I just read an interesting article by Lance Fortnow about how (and why) publishing in CS differs from most other disciplines. Beware that it is a bit ACM biased. 
Open Access by Gideon Burton (cc-by-sa)
I agree that the current situation is not perfect. E.g. the focusing on deadlines has disadvantages (but also advantages). And I must say, that the most useful parts of most conferences are the coffe breaks and birds-of-a-feather sessions. I guess something like the Linked Data Meetup would at the same time be cheaper and more enjoyable.
On the other hand I am not sure whether the best solution would be to change to a journal centric model as in other disciplines. Fortnow criticises the “least-publishable units”, which end up as conference papers. On the other hand, working in small increments is a main feature of lean processes, and also start-ups (cf. Eric Ries “minimum viable product”). For example, FOAF+SSL has been developed publicly, on the web. As Fortnow explains, the current model of publication in CS has emerged as a result of the agility of the discipline. It would be a pity to sacrifice this agility.
I guess what we need is a mixture: Web 2.0 style work within sub-communities, publishing early and often. And then maybe some kind of journal (or better open access) style process for disseminating results across communities.
Cheers, Simon
Interesting point about “lean,” one I haven’t heard used in this particular context. But I think that lean ideas do not apply to most CS research: (most) research doesn’t cater to rapidly fluctuating demands or need to be user-tested the way that, say, cars or software do. Lean is a great engineering and business technique, but it solves problems that most research — especially theory research — just doesn’t have.
Link | August 28th, 2009 at 5:32 am
Josh, good point. I also have been thinking about that. But I guess, to a certain degree, lean is well applicable. Although demands do not change quickly, you need to find out what may work and what not and which topics are worth while and which not. This would be the main souce of uncertainty in research. The main point is not to react to demand, but to make failures cheap and to progress fast. Let us assume I had a great idea, but somebody else has done exactly the same, ten years ago, but under a different title. I would be extremely happy to learn about that after two weeks instead of during a review three months from now.
This kind of feedback would be more likely if you publish early and often. The same applies for wrong assumptions underlying your approach, good ideas for evaluation measure etc. And even if everything goes well, early feedback will allow you to progress faster. Does that sound reasonable?
Link | August 28th, 2009 at 8:27 am
That does sound reasonable. But now I think you’re making an argument FOR a more journal-centric model than against it. All of the items you just listed would be better served if anyone who wanted could speak at a conference, without having to worry if their idea was suitable for publication in a top venue — in other words, the math model of conferences.
(Understand that when I say “journal-centric” I mean that journals should be the reputation/publication centers, not the centers of disseminating results.)
In fact, the example you gave happened to me *exactly because I could go to a non-publishing conference*. I started studying a question, got to present it at Midwest Theory Day (a mostly-graduate-student, anyone-can-speak, no-proceedings, one-day, biannual conference), and someone in the audience told me he had worked on it twenty years ago. (Luckily we still had some new results.)
Link | August 28th, 2009 at 4:13 pm