Idea: Sorting, not Filtering
Bogofilter splits email into spam and ham. The mechanisms for doing this are to derive a spam score and a ham score, and bring them together to come to a decision. Regularly, there are strong indications in both directions, and Bogofilter cannot make a final decision.
What we are hoping to do is take a stream of messages and sort them into categories that match a user's set of aliases or, perhaps more practically, a set of topics that have been assigned to aliases. (I'll use the term topic so we can decide later on if they are the same as an alias.) When users move a message from one topic to another, they are correcting Bogofilter and asking it to sort in another way.
Bogofilter uses a probability-of-spam for each word, and reverses it to estimate a probability-of ham. The probabilities are multiplied and as Gary Robinson suggested evened out by taking a root to the same degree as the number of multiplicands. Generalising on that, every topic could have its own probabilities-per-word, and a match with each of the topics could be derived. The top hit would then lead the sorting decision.