Re: Problems with Bayesian filtering

Well, there are plenty of other meta-data signs that we do find are 
good spam-indicators though, so we only found the opposite as regards 
the color tag in particular, not meta-data in general.

C

On Feb 26, 2004, at 9:47 AM, <[email protected]> wrote:

> FWIW, in our testing, we found the *exact* opposite.  Matching on HTML
> colours is very error-prone.   I suspect Paul's high hit rate was a 
> result
> of him not receiving very little mail in HTML that was not spam.