I think this could be achieved by indexing only tags on blog posts. Aside from having to groom the tag corpus (maintaining sets of synonyms would probably be a big part), this would seem to be a fairly pain-free way of deriving some of these associations. It could be extended into "hmm, what other things are the people I share this with also interested in," which happens to be just about where I think the entire field of recommendation engine research should have ended. But I digress.

Yes, you would have to use tags to participate if you weren't already, but that just makes it opt-in.