It's pretty easy to map some types between dbpedia and freebase;
for instance, both systems agree on what a "Person" is, so we can say
the "Person" type is equivalent. If we find that, say, a Yago class
contains a list of Persons, then we can say that that the Yago class is
a subclass of Person, even the freebase "Person". That's all pretty easy.
Then there are the types that don't quite align. Dbpedia's
"Organization" class corresponds most closely to "Employer" in
Freebase. Employer is a duck type, that is, you'd prove that X is an
employer if there some person Y such that X employs Y. Most
significant organizations employ, and few non-organizations employ (Is
"Dalia Dippolito" an employer because she hired a hit man to kill her
husband?) There's still the problem that a volunteer FireDepartment is
not an employer, but other FireDepartments are. Dbpedia's
"Organization" which contains FireDepartment, FootballTeam, Business
and such, is a more useful type to me than "Employer" is.
We don't have a good language for describing the relationship
between "Organization" and "Employer". "Employer is a subclass of
Organization" comes close, but it might not be exactly true.
Now, a set of "Organizations" could be extracted from Freebase by:
(i) discovering freebase types which are subclasses (or
near-subclasses) of Organization, and
(ii) filtering non-Organizations out, if they exist
No we've got examples where:
(a) freebase can help dbpedia (finding the missing {ersons), and
(b) dbpedia can help freebase (creating a missing type)
There are a lot of practical issues in expressing complex type
mappings with things like OWL and SKOS, but there is a simple method
that works with well-implemented standards.
Rather than describing a type mapping, a type mapping can be
expressed as a set. For instance, I could go through the
dbpedia<-->freebase type mappings, discover a bunch of missing
Persons, and then publish an NT file that asserts the Personhood of the
missing Persons. This could then be loaded into any RDF processing
system to improve the accuracy of dbpedia.
In some cases one might want to remove triples, in which case the
easiest answer is to produce a file that replaces a file from dbpedia.
For instance, at this point I know of a handful of type assignments
that are wrong in dbpedia; I can publish an NT file that removes
assertions I don't believe and adds assertions that I do believe.
------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
Dbpedia-discussion mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dbpedia-discussion
