direction of the relation. For instance:
:d about o skos:broaderTransitive � :bt about
:d about o skos:narrowerTransitive � :nt about
11 The BestMap ontology is available from http://www.best-project.nl/owl/bestmap.owl.
ConnecTng to SKOS (3)
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 14
:cm_about
:em_about
:bm_about :rm_about
:b_about
disjoint
disjoint
:bt_about
:about
:r_about
:nm_about
:n_about
:nt_about
disjoint
disjoint
Mappings as OWL Classes
• A mapping class: – Classifies resources annotated using one vocabulary, and
– Infers annotaTons using the other vocabulary – … it may be directed
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 15
:cm_about
:em_about
:bm_about :rm_about
:b_about
disjoint
disjoint
:bt_about
:about
:r_about
:nm_about
:n_about
:nt_about
disjoint
disjoint
Fig. 2. Redefinition of SKOS relations using :about.
follows:
ex:AO Mapping ≡ :about value lv:animal � :about value lv:company �:about value lv:dangerous action
� :d about value tv:animal � :d about value tv:animal owner �:d about value tv:damage
Admittedly, this mapping is quite unrestricted as it triggers even for resourcesannotated with concepts that only have a skos:related relation with the conceptsof the restriction. It is easy to see how the more specific relations of Figure 2 canbe used in the class axiom to further restrict the applicability of the mapping.
The mapping is directed since a resource annotated with the legal concepts ofex:AO Mapping will not be classified as an instance of that class. The directioncan be reversed by swapping the subclass and equivalent class restrictions onannotations. Furthermore, two annotations can be defined as equivalent by usingonly equivalent class restrictions. There is of course no restriction on the use ofadditional OWL 2 expressions in the class axiom. However, for the mappingit makes no real sense to use a union in the equivalent class restriction: thiswill not lead to new (known) :d about relations between matching resourcesand concepts. One step further is the possibility to exclude resources from amapping where they are annotated using some concept. This can be done byintroducing the complement of a value restriction in the equivalent class axiomof a mapping. However, because of the open world assumption this does notmean that resources that are not annotated will be members of the mappingclass: this requires explicit negative property assertions on the :d about property(thankfully available in OWL 2).
Example
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 16
lv:horse
lv:riding_school
lv:kick
:ao_example
:d_about
:d_about
:d_about
tv:animal
tv:animal_owner
lv:animal
lv:company
lv:dangerous_action
skos:broader
skos:broader
skos:broader tv:damage
:AO_Example
rdf:type
:d_about
:d_about
:d_about
:about
:about
:about
Discussion
• Extensible – Any OWL axiom may be used in a mapping (e.g. someValuesFrom etc.)
– Reusable (parTal) mappings – Exclude resources annotated with a parTcular concept
• NegaTve property asserTons • Novel – “ReificaTon” wrt. normal OWL ontologies
• Overcomes limitaTons of SKOS semanTcs – Makes explicit the extensional perspecTve that underlies SKOS semanTcs
– Non-‐intrusive
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 17
Discussion
• OWL 1 vs OWL 2 – Property chains – Disjoint properTes – NegaTve property asserTons
• The bad – Cannot enforce that the mapping holds between two disTnct concept schemes
– Property chains are not equivalent to super property
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 18
Future Work
• Apply BestMap to other domains – Legal assessment based on spaTal plans – …
• Further development of BestPortal – Do the mappings actually work? – Structured mappings
• (case frames)
– Does BestPortal really improve access to court proceedings?
– Connect to the linked data cloud
24 October 2009 Rinke Hoekstra -‐ OWLED 2009 19