LOD constellation concerns
apohllo
For a long time I've worked with the Cyc ontology, and I am quite
familiar with it. As far as I know, Umbel was created as a kind of
clean-up of Cyc, so many of the concepts (and their relations) present
in Umbel are simply taken from Cyc.
On the other hand, the LOD constellation diagram http://umbel.org/lod_constellation.html
shows that (maybe I am wrong here) there are more class level links
from Umbel to FOAF and other ontologies, than to Cyc (I assume that
the weight of the arrow indicates the number of links, as it is in the
original LOD diagaram).
But what is my biggest concern is the methodology of the diagram.
Working with Cyc, I have discovered many class-level links in that
ontology, mostly with the DBpedia (e.g. check the Dog concept). The
problem is that they are expressed not via the equivalentClass or
similar predicates, but via the sameAs predicate.
As far as I can understand the definition of the sameAs predicate, it
is not restricted to instances, so in my opinion the diagram should be
corrected, to include that predicate, i.e. it should also include the
links between owl classes linked via the sameAs predicate.
Kind regards,
Aleksander
Mike Bergman
On 6/21/2010 4:24 AM, apohllo wrote:
> Hello!
>
> For a long time I've worked with the Cyc ontology, and I am quite
> familiar with it. As far as I know, Umbel was created as a kind of
> clean-up of Cyc, so many of the concepts (and their relations) present
> in Umbel are simply taken from Cyc.
ALL concepts in UMBEL are also found in OpenCyc, and NO concepts are in
UMBEL that are not in Opencyc.
>
> On the other hand, the LOD constellation diagram http://umbel.org/lod_constellation.html
> shows that (maybe I am wrong here) there are more class level links
> from Umbel to FOAF and other ontologies, than to Cyc (I assume that
> the weight of the arrow indicates the number of links, as it is in the
> original LOD diagaram).
> But what is my biggest concern is the methodology of the diagram.
> Working with Cyc, I have discovered many class-level links in that
> ontology, mostly with the DBpedia (e.g. check the Dog concept). The
> problem is that they are expressed not via the equivalentClass or
> similar predicates, but via the sameAs predicate.
See [1], esp the [2]file on that page, which is the basis for generating
the link weights on the diagram.
It is perhaps not as clear as it could be. The diagram does NOT reflect
the number of concepts with linkages, but the number of class predicates
used.
You may also want to see the background post on this diagram [3]. It says:
"The bubble sizes are related to number of linked structures
(ontologies) to which the node has class linkages. The arrow thicknesses
are related to number of linking predicates between the nodes. Two-way
arrows are shown as darker and indicate equivalentClass or matching
superClassOf and subClassOf; single arrows represent subClassOf
relationships only."
But, I agree, this presentation is not the only way to represent
linkages, nor perhaps even the best one. ;)
From the reference [8] on [3], you can obtain the raw data files for
calculating this diagram on other bases.
Hope that helps!
Thanks, Mike
[1]http://umbel.org/lod_constellation.html
[2]http://www.umbel.org/lod_constellation/class_level_lod_constellation.csv
[3]http://www.mkbergman.com/457/a-new-constellation-in-the-linking-open-data-lod-sky/
>
> As far as I can understand the definition of the sameAs predicate, it
> is not restricted to instances, so in my opinion the diagram should be
> corrected, to include that predicate, i.e. it should also include the
> links between owl classes linked via the sameAs predicate.
>
> Kind regards,
> Aleksander
>
--
__________________________________________
Michael K. Bergman
CEO Structured Dynamics LLC
319.621.5225
skype:michaelkbergman
http://structureddynamics.com
http://mkbergman.com
http://www.linkedin.com/in/mkbergman
__________________________________________
apohllo
> > ALL concepts in UMBEL are also found in OpenCyc, and NO concepts are
> > in UMBEL that are not in Opencyc.
> > generating the link weights on the diagram.
> > It is perhaps not as clear as it could be. The diagram does NOT
> > reflect the number of concepts with linkages, but the number of class
> > predicates used.
Umbel which would relate them with appropriate classes from Cyc? So
far,
only sameAs predicate is used.
> > You may also want to see the background post on this diagram [3]. It
> > says:
> >
> > "The bubble sizes are related to number of linked structures
> > (ontologies) to which the node has class linkages. The arrow
> > thicknesses are related to number of linking predicates between the
> > nodes. Two-way arrows are shown as darker and indicate equivalentClass
> > or matching superClassOf and subClassOf; single arrows represent
> > subClassOf relationships only."
inferences, which might be drawn from the data. It just summarizes the
information which is stated in these ontologies.
> > But, I agree, this presentation is not the only way to represent
> > linkages, nor perhaps even the best one. ;)
For
instance I tried to "translate" the data which is available in the
Musicbrainz [1] and uses Music Ontology mostly, to the Cyc conceptual
scheme. Even though on the Linked Open Data cloud, there were links
between Cyc and DBpeida on the one hand, and DBpedia and Musicbrainz
on
the other, to my surprise, the traversal from Cyc to Musicbrainz was
not
possible, since the classes were not linked (via sameAs predicate, or
any other).
Then I found the Linked Open Data constellation diagram (on Wikipedia
by
the way), which on the first sight suggests, that there are more class
level links between MusicOntology and Umbel than between Umbel and
Cyc.
This was strange to me, and now I know why.
Still I would be happy to see a diagram which includes not only the
class predicates, but also the sameAs predicates used for classes. I
think it would be very helpful for people, which try to make out most
of
the Linked Open Data. I know there are services such as sameas.org,
but
without some background, verified knowledge about what might be found
there (like the discussed LOD constellation diagram), it is hard to
use
them.
Anyway - if you think that such an idea is worth developing, I would
be
happy to help you with it.
Kind regards,
Aleksander
[1] http://dbtune.org/musicbrainz/