Are any relationships or relationship-metrics bidirectional? (is friendship itself?)

I’d argue “yes”, but the arrows can be weighted differently (and usually are, I expect).

num_followers_in_foaf_network / (1 + log(num_followers_globally)

This, intuitively, seems better — but I’d love to see a histogram of the Twitter population’s number of followers/follows. I wonder how normal the distribution is. I would expect “not at all”: a bunch of peaks, out-of-scale at the very bottom, and I would not be surprised to see a dearth of users with number of followers between, say, 50k and 500k.

I should probably also eliminate the “celebrities” of the network

Maybe, but that’d be pretty hard. And also misleading for the celebs who use their big account as their personal account. It’s maybe along the lines of “I hope I don’t become famous. It’d be a pain to change all my avatars on sites that don’t allow one to use pictures of famous people.” It seems by this consideration that celebs couldn’t have real friends — or, at the very least, information about friendship would be discarded by your model. It’s a small population, but “celebrities are people too”. ;-)

Add on a minimum reply threshold (say, 3), and there’s a good proxy for friendship.

Ah, I see you are unacquainted with the concept of “flame war”. Welcome to the Internet, it’s a fine place. ;-)