View Single Post
Old 12 Sep 2017, 10:59 PM   #10
jhollington
Essential Contributor
 
Join Date: Apr 2008
Posts: 371
Quote:
Originally Posted by n5bb View Post
Yeah ... I was confused about this until I re-read the latest help. I now understand the trap Fastmail is worried we might get into if we mark the Spam folder as learn-as-spam.
This is kind of an interesting point, and it took a moment or two to wrap my head around where they were going with this, but I can see how it makes sense.

However, I've had my Spam folder set as learning spam for years, and haven't really had any problems with it. Part of the reason for me is that I primarily use Apple Mail so it's just an easier way to deal with getting stuff taught as spam — I very rarely use the web interface. I also strongly suspect that FastMail won't recognize deletion of mails from the spam folder via IMAP —*probably also not surprising since most IMAP clients will do a move-to-trash rather than an expunge. I realize I could point Apple Mail to use any arbitrary folder as its own "Junk Mail" folder, but again, I haven't really had a problem and I don't relish the idea of having two spam folders

In my case, however, I've also set several other folders to learn "not-spam," including my Inbox and Archive folders. So in the event that something does get into a false positive loop, I could very well be misunderstanding how the Bayes training works, but I've always assumed the process of moving it back to the Inbox or Archive folders would effectively reverse the spam training for that message the next time those folders are scanned.

Further, since I check my Inbox every single day and the scanning only occurs on a cycle, there doesn't seem to be a high risk of false negatives (e.g. learning spam messages as legit) from having that set as a not-spam training folder — chances are anything that lands in there that's spam will be dealt with before the next FastMail scan of that folder.

Now, all of that having been said, I should note that I've been at this for a very long time with FastMail, so my Bayes database numbers for both spam and non-spam are into the tens of thousands. It might be a very different scenario for somebody who is just getting started.
jhollington is offline   Reply With Quote