Re: Final CFP: In-Use Track ISWC 2013

Norman Gray <norman@astro.gla.ac.uk> writes:
>> I am not completely familiar with DOI. Am I right, that it more or less
>> provides the same service as http://purl.org .
>> DOI links on the resource-level. You would still need frag ids to link to parts.
>> Firefox can actually handle this:
>> http://dx.doi.org/10.1038%2Fscientificamerican1210-80#atl
>
> It's not the same thing as purl.org.

The mechanism by which DOIs and purls are resolved is more or less
identical. Under the hood, DOIs use handles, purl.org uses a triple
store. In practice, users don't interact with either directly.

> A DOI (parsed as "digital (object identifier)") is an opaque ID for an object
> of some time, which you look up in a distributed registry of resources. Thus
> your example of doi:10.1038/scientificamerican1210-80 is a name for that
> article. DOIs can also be looked up using the dx.doi.org service, but that's
> just a convenience interface to the underlying API, which is based on the
> broader-remit Handle system. Since there are no fragment IDs defined in the
> doi: URI scheme (as far as I recall), there's no meaning can be attached to
> the fragment in the dx.doi.org HTTP URI.
>
> It's also -- I _think_ -- not specified what precisely it is that the DOI denotes.


This is the same as purls, after the DNS part of the system. Of course,
anyone can set up a new purl server; and the domain name of this depends
on DNS. Strictly this is not true of DOIs, although it is true of DOI
URIs (http://dx.doi.org/10.xxxx)

> The other big difference is that DOIs cost actual money, of the order of
> $1/DOI, though there's lots of variation. This is the sustainability model for
> DOIs: if one registry disappears, others can take over.
>
> The most common objects which are given DOIs are journal articles, of course,
> but there's currently a lot of effort going into the detailed mechanics of how
> you acquire a DOI for a dataset, what precisely that means, and what the cost
> model should be for registering DOIs in this context and in these numbers. See
> <http://www.datacite.org>


This really depends on the registration agency of which there are 8.
CrossRef DOIs for subparts can cost as little as 0.06$. DataCite DOIs
come with a different set of guarantees to CrossRefs as far as I can
see. So, CrossRef provides a guarantee of one DOI to one object, which
DataCite doesn't. I *think* datacite says "what is resolved doesn't
change", while CrossRef only says "it should maintain it's logical
identity". 

All DOIs provide metadata, although only at the Handle level. DataCite
and CrossRef DOIs also do content negotiation in the HTTP format;
unfortunately, at the HTTP level it is not possible to distinguish the
different DOIs from each other. The metadata you get back is not
entirely standardized, even between crossref and datacite.


>
> That's the edited highlights: more details at <http://en.wikipedia.org/wiki/Digital_object_identifier>
>
>> If I am right, DOI also wouldn't be able to provide links to the 40
>> million mentions contained in the Wiki links corpus:
>> http://techcrunch.com/2013/03/08/google-research-releases-wikilinks-corpus-with-40m-mentions-and-3m-entities/
>> That's 40 million DOIs ....
>
> I don't there would be such DOIs, unless someone has spent quite a lot of money registering them.


A purl would be much better in this case anyway, since purls support
partial redirection, which to my knowledge, DOIs do not. With DOIs you
would need 40 million DOIs. With purls, you would create a single
partial redirect purl and handle the rest locally.

Phil

Received on Thursday, 9 May 2013 12:09:23 UTC