Problem is, if the questions can be automatically generated from publicly available information (the contents of a blog), so can the answer. It will work as long as the system is not commonly available, because spammers won't care - but that's true for any new anti-spam system.

Manually creating three options and selection one as the "right answer" won't work either, a spammer will just try three times, until he gets it right.

I don't like the affinity test - it may work for certain types of sites, but consider the case where you criticize a company on a blog, and a rep wants to reply and participate in the discussion. If he has to do a "nontrivial amount of research" to post, he's probably not going to, which would be a shame.