I think the origin of crossref doi representing as abstract "work" came out of an emphasis that equates the print version of an article with the electronic version. Your example of the semantically enhanced article comes close to breaking that equivalency (any digital version of print is to some extent semantically enhanced.) and I think you ought to re-examine the implications.

@eric I agree with you that "any digital version of print is to some extent semantically enhanced". I suppose my overall point is that we need to stop trying to define a hard-and-fast rule that ties the CrossRef DOI to a particular layer of FRBR and instead think about what we are trying to do in citing something. And the rule-of-thumb I'm advocating is nothing more than a variation on Bateson's "difference that makes a difference". Some semantic enhancements make a difference. Some don't. This will be an editorial judgement call, which I expect will make some geeks (including me) very twitchy because it can not really be codified.

"Fundamentally, CrossRef DOIs are aimed at providing a persistent online citation infrastructure for scholarly and professional publishers."

Geoff, I would see that as being the baseline case. DOIs (even CrossRef DOIs) have a more generic role to play in managing and organizing content online.

The view you present above (reference linking) and the concerns you bring up (structuring, versioning) might tend to position CrossRef as essentially nothing more than a global CMS for scholarly publishing. In that view the primary objects are manifestations and would in Web architecture terms be Information Resources. So, for example, two 'equivalent' manifestations (e.g. HTML and PDF) will be managed by the same DOI and can be accessed using regular HTTP content negotiation.

And I can see how versioning concerns (and dynamic content) could lead you even further down that way of thinking.

(Aside: Btw, versioning and dynamic content immediately brings to mind Memento as a candidate technology which essentially utilizes time-based content negotiation.)

On the other hand, object structuring especially seems to drive us IMO in the opposite direction. Here we are principally concerned with a structure that contains substructre which might be recognized as an entity in its own right, i.e. a work within a work.

A instructive working example might be an image within an article. The abstract article (the work) has an identifier - the DOI - and the image (also a work) has its own identifier - the DOI. Now let's say our image has two formats PNG and JPEG. No problem we have a single identifier and use content negotiation to mediate the formats. Now let's say we focus on just one format (JPEG, say) and consider two or three display sizes plus a thumbnail version. Are we going to assign seperate DOIs for each sizing? No, of course not. They will still be managed by a single DOI which will be a work (or possibly expression) identifier - but it will nonetheless be an abstract Non-Information Resource.

Similar considerations can be applied to other component types. Substructure will likely be dealt with using some level of asbtraction. And the same is likely true of versioning too.

It is evidently a more scalable solution to deal with this kind of an object oriented approach (with a single public identifier plus accessor methods, e.g content negotiation, etc) than using separate public identifiers with their own metadata records for individual components and types.

You say that:

"In short, we need to update our guidelines."

Yes - indeed. But this also begs the question of how (or who) CrossRef will consult to arrive at a more informed set of guidelines. I presume that this will be taken to the Technical Working Group for its input?

@tony

I think the "baseline case" is the "de facto case." CrossRef DOIs were originally implemented to facilitate linking of citations (Hence the name of CrossRef's parent organization "Publishers International Linking Association")

I am not as steeped in the history of CrossRef guidelines as some, but it seems to me that use of the word "work" at the time was probably not in the FRBR sense and had more to do (as Eric said) with the notion of equating print with electronic.

I am not at all sure what you mean by "global CMS" but it certainly doesn't sound like anything I would advocate. For one thing, we don't manage *any* content. At best we manage citation pointers to content.

And, as I said above, I'm not sure we should be trying to tie the CrossRef DOI to a particular FRBR layer. Sometimes one cites things at the work level, sometimes at the expression level and sometimes at the manifestation level. I suspect that we are more often concerned about the expression level, but that could change if manifestations start to become too disimilar.

So, as you say, in the "baseline case" you might be able to use content negotiation to access either the PDF or HTML representation of an article. But first we need a mechanism to distinguish between when we are getting the thing pointed to by the CrossRef DOI (the PDF , HTML, etc.) as opposed to "something about the thing" (e.g. the landing page, metadata record, etc.). This might not seem like a particularly ambitious approach- but it seems to me to have the advantage of being pragmatic and roughly aligned with current practice.

And, of course, I think we need to figure out how to best map relationships between citable objects- but that is where I think we might want to draw a pragmatic line. What you seem to be advocating above is that we get into tracking compound documents in their entirety. I'm not sure what business we have keeping track of image thumbnails or other display-related detritus of the publication process. You say CrossRef DOIs have "more generic role to play in managing and organizing content online". That truly does sound like the asset management function of a CMS and that really seems to be stretching CrossRef's mandate.

As we try to figure out how to make the CrossRef DOI play better in the linked data world, we will, as always, consult with our members (including the TWG), knowledgeable people in the industry, etc. As is the case with all of our major projects, we will also form a special working group. And, of course, anything that we do will ultimately be approved by the PILA board.

Just to toss in another thing to think about. Suppose we have an author id (i.e. ORCID) and we want to list the articles by the author. You'd think of it as an error if two sort-of-the-same versions of an article were listed separately. Following this line of thought, you're led to the notion of having stronger typing of crossref doi's.

Contemplating changing the guidelines has impacts on the fee structure as well, I think. From the CrossRef Publisher Fees page:

DOI Conflict Penalty Fee
DOI Conflict Penalty Fee: $2.00 per conflict, to be charged at CrossRef's discretion. As stated in CrossRef's DOI Guidelines (http://www.crossref.org/02publishers/doi-guidelines.pdf) it is our policy that only one DOI be created for a given work. However, it is very difficult to prevent the assignment of multiple DOIs to the same work owing to the reliability and consistency of the metadata provided in a deposit and to accepted publisher practices. For further information please refer to the following document (http://www.crossref.org/02publishers/DOIConflicts.pdf).
A conflict exists when two or more DOIs have been assigned to the same logical work as represented by identical metadata or when subtle metadata differences may exist (e.g. a slight variation in the publication title or style differences in the item title).

Even beyond ejournal/ebook distinctions, there continues to be very little differentiation in the industry between the linking and standards requirements for of different classes of ebooks. Reference ebooks and databases have fallen to the lowest rung of the “books and databases" discussion ladder. And, from Paratext’s standpoint, it is easy to see that lack of engagement with discovery, cross search, and linking of reference works has resulted in declining use of reference in libraries.

Anyone who is intimately familiar with reference will recognize that is essential it is for publishers to assign DOIs that enable durable linking at the article/entry level. This is the only way to facilitate cross-searching and/or discovery in ways that support successful reference research.

For those less reference-minded, this is because the title tells very little about what lies within a reference work. Full-text search does not overcome the problem because it does not interpret context precisely enough to give a researcher what they need at the top of the results list—too much is missed.

For most reference publishers, assigning backfile DOIs at the article or entry level will involve hundreds of thousands of assignments. To the best of my knowledge, the DOI pricing structure is oriented to title-level DOIs. So, the pricing structure breaks down when we get to this volume.

For e-reference materials to function well, publishers should quickly commit to making DOI assignments. However, the business model will need to help them get there. Few publishers facing declining usage will be able to rationalize a significant infrastructure. Would it be simple enough to find a middle-ground with one pricing tier at the title level and fractional pricing for DOIs assigned to smaller chunks?

@eric

Basically, I agree that getting undue credit just because you have sort-of-the-same content identified by separate DOIs could be a problem. I say "could", because this depends on the context, doesn't it? Listing the several major editions of the monograph you have authored might be a good thing whereas listing every instance in which your dataset has changed might not. In both cases, for citation purposes you would want to assign separate DOIs- but whether you list them separately in your publication list depends on the situation.

I also think you might be presuming that the ORCID system would depend exclusively on the fully automatic assignment of publications to authors via DOIs. The current vision of the ORCID system is to combine externally-asserted claims (which might be automated or not) and self-asserted claims and to clearly distinguish between the types of assertions. This will allow third parties (ORCID has no intention of getting into the metrics business. Neither does CrossRef.) to make judgements about what to count and how.

And so we get back to this issue of "judgement." The publisher is in a fairly good position to make a judgment as to whether the difference between A & B is likely to make a difference in the crediting or interpretation of the content (important for citation) However, publishers are not necessarily in a good position to judge how (or even whether) the difference between A & B is important in the context of promotion, grant funding, etc.

So finally, do we need "typed DOIs" to make CrossRef DOIs more tractable for linked data applications? I certainly think we need something, but I'm not sure yet whether it is "typing"- at least not in FRBRish way. But perhaps that is not what you meant?

@alix

Great to see you here. We've actually recently published new guidelines for assigning DOIs to books including reference works:

http://www.crossref.org/06members/best_practices_for_books.html

And we've also just revised our fee structure in order to encourage more granular assignment of DOIs within books.

http://www.crossref.org/02publishers/20pub_fees.html#_2010_deposit

In this latter case, we were particularly interested in making sure what we did would work for reference work publishers (and we had quite a few of those on the working group).

So we are very much focused on increasing the use of DOIs in books, If these revised fees and guidelines don't address your concerns, then we'd be interested in talking to you directly.

@peter

Good point, but aside from wording, I actually think the issues around fees would be largely the same. When content changes to such a degree that it could effect the crediting or interpretation of the work, there is also generally a way to provide distinguishing metadata that goes along with it.

@geoff

I do think we need to separate out a) the issue of "type" of DOI (i.e. is it an Information Reource, or not), from that of b) when to assign a new DOI based on change in content. As you have suggested the second issue b) must generally be left to the publisher to determine, but the first issue a) is something that really needs to be defined at a higher level in order to provide a common semantics.

(Btw, I don't really like the word "type". Seems a bit weak when we are really considering the difference between "Abstract" and "Concrete". But if it serves then it'll do.)