Re: Request to register "identifier" relation type

[Michael Nelson <[e-mail]>]

Hi Bjartur,

I reply inline below, but I also encourage you to check out a new blog 
post that we did that gives some more examples:

http://ws-dl.blogspot.com/2017/08/2017-08-07-relcanonical-does-not-mean.html

as well as the draft itself which covers additional scenarios:

https://datatracker.ietf.org/doc/draft-vandesompel-identifier/

> If the content is different parts of a paper, then the subpages should <link rel=up
> hef="/to/the/paper" >, should they not? 
>

They certainly can; navigation is orthogonal to identification.  Of 
course, if they page hierarchy is multiple levels, then the URIs used for 
things like "up" and "identifier" will likely diverge.

> The DOI is the canonical IRI. There may be a couple of alternate representations of the
> whole paper (HTML, PDF,  and paper). Then there are parts with their own IRIs (images,
> tables, even chapters). The parts should link up to the DOI (explicitly using rel=up).
>

Dereferencing a (http version of a) DOI doesn't always lead you to an 
"article".  Most of the time it leads you to a landing page, from which 
all the various parts are available (and are somehow collectively 
identified by the DOI).

Dereferencing:

https://doi.org/10.1145/3041656

leads to:

http://dl.acm.org/citation.cfm?doid=3077622.3041656

but what we'd conventionally consider the "paper" (in this case, available 
only as a PDF) is really at:

http://dl.acm.org/ft_gateway.cfm?id=3041656&ftid=1879413&dwn=1&CFID=593352672&CFTOKEN=50692319

If the PDF claimed:

Link: <https://doi.org/10.1145/3041656>; rel="canonical"

Then it would be telling Google et al. to index the content at 
https://doi.org/10.1145/3041656 (and then ultimately 
http://dl.acm.org/citation.cfm?doid=3077622.3041656) *instead* of the 
content at the PDF.

The PDF can claim:

Link: <http://dl.acm.org/citation.cfm?doid=3077622.3041656>; rel="up"

Because the HTML landing page really is the parent doc for the PDF.  In 
summary, the PDF "should" claim:

Link: <http://dl.acm.org/citation.cfm?doid=3077622.3041656>; rel="up",
   <https://doi.org/10.1145/3041656>, rel="identifier"

And the landing page should claim:

Link: <http://dl.acm.org/citation.cfm?doid=3077622.3041656>; rel="canonical"

to distinguish itself from all the crazy URL arguments that the ACM DL 
adds, like this real URL extracted from a search session inside acm.org:

http://dl.acm.org/citation.cfm?id=3041656&CFID=593352672&CFTOKEN=50692319

Which is really just the same page with a different URI.

> To me, rel=identifier sounds like a hypernym of alternate and up specifically intended for
> pointing to the nearest DOI. If it's only intended for DOIs, why not just call it rel=doi?
> In non-DOI instances the distinction between alternate and up seems important. 
>

But it's not just DOIs.  arXiv doesn't use DOIs, and it's a great 
candidate for this.  Consider:

https://arxiv.org/abs/1304.5213

This data set is clearly not rel="canonical" for arXiv:1304.5213:

https://arxiv.org/src/1304.5213v1/anc/dataset.csv

It could claim "up", but that's really for navigation.  Consider the DVI 
version:

https://arxiv.org/dvi/1304.5213

It should claim:

Link: <https://arxiv.org/format/1304.5213>; rel="up"
       <https://arxiv.org/abs/1304.5213>; rel="identifier"

It can't claim to be canonical or alternate for 
https://arxiv.org/abs/1304.5213, because https://arxiv.org/abs/1304.5213 
is an HTML landing page, not the PDF or PS version: so the DVI is not an 
alternate version of the landing page, and assuming Google "wanted" to 
index DVI, the HTML landing page is not a replacement version for the DVI.

The rel="up" link is there because the DVI is becaue the link for the DVI 
page is really only from https://arxiv.org/format/1304.5213, *not* from 
https://arxiv.org/abs/1304.5213.

You can do things like rel="up up", but that's even more clearly for 
navigation and not identification.

The summary is that if you wanted to create a single link to the resource 
in arxiv with the title "Carbon Dating The Web: Estimating the Age of Web 
Resources", then:

https://arxiv.org/abs/1304.5213

really is the best available single URI you can use, all the constituent 
resources (e.g., PDF, DVI, csv) are not canonical or alternate for each 
other, and since the arXiv model can go two layers deep, using "up" 
won't always get you back to "https://arxiv.org/abs/1304.5213".

The blog post linked above also covers a Wikipedia example, which I won't 
repeat here.

> To be fair, RSS and Atom go so far as using alternate where index would be more appropriate.
> So specifying the desired semantics explicitly, under a new rel, and then moving to existing
> ones seems like a more forwards-compatible strategy.

Yeah, RSS and Atom are kind of grandfathered in with the tuples of 
rel="alternate" and the corresponding mime types.  Of course, back in the 
syndication days the RSS feed was often a legitimate "alternate" 
representation of the corresponding HTML page, but that sort of diverged 
through time.

regards,

Michael

> 
> Regards, 
> Bjartur Thorlacius
> 
>

----
Michael L. Nelson [e-mail] http://www.cs.odu.edu/~mln/
Dept of Computer Science, Old Dominion University, Norfolk VA 23529
[phone] [phone] (f)