Yes, I would like to do that--combine it with the rating prediction, so e.g. if two newsletters both have a predicted rating of 4 stars, prefer the one with content that's most dissimilar to the users' positively rated newsletters so far.