1. CiteULike is a social bookmarking system oriented towards academic publications. When I find an interesting article on PubMed or Ingenta, for example, or when I learn about a good book on Amazon, I click on the CiteULike bookmarklet which automatically fills a full bibliographic notice for the reference and adds it to my library.

   
   A sample from my CiteULike library. Stars indicate the priority-to-read rating (a tick means the reference has already been read). Books are automatically added with their cover. The usual folksonomy functions (shared keywords, and n others…) are featured.

   CiteULike’s added values to a plain reference list are the shared keywords, the priority-to-read rating and the capacity to attach notes to the reference. One of the most interesting capabilities may be the PDF uploader function, which makes it possible to attach the full article to the reference, directly on the CiteULike server.

   
   The article on the top of this screenshot has been annotated. The bottom one comes with its PDF file. CiteULike is not limiting file size or total storage capacity at the moment.

2. Selections of references (by user, keyword, or author) can be watched, or exported to EndNote (RIS), to BibTex (BIB), or even followed through RSS syndication.

   
   This example selects CiteULike references through two criteria: user (phnk) and keyword (public_health).

   Below, I have exported this selection to the BibTeX format, and will use BibDesk to manage it further. BibDesk is a free BibTeX files manager, with a nice intuitive Mac OS X interface and a lot of interesting features (such as automatic filing and export capabilities to formats like HTML, RTF, Microsoft Word…).

3. The BibTeX export function is far from perfect but it is reasonably efficient. The following screenshot shows that references were exported thoroughly (abstracts, notably, are present in the BibTeX file) and without any encoding problems (for instance, Aïach was correctly exported).

   

4. BibDesk has a batch function that generates smart cite-keys for references. This function will correct one of the few weaknesses of the CiteULike-to-BibTeX export process: the cite-keys are quite lame, since CiteULike has no cite-key guessing capability.

   

5. My workflow currently ends at this stage. So far, it is still handicapped by pretty serious bugs, such as the problem with keywords export in CiteULike (an article tagged with political_science and health_policy will be exported with a single, nonsensical political_science health_policy keyword in BibTex). But the main necessities are already there.

   
   The cite-keys problem is solved, but the keywords problem remains at this stage.

This workflow is close to perfect, only little improvement is needed to make it fully functional. Overall, I feel very confident in the efforts of both the developers of CiteULike and BibDesk to make their tools interact properly. Perhaps the Local-URL autofiling function in BibDesk could be articulated with the PDF upload in CiteULike.

The reverse operation, importing from BibDesk to CiteULike, is extremely simple too:

1. BibDesk can place any number of references into the clipboard, as BibTeX or minimal BibTeX records:

   
   BibDesk has many more commands available through right-click and drag-and-drop than simply copying records..

2. In parallel, CiteULike can either import .bib files, or parse BibTeX code from a text box:


Similarly to what can be achieved with the Flickr uploadr, for instance, the user can batch-assign tags and reading priorities to the references he imports.

This last process seems to be very clean: for example, URLs and BibTeX keys are perfectly imported. Alas, the killer-trick that would automatically upload files to CiteULike using Local-URL data from BibDesk files is not yet implemented, and I do not think this is actually conceivable in any way. Let aside this tiny drawback, the correspondence between CiteULike and BibDesk is truly amazing.

Update, 13 October 2007: recent BibDesk nightlies contain a new Web import tool that automatically scans Web pages for BibTeX references. It seems to work fine with CiteULike, although it ‘stalled’ after my first test.