In message <[log in to unmask]>, Paul Walk
<[log in to unmask]> writes
>
>However, I've always been a little unconvinced by the 'Cool URIs don't
>change' piece (http://www.w3.org/Provider/Style/URI) linked to by the
>W3C Note you referenced. While I agree with the sentiment, I think the
>W3C statement does not quite gel with my practical experience.
>
>The gluing together of the object identifier with the organisational
>domain part of a URL is seen by some as a potential problem, and is
>addressed for example in John Kunze's paper on ARK
>(http://colab.mpdl.mpg.de/mediawiki/images/e/eb/ESci08_Sem_1_n2t_a_viabl
>e_pid_solution_Kunze.pdf )
>
>The W3C would maintain that this is unnecessary. I tend to lean
>towards the W3C in this respect - but really because I think the
>deleterious effect of changing URIs can be overstated, so it's not
>necessarily worth introducing another layer of complexity.
Paul,
Your comments on feasibility are well taken, but I think that in the
particular context of the Semantic Web, stable URIs are pretty central
to whether the whole thing will work at all. This isn't just related to
whether they resolve to a physical resource (the point John Kunze
discusses in the presentation you quote): not so long ago the same job
was to be done by "URNs", which didn't (and in fact couldn't) point to
anything.
The other, crucial, job that stable URIs do is that they become a common
currency, expressing a well-defined class or property. People use them
in their own RDF, and the magic of Semantic Web data linking starts to
happen. If your RDF instances are just strings (rather than rdf:about
links), it's effectively just a local resource. Ditto if all your
rdf:about links just point to somewhere else in your own data set.
The above comments are based on my own experiments in this area, and may
well be misguided or plain wrong. If so, I would be delighted to be put
right. Meanwhile, let me explain what I have been up to since the last
message in this thread ...
Just to see what is involved in implementing the W3C advice on Cool
URIs, Open Data, etc., I have built a front end to the Wordsworth Trust
to expose their collection as RDF. While this is very much a work in
progress, it has got to the point where there is something to see.
The whole thing is built on the basis that none of the resources
identified by the URLs actually exist (!) ... in other words the
redirection work is all done by a customized "404 page not found" error
handler. (Since the server in question is IIS, this is an Active Server
Page.) The object URLs all have the form:
http://collections.wordsworth.org.uk/object/[identity number]
e.g.
http://collections.wordsworth.org.uk/object/GRMDC.C104.15
If you request this URL in a normal web context, you are redirected to
the object's page on the standard site (try it). If your HTTP request
includes the header "Accept: application/rdf+xml", it will instead issue
a "303 See Also" response, and give the URL which delivers RDF:
http://collections.wordsworth.org.uk/object/data/GRMDC.C104.15
You can request this in your browser, and if you have something like the
RDF Tabulator extension to Firefox, you can see it as a nicely rendered
table of properties and values. Otherwise you'll get an RDF XML dump -
always a pleasure ;-) (BTW, I don't get anything using IE - maybe I'm
not setting the MIME type properly.) You can also open it in the RDF
browser provided by OpenLink:
http://demo.openlinksw.com/DAV/JS/rdfbrowser/index.html
Once it's loaded some triples, click on the "Raw triples" tab to see
them.
Within my transform, I look up the most specific place name in Geonames,
and if I get a good match I include the Geonames URL in place of the
keyword. If you follow the "data link" after the Geonames URL, this
loads information about the place, and then you can switch to the "Yahoo
Map" tab and see it as the familiar pin on a map. Load a few of the
C104 series in the same way, and you can see the distribution of this
set of prints - all without doing extra work on the catalogue record.
Similarly, by using the W3C Calendar namespace
(http://www.w3.org/2002/12/cal/icaltzd) for my dates, I get entries on
the timeline.
The RDF is produced by an XSLT transform on the live source record
(which is already XML). At present it almost totally fails to meet my
criterion for successful Open Linking, but does raise some interesting
issues:
1. The properties I use are mostly taken from DBpedia. They are
effectively what has been scraped from Wikipedia. Should the museum
community be publishing an ontology of properties that it is interested
in?
2. Are we interested in finding or creating ontologies for common stuff
like object type ("simple name") and materials?
3. How do we deal with the biggies? - people are the obvious one.
In general, the main conclusion I have drawn from this exercise is that
if we are to contribute to, and benefit from, this initiative, we need
to look way outside our own little community of practice.
Thanks for reading if you got this far ...
Richard
--
Richard Light
XML/XSLT and Museum Information Consultancy
[log in to unmask]
**************************************************
For mcg information and to manage your subscription to the list, visit the website at http://www.museumscomputergroup.org.uk
**************************************************
|