From: "Robert Brooks" <[EMAIL PROTECTED]>
> Jeremy Kister wrote:
> > On Monday, May 17, 2004 1:38 AM, I wrote:
> >
> >>while 20 words does seem accurate in detecting poison, there are too
many
> >>false positives in ham, such as
> >
> >
> > playing with the regex just a bit more produces a far more accurate
result
> > in both ham and spam:
> >
> > rawbody BAYES_POISON_01 /([a-z]{3,}\s+){20}/
>
> a lot of the bayes poison is either in a tiny font or in the background
colour,
> if we could exclude text on this basis then it would be a winner - SA3.0?
There are already sets of external rules that handle these issues
quite well, individually. It'd be duck soup to make meta rules for
combined strings of long words without short words or punctuation
with micro or invisible text.
{^_^}
