[log in to unmask]">
On Thu, 13 Mar 2003, Andy Powell wrote:
In the OAI world, requirement 1) has been met by the use of 'oai' URIs to
identify each 'item' in OAI repositories. An example URI is
oai:rdn.ac.uk:12345-67890
Many apologies, but having read this back to myself (and checked with the
OAI-PMH spec) I realise that what I said here is incorrect - and I'd
better correct myself before someone else does it for me :-).
In the OAI-PMH, the 'oai' URI is used to identify an 'item'. An 'item' is
'a constituent of a repository from which metadata about a resource can be
disseminated'. So, what this means is that, in the contaxt of a learning
object repository (like, say, Jorum):
- the learning object is the 'resource'
- the set of possible metadata descriptions about the resource (dc, ims,
etc.) is the 'item'
- each specific description (dc or ims or ...) is a 'record'
The 'oai' URI scheme is used to identify the 'item' not the 'resource'.
See
http://www.openarchives.org/
for details.
The rest of my argument still stands I think, namely that we should use
some form of URI to identify learning objects and that we should use the
PURL or Handle systems to resolve those URIs.
So, the question is, what form of URI should we use given that we can't
use the 'oai' scheme (for resources)?
Looking thru the list of registered schemes at IANA, I don't see anything
obvious, other than the 'urn', but I'm not convinced we should bother with
URNs for various reasons. Looking thru the list of unregistered schemes
at
http://www.w3.org/Addressing/schemes
I also don't see much of interest. That leaves us having to invent our
own scheme. One possibility would be to invent a learning object 'lo'
scheme (with a view to registering it in the longer term).
If so, I'd suggest following the layout of the 'oai' scheme, i.e.
lo:DNSdomain:identifer_within_domain
e.g.
lo:bath.ac.uk:12345-67890
Another possibility would be to simply use PURLs, adopting a similar
format, e.g.
http://purl.org/lo/bath.ac.uk/12345-67890
PURLs of this form have the advantage of being able to be independently
assigned by learning object creators (because of the DNS domain part) and
they can easily be mapped to URIs of the form
lo:bath.ac.uk:12345-67890
if necessary in the future, and to 'oai' URIs fo the form
oai:bath.ac.uk:12345-67890
I.e., we could maintain a simple relationship between the learning object
'resource' PURLs and the corresponding OAI 'item' URIs.
I have a strrong dislike of inventing our own URI scheme, so my gut
feeling is to use PURLs but I would really appreciate other's views on
this!
Hope this all makes some sense! If it would help, I could write this up
as a firm proposal for 'learning object identifiers'?
Andy.
The 'oai' part says this is an 'oai' URI, i.e. it conforms to the 'oai'
URI scheme. The 'rdn.ac.uk' bit says that this URI was assigned by the
RDN. The '12345-67890' part is an identifier that is unique within the
RDN. Overall, uniqueness is guaranteed (as long as the 'rdn.ac.uk' part
is unique - which is why the use of DNS domains for this part is strongly
encouraged).
At the moment, there is no 'resolver' service for 'oai' URIs. If I type
'oai:rdn.ac.uk:12345-67890' into my browser Address bar I'll get an error
- because my browser doesn't understand the 'oai' URI scheme.
One could build a resolver service for these URIs by harvesting metadata
records from all known OAI repositories and building a list of all 'oai'
URIs and their associated URLs (possibly taken from the metadata record
for that item). As far as I know, no-one has done this yet - but it would
be quite a neat idea. Then one would need a technology for actually
doing the resolution. It would be possible to use either the Handle
system or the PURL system for this.
For example, one could register all the 'oai' URIs with the PURL resolver
at
http://purl.org/
using a convention like
http://purl.org/oai/oai:rdn.ac.uk:12345-67890
This would be pretty trivial to do. Doing so would allow people to
actually use the 'oai' URIs to link to things.
Doing something similar with the handle system would, I think, be fairly
straight-forward.
OK, so here's my suggestion. We find an existing URI scheme (or if
absolutely necessary invent our own one) that, like the 'oai' scheme,
allows us to idependently assign unique identifiers to learning objects.
We then build a resolver service for those URIs using the PURL or Handle
systems. (Initially, we would not actually run a copy of the PURL or
Handle resolver software locally in the UK - but longer term we could
think about doing that if performance issues deemed it necessary).
If we could assume that all learning object repositories were going to
expose their metadata using the OAI-PMH, then we could do this by using
the 'oai' URI scheme. That would be my preferred approach.
Clearly, even without the need to run our own resolver, there is still a
need to register the 'oai' URIs that we assign with the PURL resolver -
this would have to be undertaken centrally, based on regular OAI
harvesting. Therefore this would have to be a funded 'service'.
Therefore someone would have to make a proposal to JISC (or whoever) to
run this service.
The advantage of this approach is that, assuming that we can agree to it
in principle, we can start assigning 'oai' identifiers now, without
waiting for the resolver to be deployed. I.e. we meet requirement 1) in
the short term and requirement 2) in the medium term?
Comments?
Andy
--
Distributed Systems, UKOLN, University of Bath, Bath, BA2 7AY, UK
http://www.ukoln.ac.uk/ukoln/staff/a.powell +44 1225 383933
Resource Discovery Network http://www.rdn.ac.uk/
Andy
--
Distributed Systems, UKOLN, University of Bath, Bath, BA2 7AY, UK
http://www.ukoln.ac.uk/ukoln/staff/a.powell +44 1225 383933
Resource Discovery Network http://www.rdn.ac.uk/
.
