Doug Miller’s written about the categorizer he and Frank Tansey cooked up at Tinderbox Weekend SF.
Even if you’re not familiar with Tinderbox, it’s worth a read.
The genesis of the method was in a couple of exercises where we were given collections of nodes in a Tinderbox hypertext and asked to group them. It was easy enough to assign a keyword to a node once you’ve searched on a term, but he wanted a way to assign new keywords without clobbering ones already there.
Categorizing after you’ve collected some data ought to reduce the problem where you create taxonomy as you add entries, and wind up with an explosion of categories: see this weblog and del.icio.us for examples.
