Submit spam and non spam to the filter

[rex.withers]

I would like my users to submit spam and non spam to the filter by sending to spam@domain.com and nonspam@domain.com can I get them to attach multiple mails and just send them that way or do I have to get them to forward the mails individually?

[Jodrik]

I've tried something similair to this only with automated RBL via mail forwarding to a single adres like spam@domain.com

What I found out is that AMS does not yet support you extracting anything from inbound mails and doing something with it. I always take the time to mail support about these things since Code Crafters has implemented almost everything I ever had the idea to suggest to them.

From the reply on my latest mail:

I should note that we also have considered how content filtering could be designed to obtain specific custom strings from the mail body and headers. This means that in a future update there is a good chance we may introduce something to do just this.

I'm not sure this will help in any way on the Bayesian filter, my novel way of dealing with the Bayesian filter is to let people use IMAP4 and created two permanent serverside included folders SPAM & NON-SPAM incase something is read wrong. Every 24 hours my server automatically scans the SPAM folder for content and I just set up a simple mail adres to allow users to mail me when Bayesian filterd something out they deffinatly don't want filterd out.

But for a more usuable filter I highly recommend using a simple DNS server to create your own RBL. It's fairly easy to script the means to add a DNS record via a simple webapplet using something like PHP.

EDIT: Note that it is advisory to inform your users that when using IMAP4 over the normal port that their username and password is send as plaintext over the internet to your server, if you need to provide a secure envoirment you should deffinatly considder using the SSL version of IMAP4.

[Code Crafters]

There are 2 ways to train Bayesian filtering based on the protocols you use:

1. IMAP4 and WebMail: As Jodrik said, you can get users to archive their mail into SPAM and non-SPAM folders. You can then set up Bayesian filtering to train from these folders using the Auto-Learn From Users training mode only for best accuracy. If a mail is incorrectly trianed from you can simply put it in the other folder and it will be retrained effectively undoing the incorrect training done previously.

2. POP3: Users that access mail via POP3 are limited to only having an Inbox and obviously can't sort mails into folders. Instead set up 2 users for Bayesian training (e.g. spam@yourdomain.com, non-spam@yourdomain.com). Get POP3 users to forward mails appropriately to these users. Then set Bayesian filtering to train from these users for SPAM and non-SPAM respectively. You can also use content filtering to remove or edit the from header line for these Bayesian training users since you don't want Bayesian training to pay any attention to the From address which will now be the local user who forwarded it. You can either forward mails one at a time or multiple combined as long as the mails are still stored as plain text attachments and not base-64 encoded or similar as the Bayesian needs to read the tokens and won't unencode attachments before doing this.

When using IMAP4 or POP3 you should ideally use either implicit SSL (whole conversation encrypted), explicit SSL (conversation encrypted before login) or SASL secure login mechinisms (e.g. CRAM-MD5) to avoid your password being sent in plain text. You should also access WebMail using SSL if possible.

[rex.withers]

Thanks for that super comprehensive reply if didn’t really understand the final bit (more study required) but you answered my question well.

Thanks again