[sylpheed:30612] Re: some tutorial about using the spam filter ?

[Stefaan A Eeckels]

On Fri, 22 Dec 2006 10:26:48 -0600
Gene Goldenfeld <genegold at highstream.net> wrote:

>  One thing I've never
> understood is if one has to look every time anyway, why not use a
> front end app like Mailwasher (free or Pro)?

No filter is 100% reliable. If you decide to use Mailwasher you'll also
have false negatives and false positives. The advantage of bogofilter
(which I use centrally at my incoming MTA) is that it is trained with
the SPAM, and more importantly the HAM you get. It's also a
disadvantage because you need to train it, and as long as it isn't
well trained, its ability to distinguish between HAM and SPAM can be
less than a non-bayesian filter.

I have found bogofilter to give superior results when you train it with
several hundreds (the more the merrier, really) of SPAM and HAM
messages, and if you configure it so that every message you receive is
integrated in the database.

This is how I've configured bogofilter for my network, and none of my
users gets more than a couple of false negatives per month (mostly new
scams, or scams translated in another language). The false positives we
get are always solicited commercial messages, which are often
content-wise pretty close to SPAM anyway. The solution is then simply
to place a filter rule recognising that specific originator before the
filter rule that triggers on the "X-Bogosity: " header inserted by
bogofilter - call it white-listing if you want.

Take care,

-- 
Stefaan
-- 
As complexity rises, precise statements lose meaning,
and meaningful statements lose precision. -- Lotfi Zadeh