Redelivery of ham in Spam training
list at joreybump.com
Mon May 1 19:52:40 EDT 2006
John McMonagle wrote:
> Jorey Bump wrote:
>> John McMonagle wrote:
>>> Any simple way to find the local recipient the message was going to?
>> Deliver it to them first. Then they can volunteer a copy and you can
>> discard it after training. It's still a bad idea, but may be useful if
>> the participants are restricted and trustworthy.
> Mailbox is shared by a few trusted users.
> Hope were back on track.
In that case, it's still easier to let these users do your
classification. Instead of moving the trained messages, delete them and
purge. Since you are training your bayes database, it will help in the
The problem is that once the message is delivered, recipient information
is usually lost. The MTA keeps track of destination addresses while the
messages are in its queue, but an IMAP server only manages recipient
mailboxes. I use postfix, and put messages with marginal SpamAssassin
scores in the HOLD queue for inspection. I delete 99% of these, so I
usually don't bother to retrain on the false positives. I will retrain
on the junk that leaks through to my own account, though. Periodic
SpamAssassin upgrades are more helpful, and new rules bootstrap the
bayes database pretty well.
More information about the Info-cyrus