On Thu, 25 Oct 2012 18:59:03 +0200 Simon Loewenthal top-posted: > RW <rwmailli...@googlemail.com> wrote: > > >On Thu, 25 Oct 2012 16:47:20 +0200 > >Simon Loewenthal wrote:
> >> * 2.0 DEAR_SOMETHING BODY: Contains 'Dear (something)' > >> > >> > >> Does anyone know the rational behind this, > > > >So it wont hit Dear Bob, but will hit Dear Sir etc. It seems > >reasonable, they're all forms of address that typically wouldn't be > >used if the recipient's name were known to the sender. > > Except for formal letters to administrative addresses. Such addresses tend to get a lot of spam, while the legitimate mail is unlikely to look particularly spammy, and likely to do well in Bayes. I suspect that this type of DEAR_SOMETHING FP has a relatively small effect on the classification FP rate at the default threshold. If you have a particular problem with it you might want to shift the bulk of the score to a meta rule that excludes problematic addresses.
