Kris: You mention "and Outlook is a headache to do properly but it *is* possible."
That got me thinking of how to do it. If I craft an email with no content but the FP or FN email drag-n-droped into the message body, I get a message/rfc822 Content-Type. If I send this to the spam or ham learning address, will sa-learn ignore my outer message and parse only the rfc822 part? The way I train FP's is just forward the whole thing to the mbox file where sa-learn will train - SpamAssassin has made an rfc822 attachment out of the email as part of it's reporting. For FN's, I've not found a good way, but your comment got me to thinking about the drag-n-drop. (Drag to the desktop, and include that file does the same thing. SaveAs does NOT!). I guess sa-learn has a -D switch, but would I be able to tell what it was doing at the level required here (which message bits are being learned)? I may have to try that unless someone (or Kris) already knows the answer and can save me the trouble. Dan -----Original Message----- From: Kris Deugau [mailto:[EMAIL PROTECTED] Sent: Tuesday, January 30, 2007 2:24 PM To: Oenus Tech Services Cc: users@spamassassin.apache.org Subject: Re: sa-learn on dedicated spamabuse email account Oenus Tech Services wrote: > Yes, but this would not work with our more than 1500 customers that have > only pop3 access and do not have access to any shared or private folder > in the servers. We needed to implement some way for these pop3-only > customers to report spam back to us, and for now we've only thought of > forwarding spam to an spamabuse account where some scripts could check > its inbox and do bayesian learning. However, does sa-learn take into > account that those emails are being forwarded That depends on what you mean. If you just click "Forward" in most mail clients, the result will be useless for feeding Bayes. (Mostly. The exceptions are rare.) However, if you spend the effort to teach people how to "Forward as attachment" (most mail programs support this properly; Eudora is the only one I know of where it's simply not possible - someone *please* prove me wrong? - and Outlook is a headache to do properly but it *is* possible.) Hand-check the submissions for a while though, because customers *will* misclick something (you do NOT want to train ham as spam) or submit something they really *don't* want, but which is still ham because it's a "Joke-A-Day"-type newsletter that many more people want - so you have to blacklist it for that user. Of course, if you've got the server resources for per-user Bayes - automate away, customers can't hurt each others' mail flow. >;) You might also want to keep submissions for a week or so in case $user calls in complaining their favourite "Please Spam Me And My Friends" newsletter isn't showing up in their inbox - you can dig it out and hand-train it correctly. -kgd
