Less is definitely more

After suffering through a big, steaming pile of (mostly) poo of 18-page(!!) paper submissions for a conference I shall not name, I was ecstatic to discover that my next batch of reviews consisted of 3-page papers. A bit on the other extreme but overall I think that, at least in data mining / machine learning, lower page limits are appropriate for conference submissions (journals are another story).

Say, one page for a short and to-the-point introduction and motivation, one page for the model and equations, and a couple of pages for main experiments and plots. This should be sufficient to get at least the main idea across, without distractions. Anything more than that risks diluting the message, unless the authors are very skilled and careful writers. Further details do have value, but they could be an extended appendix or, even better, an online demo.