Hi,
following the discussion regarding deficiencies of the DBpedia ontology, I
would like to make you know that a team of collaborators sponsored by
Structured Dynamics is working on classification of the Wikipedia articles
into the Umbel [1] ontology. Yet another ontology, you may think, but Umbel is
exceptional in the regard that it is entirely based on Cyc [2] - a more than
20-year effort of building an ontology large enough to capture entire common
sense knowledge. We address many of the problems that are present both in
Wikipedia and DBpedia, e.g. things like list and disambiguation pages,
administrative categories in Wikipedia, compound categories in Wikipedia, etc.
The good news is that exactly this week we've produced the first "final" result
of the classification, i.e. around 90% of the Wikipedia articles has a Umbel
type (we use ~4k different types in the classification) assigned via a mapping
between Wikipedia categories and Umbel types. I am writing in quotes, since
that result still has to be validated and curated and we are probably not going
to publish it just at the moment. Still, I expect some consumable results to be
produced in May, so stay tuned.
What is more - regarding the problem of integrating the data from different
data sources - I used Cyc, especially its conflict detection mechanism, in the
past [3] to integrate classification results produced by various type
assignment methods. For me it seems to be the best resource applicable in that
problem, since besides the generalization relation, which is found in all the
ontologies, most of the classes have disjointWith statements, even on a level
of Cats and Dogs (i.e. individual species). As a result it is relatively easy
to keep the data coherent.
Kind regards,
Aleksander
[1] http://umbel.org
[2] http://www.cyc.com/platform/opencyc
[3] http://ceur-ws.org/Vol-906/paper2.pdf
---- Wł. Śr, 09 kwi 2014 23:25:25 +0200 Kingsley
Idehen<kide...@openlinksw.com> napisał(a) ----
On 4/9/14 4:53 PM, Paul Houle wrote:
> The type assignments in DBpedia are very precise (few false
> statements) but not accurate in the sense that recall is poor; many
> things fall through the cracks. The real problem is that the the
> mappings are the map, not the territory. Wikipedia is an
> encyclopedia for humans, not for machines, so DBpedia has to parse
> whatever unsane markup they give us.
>
> Systems like Wikidata and Freebase can be edited by machines and
> human ontologists and get better recall for types.
>
> http://basekb.com/
>
> is a conversion from Freebase to industry standard RDF. You could
> use :BaseKB as a substitute for DBpedia, but DBpedia has advantages
> too because in addition to the 4 million things important enough to be
> in DBpedia, there is another 37 million unimportant things in :BaseKB
> that matter only to librarians, video store clerks and professional
> discographers.
>
> These unimportant things will drive you crazy unless you master
> them, and the easiest way to turn down the noise is to restrict
> search to the 4 million things.
>
> I could make you an RDF file that has statements such as
>
> ?dbpediaTopic a ?freebaseType .
>
> you could load that together with the rest of DBpedia. That would
> get you a long way towards good lists. The trouble at this point is
> that you don't have the freebase types connected to the DBpedia types
> so you can't join them against the schema to find properties and such.
> Mapping the types to the DBpedia types would not be that hard either,
> since the two systems are well aligned. Then you get something that
> looks like DBpedia but has more accurate types.
>
> Freebase has more accurate and better populated data for things
> like ticker symbols, geo-coordinates, genders, birth dates and the
> like. It would not be hard to rewrite Freebase statements to
>
> ?dbpediaTopic ?freebasePredicate ?anotherDbpediaTopic .
>
> and that would produce something that would be remarkably user friendly.
:baseKB could (and maybe should) pitched as a human-and-machine curated
bridge between Freebase, DBpedia, and Wikidata (I think).
Have you considered mapping the classes and properties across DBpedia,
Freebase, and Wikidata?
--
Regards,
Kingsley Idehen
Founder & CEO
OpenLink Software
Company Web: http://www.openlinksw.com
Personal Weblog: http://www.openlinksw.com/blog/~kidehen
Twitter Profile: https://twitter.com/kidehen
Google+ Profile: https://plus.google.com/+KingsleyIdehen/about
LinkedIn Profile: http://www.linkedin.com/in/kidehen
------------------------------------------------------------------------------
Put Bad Developers to Shame
Dominate Development with Jenkins Continuous Integration
Continuously Automate Build, Test & Deployment
Start a new project now. Try Jenkins in the cloud.
http://p.sf.net/sfu/13600_Cloudbees_______________________________________________
Dbpedia-discussion mailing list
Dbpedia-discussion@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dbpedia-discussion
------------------------------------------------------------------------------
Put Bad Developers to Shame
Dominate Development with Jenkins Continuous Integration
Continuously Automate Build, Test & Deployment
Start a new project now. Try Jenkins in the cloud.
http://p.sf.net/sfu/13600_Cloudbees
_______________________________________________
Dbpedia-discussion mailing list
Dbpedia-discussion@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dbpedia-discussion
