The problem I see with most attempts at using the social graph for problems like this is that most social graphs have no notion of expertise. Tom may be a good friend of mine who shares my interest in cooking and scuba diving, but he works in a Windows shop coding in Java for enterprise apps while I am a Mac user who does large-scale consumer service development using Python and Erlang. He also likes the Patriots, which calls into question any opinion he may have about sports :) Capturing the different facets of "relevancy" is hard, and it does not sound like this comes close to being able to distinguish signal from noise when coming from a source that otherwise might appear to be trustworthy.