alignment of ontology design patterns: class as property value, value partition and normalisation

Alignment of Ontology Design Pa2erns: Class As Property Value, Value Par::on and Normalisa:on

Bene Rodriguez-‐Castro, Mouzhi Ge and Mar6n Hepp

The 11th Interna6onal Conference on Ontologies, DataBases, and Applica6ons of Seman6cs (ODBASE) 2012 – Rome (Italy)

Outline

•  Introduc:on

•  Revisi:ng the Class as Property Value (CPV) ODP

•  Revisi:ng the Value Par::on (VP) ODP

•  Revisi:ng the Normalisa:on ODP

•  Alignment of the CPV, VP and Normalisa:on ODPs

•  Conclusions

•  Future Work

2 B. Rodriguez-‐Castro, M. Ge, M. Hepp

Introduc:on

•  Ontology Design Pa2erns (ODPs) evolved from design pa2ern: „archetypal solu.ons to design problems in a certain context“

•  ODPs are receiving significant a2en:on due to the success

of so0ware DPs in soTware engineering •  Pa2erns that may not be applied consistently, can lead to

interoperability issues among the ontology models involved

•  Ideally, the outcome of this work will reduce the

opportunity for unintended inconsistencies.


Revisi:ng the Class as Property Value (CPV) ODP (Noy, 2005)

4

Approach 5: Use classes directly as annota:on property values Approach 4: Create a special restric6on

in lieu of using a specific value

Approach 3: Create a parallel hierarchy of instances as property values

Approach 2: Create special instances of the class to be used as property values

Approach 1: Use classes directly as property values

B. Rodriguez-‐Castro, M. Ge, M. Hepp

Approach 4: Create a special restric:on in lieu of using a specific value


Key Characteris:cs of Approach 4 of the CPV ODP

6

•  Compliance to OWL 1 DL profile.

•  Automa:c classifica.on of books based on subject by a standard OWL DL reasoner.

•  Use of anonymous individuals to approximate the use of a class as a property value.


Decoupling the Class as Property Value OPD

7

•  Implicit modica:on of the originally intended seman:c of the exis:ng class hierarchy subsumed by :Animal

–  Any instance of :Animal represents originally an actual animal in the real world

•  the original seman:cs of the classes that provide the values (subsumed by :Animal)

–  When an instance of :Animal is used as the value of the property dc:subject, it stands for an anonymous generic animal interpreted as the subject of a book

•  the seman:cs of the expected range of dc:subject

•  Coupling inadvertently two dis:nct modelling problems into one

–  The problem of using a class as a property value per se (strict-‐CPV)

–  The issue of not only using a class as a property value, but also, the possibility of altering its original intended meaning in the process as a result (coarse-‐CPV)


Approach 4: Changing the Target Domain from ``Book´´ to ``Zoo´´

8

Animal

Book About

Animals

“Lions: Life in the Pride”

Lion

African Lion

rdfs:subclassOf

Unidentified Lion(s)

Unidentified African Lion(s)

“The African Lion”

rdfs:subclassOf

rdf:type

rdf:type rdf:type

rdf:type

dc:subject

dc:subject

Book

rdfs:subclassOf

Zoo

Zoo With Animals

London Zoo

Munich Zoo

rdfs:subclassOf

rdf:type

rdf:type

:hasAnimal

:hasAnimal

``Book´´ Target Domain

``Zoo´´ Target Domain

An anonymous instance of :Animal does not represent an actual animal but the subject of a book


2

1

Approach 4: Changing the Target Domain from ``Book´´ to ``Zoo´´

9

Animal

Lion

African Lion

rdfs:subclassOf

Unidentified Lion(s)

Unidentified African Lion(s)

rdfs:subclassOf

rdf:type

rdf:type

Zoo

Zoo With Animals

London Zoo

Munich Zoo

rdfs:subclassOf

rdf:type

rdf:type

:hasAnimal

:hasAnimal

An anonymous instance of :Animal represents an actual animal


Example of coarse-‐ and strict-‐ CPV ODP


coarse-‐CPV ``Book´´ Target Domain

strict-‐CPV ``Zoo´´ Target Domain

strict-‐CPV versus coarse-‐CPV ODP

12

•  In the strict-‐CPV, the natural range (rdfs:range) of the property :hasAnimal does align with what the class :Animal originally represents

•  In the coarse-‐CPV, the natural range of the property dc:subject in the context of Approach 4 of the CPV ODP, does not align with the orignal seman:c of the class :Animal

•  Syntac.cally, the implementa:on of the coarse-‐CPV and strict-‐CPV ODPs, is essen:ally equivalent

–  Elements placed at equivalent posi:ons on the ontological structure of both pa2erns, perform equivalent func:ons and are implemented following the same set of OWL idioms

•  Seman.cally, the implica:ons of each variant can be par:cularly different.


Revisi:ng the Value Par::on (VP) ODP (Rector, 2005)

13

Pa2ern 1: Values as sets of individuals

Pa2ern 2. Representa:on variant 1: Using a fact about the individual

PaQern 2. Representa6on using variant 2: Placing an existen6al restric6on on the individual


Pa2ern 2. Representa:on using variant 2: Placing an existen:al restric:on on the individual


An anonymous instance of :Good_health_value (e.g., :Johns_Health) does represent an actual health value

Key Characteris:cs of Pa2ern 2 – Variant 2 of the VP ODP

15

•  Compliance to OWL 1 DL profile.

•  The use of classes instead of individuals to represent the feature space (good, medium, poor) of the feature class (health)

–  Use of anonymous individuals to represent the health status of a person (e.g., :Johns_Health)

•  Automa:c classifica.on of people based on their health status by a standard OWL DL reasoner.


Revisi:ng the Normalisa:on ODP (Egana-‐Aranguren, 2009)


poly-‐hierarchy (tangled ontology)

Non-‐normalised ontology: the subsump:on rela:ons that create the poly-‐hierarchies are manually asserted

Normalisa:on ODP (Asserted)


•  the exis:ng poly-‐hierarchies in the structure of the normalized asserted ontology model are removed

•  the subsump:on rela:ons are

implemented explicitly as restric:ons

single-‐inheritance (untangled ontology)

Normalised ontology (asserted): the pa2ern allows exactly one unlabelled flavour of is-‐a link, which translates into a single-‐inheritance structure of the asserted subsump:on rela:ons.

Normalisa:on ODP (Inferred)


Normalised ontology (inferred): the implementa:on of the subsump:on rela:ons explicitly as restric.ons enables a standard DL reasoner to automa.cally maintain the original poly-‐hierarchies in the inferred ontology model

poly-‐hierarchy (maintained automa:cally by a standard DL reasoner)

Key Characteris:cs of the Normalisa:on ODP

19

•  Compliance to OWL 1 DL profile

•  The use of classes (e.g., classes subsumed by :Func:on) to represent explicitly a principle of division of the ontology central domain concept (e.g., :Cell)

–  Use of anonymous individuals (e.g., individuals subsumed by :Func:on) to represent the func:on performed by a cell

•  Automa:c classifica.on and maintainance of the mul:ple and complex subsump:on rela:ons by a standard OWL DL reasoner


20

Alignment of the CPV, VP and Normalisa:on ODPs


Result: Common Func:onal Groups in ODPs

21

•  Target Domain •  Domain Elements •  Domain Defined Classes •  Core Property •  Range Subsump:on Class Hierarchy

–  Range Anonymous Individuals

(For a defini:on of each common func:onal group, please see full paper available at: h2p://dx.doi.org/10.1007/978-‐3-‐642-‐33615-‐7_16)


22

Result: Example of Common Func:onal Groups

Domain Elements

Domain Defined Classes

Range Subsumpt. Class Hierar.

Target Domain

Core Property


Similari:es and Differences in ODPs Func:onal Groups


ODP Func6onal Group Similar Func6onality Different Func6onality

Target Domain All None

Domain Elements coarse-‐CPV, strict-‐CPV, Normalisa:on

VP(one-‐to-‐one vs. one-‐to many)

Domain Defined Classes All None

Core Property coarse-‐CPV, strict-‐CPV, Normalisa:on

VP(owl:Func:onalProperty)

Range Subsump:on Class Hierarchy

coarse-‐CPV, strict-‐CPV

Normalisa:on (disjointness), VP (par::on)

Anonymous Individuals strict-‐CPV, Normalisa:on, VP

coarse-‐CPV (altered seman:cs)

Conclusions

•  Approach 4 of the CPV ODP can be decoupled into two •  A comparison of all ODPs, shows five different func.onal

groups based on the ontological structure of the pa2ern and the syntac:c and seman:c func:on that each element performs

•  Nested-‐doll effect -‐ All instan:a:ons of the Pa2ern 2 -‐

Variant 2 of the VP ODP and the Normalisa:on ODP, use implicitly the strict-‐CPV ODP

•  Three different modelling scenarios are being addressed by

slight modica:ons over the same set of OWL idioms


Future Work

•  Punning meta-‐modelling capability available in OWL 2

•  Scale -‐ Increase number of ODPs that par:cipate in the compara:ve analysis

•  Evalua:on framework extending the current ODPs evalua:on and documenta:on templates


THANK YOU

Bene Rodriguez-‐Castro, Mouzhi Ge, Mar:n Hepp, Alignment of Ontology Design Pa2erns: Class As Property Value, Value Par::on and Normalisa:on, in: R. Meersman, H. Pane2o, T. Dillon, S. Rinderle-‐Ma, P. Dadam, X. Zhou, S. Pearson, A. Ferscha, S. Bergamaschi, I. Cruz (Eds.), On the Move to Meaningful Internet Systems: OTM 2012, Vol. 7566 of Lecture Notes in Computer Science, Springer Berlin Heidelberg, 2012, pp. 682-‐699. doi:10.1007/978-‐3-‐642-‐33615-‐7_16.

Bene Rodriguez-‐Castro [email protected]

hQp://purl.org/beroca

Reference:

Key References

1.  Noy, N.F.: Represen:ng Classes As Property Values on the Seman:c Web. Technical Report Note 5, W3C, Seman:c Web Best Prac:ces and Deployment Working Group (2005), h2p://www.w3.org/TR/swbp-‐classes-‐as-‐values/

2.  Rector, A.: Represen:ng Specied Values in OWL: „value par::ons“ and „value sets“. Technical Report Note 17, W3C, Seman:c Web Best Prac:ces and Deployment Working Group (May 2005), h2p://www.w3.org/TR/swbp-‐specified-‐values

3.  Egana-‐Aranguren, M.: Role and Applica:on of Ontology Design Pa2erns in Bio-‐ontologies. Ph.D. thesis, School of Computer Science, University of Manchester (2009), h2p://mikeleganaaranguren.files.wordpress.com/2010/01/thesis.pdf


alignment of ontology design patterns: class as property value, value partition and normalisation

Technology

type zoo zoo

type rdf

subclassof zoo zoo

subclassof rdf

hepp owl

hepp ananonymousinstance

type dc

animal lion african