A Contextualized Semantics for owl:sameAs


June 2nd, 2016, ESWC
Wouter Beek (w.g.j.beek@vu.nl), Stefan Schlobach, Frank van Harmelen

wouterbeek.github.io

Are these owl:sameAs?

Two types of meaning

Formal meaning

DL, OWL, Model Theory

Social meaning

Linked Data, SKOS, Schema.org

owl:sameAs has 2 meanings

Formal meaning

$$a = b \,\longleftrightarrow\, (\forall P)(Pa = Pb)$$

OWL 2 specification, 2012

Social meaning

“Include links to other URIs, to discover more things.”

Tim Berners-Lee, 4th Linked Data principle, 2006

Example

How formal meaning and social meaning collide

Existing solutions

Use weaker alternatives
(e.g. relatedness)

Everything is related to everything.


SKOS exactMatch indicates a high degree of confidence that two concepts can be used interchangeably across a wide range of information retrieval applications
SKOS specification, 2009

Use domain-specific identity relations

“the same chemical” (ex:sameChemical)

“the same product” (ex:sameProduct)

bbc:sameAs

bbc:sameAs

owl:sameAs

?

Change modeling practice

Adding an owl:sameAs link requires approval from the authority that is being linked to.

Our solution

(See the paper for details)

Preserve equivalence properties

The identity relation is the smallest equivalence relation.

Every equivalence relation is also an identity relation, but w.r.t. a subset of the properties.

If A and B have the same income, I cannot infer identity.

But if I am reasoning about income groups, then they are identical.

Identity subrelations that are also equivalence relations can, in some contexts, be used i.o. the identity relation.

Not the same sui generis

But… the same chemical compound ($=_{\{\text{chemical}\}}$)

The same chemical compound ($=_{\{\text{chemical}\}}$)

ns1:Aspirin,ns2:Aspirin〉and 〈ns1:Nicotine,ns2:Nicotine〉 are the same according to the same criteria.

Lattice of identity subrelations

Identity context: A consistent collection of identity subrelations from the identity lattice

Pros & cons

  • We do not change the schema
  • We do not change the data
  • We do not lose equivalence properties
  • Consistently reason over inconsistent data
  • Very few things are the same sui generis
  • Highly non-monotonic: identity subrelations change when statements are added/removed
  • Computationally challenging (parallel FCA computation)

Questions?



Formal meaning



$$a = b \,\longleftrightarrow\, (\forall P)(Pa = Pb)$$

Social meaning