Robert Nicholson <steffi2(_at_)DGS(_dot_)dgsys(_dot_)com> writes:
Just exactly what do you consider spam then?
Anything that looks like spam. The best way to "define" it is to use some kind
of "approximate function" (like neural nets). To stay within reasonable
procmail recipes, you typically use a scoring recipe which counts occurences of
typically-spam (groups of) words (like "make money fast") and then decides
whether it is most probably spam or not.
What one usually does is start with a simple recipe and adapt it over time.
Stefan