Nick Bright wrote:
The problem I'm trying to solve is "how to implement a training system
on my server".
I'd suggest a manual review step before feeding the messages to Bayes.
You **WILL** get users reporting all kinds of "unwanted today because
Reasons" but otherwise legitimate email as spam.
I suppose i could de-encapsulate an attachment with a script, before
feeding it to sa-learn?
I posted the key fragment of code I wrote for our spam report handling
address here, in December 2013:
http://mail-archives.apache.org/mod_mbox/spamassassin-users/201312.mbox/ajax/%3C52C2E950.8070804%40vianet.ca%3E
"Inline" forwards can still be of some value, but mainly in terms of
whatever further hand-modification of SA rules you do based on the
visible message body text. Pretty much anything else has been stripped
or munged.
-kgd
