Thunderbird junk mail filter problems

Question

After a hiatus of several years where I was using Apple Mail I've gone back to using Thunderbird for my mail. I'm generally pleased with Thunderbird. I like that it works on all OSes (I use several). I like that it can handle multiple accounts (I have several). I like that I find it intuitive. I want to keep using it.

However, I'm having some serious trouble getting the junk mail filters to actually work. This is on Ubuntu with Thunderbird 2.0.0.22, and MacOS 10.4 with Thunderbird 2.0.0.16. There's more than one account. There's also more than one Thunderbird instance but they are sharing Junk folders over IMAP (only way I know to train both of them).

Everything I've ever read suggests that after training on 3000-4000 messages most spam should be caught (barring waves of new varieties which come along periodically). That's been my experience in the past. At first, mostly untrained, I was getting between 20 and 200 messages a day. This didn't slow down much after a couple of days. I have a fairly large body of existing spam, some 20000 messages, and I added it all at once. That improved the true positive rate considerably, though I was still getting a handful of spam messages a day. Frustratingly many of these seemed to be the same message, and Thunderbird seems to have a great deal of trouble recognizing some of these. (I looked at a few to see if they are doing the normal tricks that give filters a hard time: images instead of text and/or random paragraphs of "known good text". In one case that was true but in others the messages appear to be short and mostly empty. No images or embedded text.) And now, in the past week or so, the rate has shot back up again to a dozen or more an hour. It's as if the filtering has just stopped.

Basic procedure and obvious gotchas, which I've already done:

• Enabling junk mail filtering in Thunderbird requires two separate settings:
  • In the Options (aka Preferences) select Privacy | Junk and enabled 'When I mark messages as junk`. You then choose to move messages to a Junk folder or delete them. I always do the former to prevent losing mail.
  • Under Account Settings for the account select Junk Settings enable Enable adaptive junk mail controls for this account.

I know these work because junk mail is being filtered. Just poorly.

Things I've tried:

• I've tried looking at the file training.dat. I've peered inside it with things like od and strings and it is mostly embedded strings. But the format is basically opaque. It does grow, but noticeably it does not grow every time I mark a message as junk.
• I've turned on Junk Filter Logging (Preferences | Privacy | Enable junk filter logging). This does nothing. There is no log. I seem to remember trying this years ago and it didn't do anything then either.
• I've considered tossing training.dat and starting over, but (a) it is changing (b) I don't really want to start over and (c) there's no way to mix training files, so you really can't go back to an old file without losing whatever you've gained.

So, questions:

• How can I tell if training is actually happening?
• Should I expect training.dat to change every time I mark something as junk?
• Why don't the logs appear?
• Do I have to mark each message individually, or can I mark a whole bunch of messages at once?
• Are there any tools for finding out when the filter runs and what it is doing when it does?
• Are there any tools for decoding the training file?
• Overall, why does it seem to have stopped working, and what can I do about it?

There's lots of links to Thunderbird junk mail filtering on Google, but they almost all boil down to basic tutorials. I am looking for more than just basic instructions: I want to know how to debug or diagnose how the filter is working -- or not working.

Update: I wasn't clear about this originally but I have successfully used Thunderbird in the past, for a period of many years. I stopped for a few years because I was using Apple Mail. So it's not a problem of not knowing the basics or not getting any filtering. Filtering is happening just sporadically and poorly.

Answer 1

You may be interested in the JunQuilla extension for Thunderbird. It requires Thunderbird 3.0 or later.

It has the ability to show the bayesian junk rating of your messages as a percentage (among other things).

Info here: http://mesquilla.com/extensions/junquilla/

Answer 2

I assume you've seen this article: Junk Mail Controls. There's also a link to this somewhat old article: Junk messages not moved. This forum thread may prove helpful to you as well: Enable junk filter logging.

Good luck!

Answer 3

See http://kb.mozillazine.org/Junk_Mail_Controls#Tweaking_the_Junk_Mail_Controls

In Thunderbird:
You can change the preference using Tools -> Options -> Advanced -> General -> Config Editor. Enter "junk" in the Filter field to show only the preferences that contain "junk" in their name, and then double-click on mail.adaptivefilters.junk_threshold, enter a value lower than the default 90 in the edit field and press the OK button. Many users report good results with values of 30 or lower.

Answer 4

I don't use the junk filter at all. Instead I use MailWasher to intercept the mail before it gets to my mailbox. I don't need to train it, since I simply set up a filter that defines everything as junk, and I add as friends only legitimate addresses.
It's not free, but the trial version used to never expire. I wouldn't know about the current version, since I paid for MailWasherPro (and it was worth it).

Answer 5

If you have webmail that is forwarding to your account, at the very least you could set up filters there to take some of the spam out of your inbox.

Answer 6

Which plugin are you using, spamassassin or bogo filter? Spamassassin should work pretty robust if you have that much spam training!

Answer 7

I would use spam assassin plugin for Tbird, or one of the other spam detectors, and turn off the automatic learning. you can also block it at the email server level (by that i mean by using the gmail,etc web interface to mark it as spam)