Personally, I lean toward thinking that sort of thing is too complex and unpredictable to determine algorithmically, but then again, dating sites have evidently done well in the last 10 years, and analog matchmaking sites did well before that, so maybe. My problem with solving the friendship/dating problem algorithmically is that it seems like a good answer to the wrong question. My best friends are the ones that I have a ton of shared experiences with (going to high school or college together, knowing each other for 10+ years, etc.), and the best "dates" I've been on have happened serendipitously. I don't know how you'd address those dynamics with something as necessarily superficial as a questionnaire. If anything, I could see a questionnaire being a good screening device for people you almost certainly would not click with, but I think it'd be almost impossible to predict the opposite. Just my two cents, though.