Jens Lehmann wrote:
Hello John,
John Goodwin wrote:
Thanks Chris and team for all your hard work getting this done. I do,
however, have a few comments regarding the OWL ontology. I think in
general the use of domain and range is perhaps a bit "dubious" in that
for many things I think it is overly specified. I can imagine anyone
re-using the Dbpedia properties getting some unexpected inferences from
the domain and range restrictions. Also the range restriction seem to be
done as an OWL intersection so if, for example, something has a
publisher x then x will be inferred to be both a Company and a Person
which is probably not what you want. Personally, in all but a few cases,
I'd be tempted to generalise or just remove the domain/range
restrictions. Any thoughts?
We specified the domains and ranges as disjunctions of classes (not
intersection). See the W3C specification of owl:unionOf [1].
The domain and range axioms help to structure DBpedia and clarify the
meaning of certain properties. While there is room for improvement, it
is not an option to remove all of them.
Currently, there are two versions of the infobox extraction: a loose one
and a strict one. In the strict one, it is guaranteed that the data
complies to the ranges specified in the ontology schema. Currently, only
the loose (probably inconsistent) one is provided.
Kind regards,
Jens
[1] http://www.w3.org/TR/owl-guide/#owl_unionOf
Jens,
What's the URL of the strict one?
We are building a DBpedia installer for Virtuoso, so at the very least I
want the users of this installer to have choice of "strict" or "loose"
infobox extraction.
--
Regards,
Kingsley Idehen Weblog: http://www.openlinksw.com/blog/~kidehen
President & CEO
OpenLink Software Web: http://www.openlinksw.com
