On Mon, Mar 31, 2008 at 9:30 AM, Kaare Rasmussen <kaare@jasonic.dk> wrote:
I've recently installed the Spamassassin filter for Kmail. It sure removes some spam, but not nearly enough. Typycally it will remove approximately 75 out of 175 spam messages, leaving 100 left for me to sift through.
I've let it learn what is spam for a week or so now, with absolutely no impact on the ratio. Obviously, I wonder if I register spam in vain, did I forget something, do I have to few messages (don't know how to see the numbers, but I guess I have 1-2.000 spam messages registered), or is it not that efficient to have identified spam as I thought?
Let it learn? Or teach it? The keys to SA is get razor working. All the network tests are useful, but if razor says its spam it is, period. Then make sure you understand how bayes databases have to be set up. Since SA does not block spam, merely tags it, you have to add kmail filters to move the probably spam to a folder that you inspect occasionally. Set SA to record tests that it hit upon, and not to mess with the subject. Then look at stuff it says is not spam (but which really is) and see which rules are fireing. If bayes_99 is not hitting then you have your bayes set up wrong, or its still not trained properly. You have to explicitly tell it which is spam and which is ham for a while. The easiest way to do this is to create a folder called MissedSpam and one called FalseSpam (or some such) and move/copy miss-categorized messages into those folders. Then use sa-learn to pass those mailboxes to the learning process to train bayes, and (if you register) report them to Razor. Without this, SA won't be anywhere near as effective as it can be. I get less than 1 spam per day using SA, and never the same spam twice. -- ----------JSA--------- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org