Open Discussion → Nion-email uses of multi-target Bayesian classifier
It is motivational to think of non-email uses of a multi-target classifier, as POPFile appears to be.
E.g. automatically creating tags for files, as in MacOS-X Finder, or Gmail labels. That lst is email, but POPFile doesn't handle multi-labels well.
POPFile currently puts a message in one and only one bucket. When I have written my own multi-target classifiers, they have either been in a fixed order - whichever matches first wins - or, more likely, independent. If choosing one and only one, the highest confidence wins; if choosing multiple labels or tags, all above a certain threshold win. Some tags might be specified as mutually exclusive.