SpamAssassin 2.50 news

[Matt Kettler]

Agreed, if you manually do the learning process, it should be fine, even
with multiple users, as long as the spam/nonspam it is fed is reasonably
representative.

However, auto-learning will be subject to the same ALL_SPAM_TO type
problems that the AWL is subject to.

Really I don't think that autolearning is all that good an idea anyway,
since it causes false positives and/or false negatives to be
self-perpetuating.

Probably the best solution is to have some kind of special email address
that users can send mail to manually to cause SA to learn from it. (ie:
learn_nonspam at mydomain.com, learn_spam at mydomain.com) and use a procmail
script to crank emails to these fake accounts to the learning commands of
SA. You'll probably also want to add some sender and/or recieved path
checks to that procmail script to avoid abuse from the outside by spammers.

At 10:59 PM 1/29/2003 -0500, Rose, Bobby wrote:
>Autolearning isn't in there yet.  Even so it'll be a configurable
>option.
>
>-----Original Message-----
>From: Steve Evans [mailto:sevans at FOUNDATION.SDSU.EDU]
>Sent: Wednesday, January 29, 2003 10:31 PM
>To: MAILSCANNER at JISCMAIL.AC.UK
>Subject: Re: SpamAssassin 2.50 news
>
>
>Have you looked into how the bayes filtering is going to work with
>MailScanner?  My understanding is that bayes is designed to learn off a
>single user, and may not have desirable affects when the learning
>process uses your entire mail flow.
>
>Steve Evans
>SDSU Foundation
>(619) 594-0653
>
>
>-----Original Message-----
>From: Julian Field [mailto:mailscanner at ECS.SOTON.AC.UK]
>Sent: Wednesday, January 29, 2003 1:45 PM
>To: MAILSCANNER at JISCMAIL.AC.UK
>Subject: SpamAssassin 2.50 news
>
>
>For all the SpamAssassin users, here is a reply I just got on the SAtalk
>mailing list.
>
> >On Wed, Jan 29, 2003 at 09:14:35PM +0000, Julian Field wrote:
> > > I know this is probably a dangerous question, but is there yet any
> > > idea of a very rough ETA for the next version of this wonderful
> > > package?
> >
> >Very rough: by end of Q1.  Less, but still, rough: by end of Februrary.
> >
> > > Just want to know whether I should start trying out CVS versions or
> > > just wait a couple of weeks for the real release.
> >
> >The main code is pretty much done, but the stuff that's in progress are
>
> >the scores.
>
>So we've got a good 6 weeks (or thereabouts) to go before the new
>version of SA will be available and settled.
>--
>Julian Field
>www.MailScanner.info
>MailScanner thanks transtec Computers for their support