I agree partially with Charlene: the challenge of engineering social relevance algorithms is, in part, about ascribing meaning to available social metadata.
As we learned with Google and search optimization SEO over the years, the trick is to calculate relevance with measurements that are smart enough to a) avoid rich-get-richer entrenchment dynamics (who wants Kutcher and Bieber tweets dominating their social search results?), while at the same time b) detecting and parsing inherent manipulation ability. In the social sphere, this ability is not just rampant, it’s inherent, and at a scale that dwarfs even SEO. (Consider widely accepted schemes like #FF for artificially inflating one’s followers, and automated tools for retweeting.)
Social relevance will likely be a complex function of followers, tweets, retweets, mentions, velocity, and context. As spammers develop their craft, algorithmic relevance will evolve to consider things like profile inlinks and tweet link click through, as a measurement of genuine interest worth recommending to others.
Read more from Danny Sullivan on the social signals Google and Bing use.