I have got a private server running postfix, spam-assassing, dovecot, sieve... a common deployment on a linux server, but my SPAM filter doesn't work very well, because my client "set" is reduced and spam emails are like ten times bigger than my ham mails.
I could of course catch any spam collection freely available on the Internet to use as learning set for my spam-assassing filter, and my own gmail ham mail as ham set, but anyway, I don't think it will work reliably because I have to access to enough mails to have a well-training spam-assassing filter.
I think of three solutions:
Is there any well-trained spam-assasing filter available on the Internet? I could take one and pass to my spam-assassing filter, replacing the current one.
Take every mail I receive from my server, and forward them to a gmail account created ad-hoc with that purpose, and see to which folder arrives. If it arrives to a SPAM folder, I just simply send the original mail to the SPAM folder of my customer. If it doesn't, then to the
curdirectory. Creating a tiny
imapclient to do that shouldn't be too hard to code. All of it, provided that I don't forward too much emails to gmail, for example, discarding emails from compromised accounts and so on.
Same idea, but instead of simply do a double-forwarding (first to gmail, then to my customer), forward them separately (without waiting for the gmail results), but use the gmail results to train my own spam-assassing filter periodically (each day, for example). In that case, spam-assassing, theoretically, should mimic slowly the gmail filter.
The think is I don't dare to use gmail for that non-intended use, but I don't know exactly how to move on to improve my spam-filter otherwise.