First UMBEL Web Services Released
mkbergman
Fred and I are pleased to announce the availability of the first 11
UMBEL Web services. The sandbox is at:
http://umbel.zitgist.com
These 11 online demos are:
* Find Subject Concepts
* Subject Concept Report
* Subject Concept Detailed Report
* List Sub-Concepts & Sub-Classes
* List Super-Concepts & Super-Classes
* List Equivalent External Classes
* Verify Sub-Class Relationship
* Verify Super-Class Relationship
* Verify Equivalent Class Relationship
* Subject Concepts Explorer
* Yago Ontology.
You can learn more about this announcement in posts by Fred and me:
http://fgiasson.com/blog/index.php/2008/04/20/exploding-the-domain-umbel-web-services-by-zitgist/
http://www.mkbergman.com/?p=438
Please note that we have not yet released the actual service endpoints
or technical documentation, which will be forthcoming shortly.
Thanks, Mike
smonroe
needed, only some way to retrieve the Subject Concept starting point
(i.e. All concepts for "American president").
-sherman
On Apr 20, 11:10 am, mkbergman <m...@mkbergman.com> wrote:
> Hi all,
>
> Fred and I are pleased to announce the availability of the first 11
> UMBEL Web services. The sandbox is at:
>
> http://umbel.zitgist.com
>
> These 11 online demos are:
>
> * Find Subject Concepts
> * Subject Concept Report
> * Subject Concept Detailed Report
> * List Sub-Concepts & Sub-Classes
> * List Super-Concepts & Super-Classes
> * List Equivalent External Classes
> * Verify Sub-Class Relationship
> * Verify Super-Class Relationship
> * Verify Equivalent Class Relationship
> * Subject Concepts Explorer
> * Yago Ontology.
>
> You can learn more about this announcement in posts by Fred and me:
>
mkbergman
On Apr 26, 7:48 pm, smonroe <sdmon...@gmail.com> wrote:
> Btw, is the dataset linked data? If so, then a RESTful API will not be
> needed, only some way to retrieve the Subject Concept starting point
> (i.e. All concepts for "American president").
In internal draft testing there is indeed a Web service with
dereferenced URIs (i.e., Linked Data). This has not yet been released
because:
1. We are still completing ontology technical documentation
2. Cycorp and the Cyc Foundation are reviewing the existing 21 K
subject concepts and making suggestions for additions and deletions,
and
3. About 8 K of the subject concepts still lack definitions (another
critical factors since definitions are the more appropriate basis for
establishing relationships than simple labels).
Though not sufficient to prevent release (as those three factors above
are) we are also simultaneously:
4. Adding to the aliases and synsets ("semsets") for each subject
concept
5. Mapping UMBEL subject concepts to leading RDF ontologies (DOAP,
SIOC, FOAF, Geonames, etc., etc.) with about a 12 already completed,
and
6. Extending the inclusion of named entity mappings from the current
250K or so to about 1.5 M (based on Yago). (BTW, this will be a
constantly expanding list with DBpedia and much of the LOD data soon
to follow as well. :) ).
While we know (and hope!) there will be future versions and are not
seeking perfection by any means at this time, we think Items #1 - 3
are pretty critical for a credible first release of UMBEL.
We're working on these as quickly as possible and I believe we can
assure full initial release will occur within weeks, not months.
If you have any ideas or suggestions for accelerating this schedule
even further, let us know! It took us quite some time to get to this
baseline status, though things are going very quickly now.
Thanks, Mike
>
> -sherman
>
> On Apr 20, 11:10 am, mkbergman <m...@mkbergman.com> wrote:
>
> > Hi all,
>
> > Fred and I are pleased to announce the availability of the first 11
> > UMBEL Web services. The sandbox is at:
>
> >http://umbel.zitgist.com
>
> > These 11 online demos are:
>
> > * Find Subject Concepts
> > * Subject Concept Report
> > * Subject Concept Detailed Report
> > * List Sub-Concepts & Sub-Classes
> > * List Super-Concepts & Super-Classes
> > * List Equivalent External Classes
> > * Verify Sub-Class Relationship
> > * Verify Super-Class Relationship
> > * Verify Equivalent Class Relationship
> > * Subject Concepts Explorer
> > * Yago Ontology.
>
> > You can learn more about this announcement in posts by Fred and me:
>
> >http://fgiasson.com/blog/index.php/2008/04/20/exploding-the-domain-um...
>
Frederick Giasson
> Btw, is the dataset linked data? If so, then a RESTful API will not be
> needed, only some way to retrieve the Subject Concept starting point
> (i.e. All concepts for "American president").
>
To add to what Mike said, sure that the instances and the ontology will
be accessible on the Web. However, the goal of these web services is not
to take what in RDF and to publish in a custom XML format. The goal is
to access the inferencing, and searching capabilities of the systems.
So, you query the web services, you get a list of subject concepts in
the resultset and then you can check in a local repository for their
definition, or dereference them on the Web.
So, the web services doesn't replace the dereferencing capabilities of
the URIs of the subject concepts.
Hope this help clarifying this point.
Thanks,
Take care,
Fred
smonroe
I sent a message before this one, which doesn't seem to have made it :
(
So I will attempt to repost from memory before replying to your latest
reply.
First off, this is very impressive work! And congratulations on this
first public release. After reading and interacting with the demos,
and have a much better understanding of the critical resource UMBEL
provides for Cypher. I would like to begin experiments with using
UMBEL as a lexicon (i.e. dictionary of Named Entities and Subject
concepts) for Cypher. I've already began devising a pluggable lexicon
architecture to facilitate this. I think my starting point would be
the Subject Concept lookup service. Some questions:
- Is there an ontology for UMBEL?
- For this example:
http://umbel.zitgist.com/view.php?concept=USCity
How is the synonym list constructed (from wordnet, cyc)? If I submit
the service one of these synonyms, will I retrieve this URI? I copied
and pasted on of them into the search interface and got an error page
followed by
"city in the u.s." is not an existing UMBEL subject concept.
Search for a related concept by clicking here
Is that a bug or a characteristic of the lookup processor?
- I think the "search related concept" option is very useful. But how
exactly am I to interpret those results? I.e., can you qualify the
relations of those results to the original search term(s)?
As for the Cypher interface, I image that Cypher would retrieve a
concept URI by submitting a word or short noun phrase (so I will need
to know what syntax to you require, e.g. do you process allomorphs of
a term such as plural forms, or does the service prefer the root form
of the word). If the output of the service is linked data, then I
could just begin transversing the graph and use those statements to
enhance semantic filtering/disambiguation. If RESTful XML output, then
I could do the transversal that way. At this point, as a consumer it
doesn't matter, but I think linked data would be the best approach
moving forward.
I predict that the impact on improvement to the quality of Cypher
output will be considerable. Kudos for taking on this important
effort.
Cheers,
-sherman
smonroe
Thanks, this clarifies a lot. I will continue then to bootstrap my
learning about UMBEL, YAGO, etc as we await the next releases. I think
my input will be most constructive after I'd had the chance to
retrieve a term/definition from Cypher.
But again this is an excellent first start. Once the foundation has
been laid (which it appears you're in the middle of), then the Law of
Increasing Returns kicks in, and I'm sure we can begin to expect
increasingly greater improvements within shorter release cycles.
The automated extraction seems to be performing quite well. Do you or
have you consider the possibility of crowd sourcing to help fill gaps
in the metadata? Do any of the above points #1-3 rely on manual entry?
Also, UMBEL covers nouns, have you considered to venture into modeling
verbs (e.g. semantic frames, Cyc microtheories, etc) and other parts
of speech?
Thanks,
smonroe
Wow, this is a very key distinction, and should be made salient in the
documentation. Most likely, or in an ideal world, it would return
dereferancable URI's and user could fetch the resource description
from the web (then perhaps cache the description in a local
repository). Thanks for clarifying. I'm really anxious to begin the
integration with Cypher.
-sherman
Frederick Giasson
> - Is there an ontology for UMBEL?
>
>
Sure, it is to come.
In fact, an ontology (describing classes and properties) and an
instantiation file of the subject concepts (their descriptions).
> - For this example:
> http://umbel.zitgist.com/view.php?concept=USCity
>
> How is the synonym list constructed (from wordnet, cyc)? If I submit
> the service one of these synonyms, will I retrieve this URI? I copied
> and pasted on of them into the search interface and got an error page
> followed by
>
With the find label service, you will get a list of potential matches.
Check the API page for more information.
> "city in the u.s." is not an existing UMBEL subject concept.
> Search for a related concept by clicking here
>
>
Normal, "city in the u.s." is an alias for USCity. This string, amongs
many others, denode the subject concept USCity. This is what we call the
semset. However, the documentation (to come too) will explain all these
differences.
But basically, you have one subject concept (with URI:
http://purl.org/umbel/sc/USCity) with multiple labels that refers to the
concept.
> - I think the "search related concept" option is very useful. But how
> exactly am I to interpret those results? I.e., can you qualify the
> relations of those results to the original search term(s)?
>
This is really just to help people to find subject concepts they are
looking for. I am not sure this is the good option for what you are
trying to do (making a software agent finding concepts of some kind?). I
do think this is good for a semi-supervised task.
> As for the Cypher interface, I image that Cypher would retrieve a
> concept URI by submitting a word or short noun phrase (so I will need
>
It would get a list of URI, yes.
> to know what syntax to you require, e.g. do you process allomorphs of
> a term such as plural forms, or does the service prefer the root form
> of the word). If the output of the service is linked data, then I
> could just begin transversing the graph and use those statements to
> enhance semantic filtering/disambiguation. If RESTful XML output, then
> I could do the transversal that way. At this point, as a consumer it
> doesn't matter, but I think linked data would be the best approach
> moving forward.
>
You will get a list of URIs, then you will be able to dereference these
URIs to get their descriptions, and their linkage with external ontologies.
> I predict that the impact on improvement to the quality of Cypher
> output will be considerable. Kudos for taking on this important
> effort.
>
This is a possibility yeah, we will see over time and experimentation :)
Take care,
Fred
Frederick Giasson
> Thanks, this clarifies a lot. I will continue then to bootstrap my
> learning about UMBEL, YAGO, etc as we await the next releases. I think
> my input will be most constructive after I'd had the chance to
> retrieve a term/definition from Cypher.
>
Yup, once the documentation, the ontology and the instantiations will be
made available, things will be clearer. The first goal of releasing
these web services was to let people start to check the UMBEL data model
and its first linkage with external ontologies. Once the doc will be
released, people will be able to start developing with it.
> But again this is an excellent first start. Once the foundation has
> been laid (which it appears you're in the middle of), then the Law of
> Increasing Returns kicks in, and I'm sure we can begin to expect
> increasingly greater improvements within shorter release cycles.
>
Yup, it is what we are working toward.
> The automated extraction seems to be performing quite well. Do you or
> have you consider the possibility of crowd sourcing to help fill gaps
> in the metadata? Do any of the above points #1-3 rely on manual entry?
> Also, UMBEL covers nouns, have you considered to venture into modeling
> verbs (e.g. semantic frames, Cyc microtheories, etc) and other parts
> of speech?
>
>
Currently all subject concept come from 25 years of work at Cyc, 6
months of cleaning by Zitgist, and a new update (to come) of new
definitions of subject concepts from Cyc, and this update is reviewed by
humans.
Will we crowd sourcing in the future? This is a possibility yeah.
About other parts of speech, it is not in our current plans, but a
future possibility. However this won't appear in the UMBEl core I think
since we really model "subject concepts".
Take care,
Fred
Frederick Giasson
> Wow, this is a very key distinction, and should be made salient in the
> documentation. Most likely, or in an ideal world, it would return
> dereferancable URI's and user could fetch the resource description
> from the web (then perhaps cache the description in a local
> repository). Thanks for clarifying. I'm really anxious to begin the
> integration with Cypher.
>
1- URI dereferencable on the Web
2- UMBEL ontology available
3- UMBEL subject concepts description available too
4- UMBEL linkage to key external ontologies (I forgot to talk about this
one).
All this will be released on time.
Take care,
Fred