A few weeks ago I blogged an idea for incorporating a "preferred bookmark link" into web pages to improve the stability of links submitted to social bookmarking sites. The comments were favourable so I think its worthwhile pushing ahead with the idea.

However, I've since decided that my proposed implementation is wrong! Originally I suggested embedding the link in a META tag in the HTML. But it's dawned on me that the LINK tag is obviously a better alternative. The LINK tag is intended to be used to convey relationships between documents, and that's essentially what we're trying to achieve. There's even a predefined link type for indicating bookmark links.

This mechanism is already in use on many blogs to identify the "permalink" for a specific article, e.g:

<a href="...some...url" rel="bookmark" title="Permalink">Permalink</a>.

So I'm going to revise my proposal so that persistent links to academic articles, e.g. DOIs, are embedded into web pages by adding a LINK tag into the HEAD of the document as follows:

<link rel="bookmark" title="DOI" href="http://dx.doi.org/10.1000/1"/>

The system is extensible as we can agree a convention, similar to RSS auto-discovery that the combination of the rel and title attributes convey information about the type of link. For example to include both a stable DOI link and a direct link to the current publisher's website, we could use the following:

<link rel="bookmark" title="DOI" href="http://dx.doi.org/10.1000/1"/>
<link rel="bookmark" title="Publisher" href="http://www.doi.org/index.html"/>

User agents (e.g. bookmarklets and other tools) and social bookmarking sites can then offer the user a choice of which link to use (to avoid security issues) or simply store both.

Labelling DOIs like this also enables them to be more easily extracted for other purposes. We're already including DOIs, expressed as info: URIs, in our embedded Dublin Core metadata, but the actual web link is useful too (if not more so!)