[rdfweb-dev] why wordnet is cool...

Libby Miller Libby.Miller at bristol.ac.uk
Fri Aug 8 10:58:27 UTC 2003



On Fri, 8 Aug 2003, Victor Lindesay wrote:

>
> Thanks Dan.
>
> Your example is neat.
>
> So my RDF agent sent to get pictures of clothing (with baked in
> knowledge about http://xmlns.com/wordnet/1.6/Clothing), will discover
> http://xmlns.com/wordnet/1.6/Cowboy_hat as a value of the
> http://rdfweb.org/people/danbri/media/2003/08/1/376j0234.jpg ->
> foaf:depicts -> http://xmlns.com/wordnet/1.6/Cowboy_hat triple, retreive
> the schema using the value directly as an endpoint and find  chain of
> derivation of  http://xmlns.com/wordnet/1.6/Cowboy_hat from the sought
> term http://xmlns.com/wordnet/1.6/Clothing. Bingo mission accomplished!
>
> Is that how it works? The only problem I foresee is that although you
> kindly allow this type of processing though the
> http://xmlns.com/wordnet/1.6/ address, most class and property uris are
> not physical locations for the schemas my agent needs at run-time.
>
> Perhaps a need for a web service (in the loosest sense) to provide
> schemas or schema information to RDF agents.
>

Hi Victor

I'm using Dan's service as a sort of web service for annotating photos
using a javascript tool. It's surprising how often you can get a hit
just from guessing a noun and then adding it to the
http://xmlns.com/wordnet/1.6/ namespace. The main issue there is just
that you don't find 'nearby' words - flowers and flower for example,
or Graffiti and Graffito, or different senses of a term (you have to
make a separate hit for each of: wn:Dog, wn:Dog-1 etc, and you don't
know how many versions there are).
It's simple to get the more specific terms, as they come in the
definition of a word.

For my search tool, the main issue is that a wn:Sky_scraper, say doesn't
come up in the search as a wn:Building, because it's not easy to get the
information connecting the two in the heirarchy as you catalogue the
information. Probably the best bet there is to do what Matt Biddulph has
done and download the wordnet dataset and precompute the wordnet
categories for each image, so that even if the cataloguer of the image
doesn't specify that it's a building as well as a skyscraper, the search
tool does have that information.

Having said that, even without those features, the wordnet service as
it stands is very useful. I love cataloging photos now*, and being able
to search for, say, 'graffito'

http://swordfish.rdfweb.org/discovery/2001/08/codepict/
http://swordfish.rdfweb.org/discovery/2001/08/codepict/querywn.jsp?wn=graffito

is pretty nifty.

Using Dan's wordnet service, and also nearestAirport data for location
(or spacenamespace), or my RDF database to search for people, all using
a simple get on a url and returning RDF/XML makes the process of creating
image metadata fast and lightweight.

My cataloging tool is here:
http://swordfish.rdfweb.org/discovery/2003/06/codjsform/shell.html

and a similar version is here:
http://swordfish.rdfweb.org/discovery/2003/06/codepictjs/shell.html

I'm in the process of cleaning up the underlying javascript code.
Underneath it uses Jim Ley's RDF parser, which means it only works with
Mozilla and IE at the moment, because it requires an XML parser and the
ability to call get on a url.

cheers

Libby


* seriously :)







More information about the foaf-dev mailing list