On Thu, Mar 25, 2004 at 12:25:38PM -0500, Mark Merchant wrote:
> > After setting this up, I
> > instructed the users to forward any spam messages they receive to the
> > address [EMAIL PROTECTED] and any incorrectly flagged spam
> > messages or sets of non-spam messages to [EMAIL PROTECTED] .
> 
> < without looking at said script>
> 
> so what happens when some joker puts mail from the company president
> in the spam folder? or like my users, can't remember what the
> difference between spam & ham is? so they put the porn stuff in the
> ham mail box? ( could be intentional i suppose ;)

Well, if you do per user Bayes databases, they are only shooting themselves in 
the foot, and not anyone else. 

I know this has issues all its own, but I'm looking into keeping each users
tasks to a minimum (or completely optional), while protecting the integrity
of the filters for everyone.

I just tell users to create a "spam" and "ham" mailbox (either, or both
works, depending on their whims).  I run a process to reap these nightly
and feed them to each users Bayes db.  I also keep a per user corpus from
these for the last 60 days on a per user basis, so re-building can happen
quickly if the Bayes db becomes corrupted.

I turn of auto learn, and feed the Bayes manually.

> 
> we could never figure out a way to do this that wasn't extremely
> vulnerable to human error.

If they make mistakes, I find Bayes is pretty tolerant, and still works
well.  Big mistakes can be fixed by manual re-learning if needed, or 
re-initializing from an edited corpus.

> 
> on the positive side, automagically making whitelists out of user
> address books works quite well...

White lists don't stop joe-jobbed email containing viruses...

-chuck

Reply via email to