Spamassassin at work is detecting spam nicely, no false positives so far although some false negatives are getting through. I don't have a big enough spam or ham corpus to get the bayesian spam filter running but when I do this could make a difference.
I have created three IMAP folders for spam:
| Possibly | This holds messages with a spam score > 5. Most spams come into this category. Messages are moved here by a procmail filter. |
| Definitely | This holds messages with a spam score > 12. Messages are moved here by a procmail filter. |
| Categorically | I use this for spam that isn't detected. When these arrive in the inbox I move them here manually. I also move messages from the 'Possibly' folder to here. |
I then have a cron job running each night to run sa-learn --spam on the 'categorically' folder and sa-learn --ham on my inbox. It runs on the 'cur' subfolder but not the 'new' subfolder so it should not pick up spam that arrives when I am not around to approve it.
The Definitely spams are not added to the bayesian filter. I think this is best, it avoids reinforcing prejudices which is not a good thing.
The spam I receive has already been through the companys surfcontrol spam filter. Draw your own conclusions.
Something I must do sometime is extract 250M of old emails from my outlook archive and put them on the IMAP server. I could do this by hooking Outlook up to the IMAP server and just moving the messages but I have a gut feeling that this will take hours and involve a lot of pain.
At home I must figure out a way to get my old email out of the clutches of Intellimail so that I can uninstall it.
Peter's Moblog
