Peter Ansell wrote:
> What would change if there were disjoint statements? Are disjoint
> declarations used for more than just verifying that dbpedia is
> consistent?
>
There are a lot of uses. I wouldn't trivialize the verification
part either: the wikipedia "street level" view hides serious
consistency problems that look quite embarrassing from 8000 feet up: if
you're making a product that gives people that 8000 foot view and you
don't want to get laughed out of town, you need to deal with them.
Also certain inference procedures will fail when applied to inconsistent
data, so providing a consistent view is an important processing step
------------
In my particular case I've identified a set of "things" that are
about a certain problem domain. Now I want to create an (internal)
broad classification of things, initially to decide which things get
further processing. For instance, I'm ready to work with Place,
Infrastructure and Organization members. My system could accomplish
something with Person, but there are practical issues I don't want to
deal with now. I suppose I could fit Work in somehow, but I don't know
how. I also got a few SupremeCourtofTheUnitedStatesCases and,
frankly, never want to see them ever again.
My categories are derived from a merger of the dbpedia ontology and
freebase types; the structure of the dbpedia ontology fits my problem
domain well, better than the FB type system does. On the other hand,
the FB type system is sometimes more specific -- some things are typed
in one system and not in the other, so merging the systems gives more
objects with types. Objects that have inconsistent assignments will be
tagged and investigated (not a lot of work, since there aren't many.)
Objects that don't have types go into an "Untyped" category -- some of
those are going to get types assigned, others are going to go in the
same holding area with Persons and Works (but still keep an "Untyped"
designation.)
Most of the work described is reasoning about objects, but reasoning
about types is important too: ~I~ had to reason about types in order to
design the procedure above, and I used my built-in "commonsense KB" to
infer that "Person and Place are disjoint" and then follow the
consequences of that. As time goes on, I'd like to automate more of
this procedure, so having that disjointness agreed upon and entered
into the dbpedia ontology would be a help.
-------------
Note that the dbpedia ontology isn't really an ontology of the
world, it's really an ontology about wikipedia entries, and it might
be a little less tight than something like Cyc would be. For instance,
the intersection between "Place" and "Organization" has 147 members,
and it's not crazy -- many organizations conduct all activities at a
single place, such as
http://en.wikipedia.org/wiki/West_Philadelphia_High_School
One certainly could create separate concepts for "WPHS the place"
and "WPHS the organization" and a set of facts describing the relation
between them, but that doesn't reflect what's in wikipedia, or the
popular understanding which conflates them. If anything, I'm surprised
that this set is as small as it is, since it could be applied to the
majority of schools. Of course you find some that are wrong too, such as
http://en.wikipedia.org/wiki/Anheuser-Busch
which an is an Organization that has activities at many places.
The intersection of "Person" and "Place" however, should be empty.
I see only 10 members of this set, one of which is wildly wrong
http://en.wikipedia.org/wiki/Achievements_of_Western_Art
and others of which are just sloppy
http://en.wikipedia.org/wiki/Reginald_A._Fessenden_House (see
http://en.wikipedia.org/wiki/Reginald_A._Fessenden)
--------------
------------------------------------------------------------------------------
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
