Spam training and Junk folder behavior

[flyerguybham]

I have noticed that Zimbra's spam training and recognition is not improving over time as I expected it would. I've concluded that I don't really understand how Zimbra performs training as well as I should.

Three related questions:

1. I know that users can help Zimbra spam training by manually using the Junk / Not Junk button on the toolbar. However, does dragging a message into the Junk folder trigger similar learning behavior? This is important for us, as most of our users interact with Zimbra via something other than the web interface (mostly Thunderbird, sprinkling of other IMAP clients). It would be nice if they could drag mail to the Junk folder and have it learned.

2. Does marking a message as Junk using Thunderbird's Junk / Not Junk button (or similar buttons in other clients) do anything useful for Zimbra training?

3. This command is in zimbra's crontab: /opt/zimbra/bin/zmtrainsa
What is the behavior of this? From running it manually, I see it is looking only at the user zimbra's folder(s), but I can't tell which folder(s) or what it is doing. Is there any mechanism by which EVERY user's Junk folder is analyzed on a regular basis?

Ultimately, what I think we want is for users to be able to help Zimbra train by dragging messages to their Junk folder. In the meantime, I am manually running zmtrainsa on my own Junk folder once a week to help the process.

Any thoughts or advice here will be much appreciated. Thanks!

[dkarp]

Originally Posted by flyerguybham

I know that users can help Zimbra spam training by manually using the Junk / Not Junk button on the toolbar. However, does dragging a message into the Junk folder trigger similar learning behavior?

In the Zimbra Web interface, you need to press the "Junk" button to train the spam filters. Dragging a message to the "Junk" folder does not do the same thing.

There's already an enhancement request filed in bugzilla to make IMAP COPY to Junk train the spam filters. Please vote for it if you'd like to see it implemented!

Originally Posted by flyerguybham

Does marking a message as Junk using Thunderbird's Junk / Not Junk button (or similar buttons in other clients) do anything useful for Zimbra training?

No. Although I believe that it trains Thunderbird's own spam filter, for what that's worth.

[wdimmit]

In the interim, you can have your users forward the messages to the spam address you defined at setup. Zimbra should pick them up and train from them there.

[LMStone]

Originally Posted by wdimmit

In the interim, you can have your users forward the messages to the spam address you defined at setup. Zimbra should pick them up and train from them there.

Not sure you want to that!!!

Forwarding an email to a spam training account causes the forwarder's email address to be trained by the spam filter.

What you want is to do a REDIRECT, which is available in MUAs like KMail.

HTH,
Mark

[dlbewley]

It seems undesirable to divulge the auto spam and ham training addresses at all, let alone trusting users to use them properly.

The ability to train on a user spam folder exists

./zmtrainsa <server> <user> <pass> <spam|ham> [folder]

That is essentially what I do on my existing non-zimbra server with a daily cronjob, but passwords are a non-issue when using Maildir.

Zmtrainsa calls zmspamextract which calls `zmjava com.zimbra.cs.util.SpamExtract`. By default no username is given and it seems to login as zimbra and extract email from presumably the "spam" and "ham" accounts (depending on a -s or -n argument). Correct?

It would be really nice to run zmtrainsa on a user without knowing their password. Can this be done?

[wdimmit]

Apparently I was mistaken / misguided in my first response - so I'll make an attempt at vindication here ;-).

Building on dlbewley's suggestion, you could also use imapsync to move or copy mail from the users spam directories into the spam mailbox directly. This doesn't solve the password problem entirely, but it does give you a few more password and auth options, which I don't believe you could (easily) have with zmtrainsa directly.

On a related note, I remember seeing a bug lately regarding admin access to user accounts over IMAP. I don't recall if the bug/rfe was for preventing or allowing that, but I'll post back if I can find it in Bugzilla.

--Will

*Update* I found the Bug / Thread I was thinking of http://www.zimbra.com/forums/showthread.php?p=23931

[phoenix]

Originally Posted by wdimmit

Building on dlbewley's suggestion, you could also use imapsync to move or copy mail from the users spam directories into the spam mailbox directly.

Why do you think you need to copy it to the spam/ham directories? Did anyone think to check the wiki?

[wdimmit]

Originally Posted by phoenix

Why do you think you need to copy it to the spam/ham directories? Did anyone think to check the wiki?

I think there's some confusion here. The discussion has been training SA from an IMAP user's "Spam" folder - without resorting to sending the user to the WebUI.

[phoenix]

Well, that just shows what an idiot you can look when you read half a thread. I started at dlbewley's post where he was talking about zmtrainsa and thought you were talking about the same thing.

There isn't a way round using zmtrainsa without passwords, I think you'd be better off voting for the bug dkarp mentioned above but that's your choice. It is rather risky, as has already been mentioned, trying to get a user to send junk/not junk to specific accounts.

[dlbewley]

Please vote for bug 14659 if you think it would be useful to train spamassassin with users arbitrarily named spam folders without supplying their password.

Think IMAP-only users, folders named other than Junk, and freshly migrated systems.