RE: DOI and other identifiers


If there is a purpose for a DOI on an annotation within a particular domain (e.g., scholarly journals—I can imagine use cases where this could be useful) that is a business issue within that domain, and it's managed by specific services in the ecosystem (e.g., Crossref). In no way should that be the canonical identifier of the annotation in the first place.

From: Dan Whaley []
Sent: Saturday, May 07, 2016 2:05 PM
To: Robert Sanderson
Cc: Web Annotation
Subject: Re: DOI and other identifiers

On Fri, May 6, 2016 at 12:02 PM, Robert Sanderson <<>> wrote:


On the call today we briefly discussed the use of other identifiers for annotations, such as DOIs.

While there's no problem assigning a DOI to an Annotation, assuming that CrossRef or some other registration agency is willing to manage the potential drastic increase in registrations, there are some questions it brings up for the working group.

While I think the number of DOIs would certainly increase (if scholarly annotation takes off) I imagine that DOIs would only be issued on request, not 1:1 for all annotations created.

* Is the DOI the canonical identifier for the Annotation?

So I may be off base here, but I think there are perhaps two different senses of the word "canonical" at play here.

From an annotation systems perspective, it seems unlikely that the DOI is ever going to be canonical in the sense that it becomes the *primary identifier* replacing the one we minted originally.  We'll want to use a consistent identifier for all our annotations internally, not different ones depending on whether a DOI was issued.  (What if someone captured the URL of the annotation *prior* to the DOI issuance?  We can't ourselves fail to resolve the "old" address of the annotation.)  I assume there may even be performance issues underlying this.  This is perhaps more true of annotation systems than regular publications because annotations would be born without DOIs and presumably get them later, and I'm not imagining that would change.  Otherwise we'd be issuing DOIs for every trivial annotation from inception, and that indeed would be massive.

Even assuming a DOI has been issued for an annotation, if someone else comes along and wants to tweet out the same annotation, and exposes our share dialog to get the link, are they going to care whether there is a DOI, and if there is one, is that the one that they necessarily want to use? (I'm presuming if a DOI was issued, we'd show both the original style, and also the DOI side by side) The tweeter doesn't really care about permanence, and they'd probably just opt for the link style they're familiar with (in our case,<><TOKEN>).  That one will also be more performant since it doesn't have to go through a resolver first.

I'll note that even for classic publishers, DOIs aren't even always canonical in the sense that the publisher themselves uses them internally.  For instance, eLife doesn't use the DOI when they reference their articles, like this one:

However, PLOS does:

So, in what sense are DOIs canonical?  Well, from a journal's perspective if you're going to cite another work in your article, they would probably prefer that you use a DOI if one is available.  From their perspective the DOI is preferable (canonical?) because there's a commitment to permanence implied. Since the journal article can't (or rarely does) change, and is often printed or copied as a PDF, it's probably better that any identifiers that might change later go through a resolver that can arbitrate.

If it isn't, then why mint one at all? To me, it defeats the purpose to have a DOI if it's not the canonical identifier for the resource.  The value of DOIs is when the publisher of the content changes, the citations and references remain the same.

If it isn't, should we have a place in the model to capture it?  Currently there's only the URI of the current location (id), the canonical URI (canonical) and the URIs of other locations from where the current representation was derived (via).  As the annotation would be new, it's neither id nor via.
I agree with Doug that non canonical aliases are best served via an extension, and would resist adding in an 'alternate' field to the core model, as it serves almost no purpose -- if you have the annotation JSON description to read the alternate field, then why do you need to know where else you can get the same JSON from?

From my perspective, no change is needed, but it would be good to discuss :)

* There was also some discussion around versioning.  From the DOI FAQ:

  7. If I have assigned a DOI name and I make a change to my material, should I assign a new DOI?

The IDF does not have any rules on this. Individual RAs adopt appropriate rules for their community and application. As a general rule, if the change is substantial and/or it is necessary to identify both the original and the changed material, assign a new DOI name.


Rob Sanderson
Semantic Architect
The Getty Trust
Los Angeles, CA 90049

Received on Saturday, 7 May 2016 20:59:42 UTC