View Single Post
Old 2 Nov 2018, 11:27 PM   #13
jhollington
Essential Contributor
 
Join Date: Apr 2008
Posts: 371
Quote:
Originally Posted by xyzzy View Post
I understand marking spam as not spam (or not spam as spam) is the way of training my spam filters before deleting the stuff. But I believe the determination of what is or is not spam is determined at delivery time, not when I delete stuff (i.e., nothing affects the spam filter training at deletion time). When I mark stuff as spam (or not spam) I am only trying to correct (re-train) the spam filters to avoid their mistake on future deliveries.
Actually, while your'e correct that messages are identified as spam at the point of delivery, it is the point of deletion at which spam messages are trained as spam (in terms of the Bayes database). This is to prevent false positives that would come from simply training based on the spam folder (although you can set your standard junk mail folder to train all messages that are in it as spam, this isn't the default setting, and in fact FastMail specifically recommends that you don't do this.

Basically, if messages that land in your "Junk" folder get trained as spam, then the Bayes database is going to include every message that lands in this folder, whether it's spam or not. While you can later mark these as "not spam" that doesn't actually undo the process of teaching FastMail that the original message was spam, but rather simply adds it to the "ham" database — so now you have two conflicting entries in the Bayes database, one that says a message is spam, and one that says it's not spam.

So unless you actually set your Junk Mail folder to be a source for learning spam, the Bayes database only gets updated when a message is specifically marked as spam or when it's deleted from this folder —*either manually or as part of an auto-purge rule.

Also keep in mind that this only applies to marking messages as spam or deleting them from the Junk folder via the FastMail web interface or FastMail mobile app. Messages deleted or marked as spam from an IMAP client don't contribute to the Bayes database — although if your IMAP client moves messages that you mark as spam into the standard Junk Mail folder, and you hae an automatic purge on that folder, then they'll eventually get learned as spam once they get automatically deleted by FastMail.

The other alternative, if you use an IMAP client primarily, is to create a second Junk Mail folder that your IMAP client uses to send spam to. Mark this second folder to be the one from which spam is learned. FastMail will put all of the spam that it detects into the default junk folder, while anything you flag as junk in your IMAP app will be moved to its own junk folder, which FastMail will scan regularly to learn more about what is spam.

See https://www.fastmail.com/help/receive/stopspam.html for more details on how this all fits together.
jhollington is offline   Reply With Quote