spam levels [was: Including html-tag contents ...]
Peter Bishop
pgb at adelard.com
Wed May 14 09:49:00 CEST 2003
On 13 May 2003 at 22:07, Tony L. Svanstrom wrote:
> I just want to tell bogofilter "this e-mail that I want you to classify has
> a
> greater/smaller than average chance of being spam; please consider it when
> you're classifying it". Resulting in the ham/spam-scale to be nudged
> towards
> either the ham- or spam-side.
>
There is some respectable maths for this using Bayes theorem.
i.e. instead of being neutral you have a "prior belief" about the
probability of different outcomes, and this is modified by the evidence to
yields a "posterior probability" of the outcomes.
Not quite sure how to map this into bogofilter though
- maybe preload bogofilter with the token distribution for that
particular user (for all prior messages), then allow the new message tokens
to destroy or reinforce that distribution
--
Peter Bishop
pgb at adelard.com
pgb at csr.city.ac.uk
More information about the Bogofilter
mailing list