[Mailman-Developers] New SpamAssassin handler on sf.

[Simone Piunno]

On Monday 17 November 2003 23:52, Richard Barrett wrote:

> Might I ask why enthusiasts for integration SpamAssassin with Mailman
> do not care about delivery of spam to other mail aliases in their
> domain. And if they do so care, why do they not concentrate on stopping
> spam reaching all of their mail aliases.
> I guess I cannot understand why you are messing around integrating
> SpammAssassin with Mailman when you should be integrating it with your
> MTA so that the spam never gets anywhere near Mailman.
> Or did I miss something important along the way.

Yes, you miss that a bayesian filter works better when the non-spam traffic is 
focused on a given argument.  If you train on many lists talking on very 
different topics, maybe even in different languages, a bayesian classifier is 
less effective.  For this reason it's much better to have a separate spam 
database for each list.   Not to mention how cool can be the Mailman UI as a 
management interface for your filter (e.g. by training on messages held for 
moderation).
This applies to all bayesian filter, not only to SpamAssassin.

-- 
Adde parvum parvo magnus acervus erit -- Ovidio