So it'd be cool if the find-similar-users tool took into account how common various interests are. A bunch of people are listed as close matches because of shared interests in computers, dvds, linux, movies, mp3s, music, perl, and programming. Yeah, me and every other geek on the planet. I'd scale down the importance of any given interest in proportion to the logarithm of its popularity; it's certainly more interesting to me if someone is into "kubrick" (94 matches) than "movies" (15000 or so), and a fellow "haskell" user (2 matches) is more interesting than a "perl" user (246) or someone who lists "programming" (958). I think I can implement this efficiently.
Anyone else have ideas for tweaks to the similarity metric?
Also, it'd be nifty to have an option to search for communities as well as users. Obviously I can write the code for that as well. :-)