[glouis at dynamicro.on.ca: Re: Dealing with wordlist mails]

[Tom Anderson]

On Wed, 2004-01-28 at 15:17, Greg Louis wrote:
> > In fact, I don't think that the default setting in the procmailrc 
> > example should have -u.
> > 
> > I think the recommendation should be 
> > - initial bulk training with a corpus
> > - followed by **manual** "train on error"
> 
> Apologies for the me-too posting, but I can't resist agreeing publicly
> and wholeheartedly with this suggestion.

I agree that this is certainly a valid way to do it, however for
end-users, it may be too much work.  Not many people have the time or
resources to collect a corpus of mail for training, and fewer have shell
access to their mail server.

I've rolled out bogofilter to several end-users on my server who do not
have shell access, and I've found that train-on-error works rather
well.  In fact, it's a really good way to "ease" into the training
mindset.  Basically, the users start out receiving everything in their
inbox as usual.  When they decide they get tired of receiving some of
the spams, they forward them to their bfproxy address for correction. 
Now they start to get some messages in their filtered box (for false
positive verification), their unsure box, and their inbox.  As they send
more spams for correction, they see more and more of their incoming mail
get filtered.  With a "learning" filter like bogofilter, this is an
intuitive result.  Teach your filter about certain spams, and emails
like them get filtered next time.  The all-at-once training method can
be too abrupt of a change for end-users.

Therefore, to the non-developers out there, I recommend no initial
training, and use train-on-error either manually for all spams, or
automatically (-u) for the "certain" onces and manually for the unsures
and mistakes.  Bfproxy is a good way for those without shell access to
do corrections, and will soon work for unsures (-s, -n) as well.

Tom

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://www.bogofilter.org/pipermail/bogofilter/attachments/20040129/5dd9d16c/attachment.sig>