Ability to "Learn" what is spam or not

[switchisp]

Like SquirrelMail that has a module for this; Zimbra could have the ability to "learn" what is spam or not after a while, when a user clicks on a "spam" or "junk " link.

AC

[KevinH]

We use spamassassin which has this ability. Currently we don't allow per-user training but it's on the roadmap.

[rmvg]

u do have a button called junk i guess all it does for now is move stuff to the junk folder? doesnt enter it into the bayes database?

[KevinH]

Correct. Just moves the message.

[anand]

there are two magic LDAP config attributes zimbraSpamIsSpamAccount and zimbraSpamIsNotSpamAccount - if set then then these addresses are emailed a copy of the message. Soon (not in the release that's coming up, but hopefully in the one after that), we will have that wired up to feed SA's bayes, and empty the accounts etc.

Any of you non-Zimbra (yet) folks using SA and have Bayes turned on in a multi-mailbox setup?

[rilke1999]

Just gave Zimbra a first look on an virtual FC 4 in VMWARE ESX and it works great.
I am currently running an Squirrelmail Setup for 150 Users ( http://koelnerabend.de ) and I allow user based training of the bayes filter. Due to a lot of stupid users, I had to switch from overall training to a user based setup. With a good training, the user will get up to 97% accuracy.
I seriously consider to change to Zimbra, but I will do some load tests first.

thanks for this great peace of software

[jstoltz]

Does the current release learn spam from the zimbraSpamIsSpamAccount automatically or should I run zmtrainsa against it periodically?

If I should use zmtrainsa how can I change or find the password for the zimbraSpamIsSpamAccount?

Thanks

[marcmac]

Currently, the "learn" feature isn't set up. To run trainsa, just reset the password on the account (it's set to a random string on install).

[jstoltz]

Thanks I have it working now. I didn't realize zmtrainsa was using fetchmail. I disabled pop3 and imap access to those accounts and was getting authentication failures.