----- Original Message -----
From: "Jonathan Tai" <[EMAIL PROTECTED]>
On Mon, 2004-02-16 at 00:02, Asif Iqbal wrote:
>> Hi All
>>
>> I am using spamd/spamc with Qmail. My spamd/spamc are owned by qmailq
>> user. Now for emails that are miscaught as spam or failed to catch as
spam
>> needs to be learned by a user. Now I am not sure how I can pick those
>> slipped and false spams and pipe them through sa-learn so that only one
>> user learns all these
>>
>> I can create a user called spam and forward all the spams that are
>> missed and have a dot-qmail file with sa-learn. But it is not the same
>> user who owns spamd. Also I can create a user called ham and forward all
>> emails that are falsely labeled as spam and have a dot-qmail file
>> sa-learn to learn it as ham.
>>
> You shouldn't forward mail to be learned by bayes because bayes will
> pick up your FWD headers and tag those as spam.
>
> I would suggest setting up a cron job that executes as the user spamc is
> running as to search through a Spam folder, learn it as Spam, then
> search through Inbox/Sent/whatever and learn it as Ham.
I understand the downside part about forwarding spam back to myself at
a spam account well enough. Is there a procmailrc entry that can be used
to strip off the forwarding headers and then forward that to the sa-learn
tool as the "From:" addressee user? That might be a handy way to make
training easier from non-Linux email software like Outlook Express.
{^_^}
