I've been waiting for actual spam to bother me enough to make a naive bayes MRF for Pleroma. It is indeed a good solution