On Fri, 2005-03-18 at 15:13 +0000, David Woodhouse wrote: > On Fri, 2005-03-18 at 10:56 -0400, Patrick O'Callaghan wrote: > > Meaning you have to send the junk message back to the server, but to a > > special "learn spam" address. Ditto for false positives. You don't > > gain anything special from Sieve filters in this case; it's still a > > hack. > > Either to a special address, or just dump it into a special IMAP folder > which is fed to sa-learn and purged automatically.
That's our hack of choice for now. > You could also use > IMAP flags for Junk, and an IMAP server which automatically invokes sa- > learn when the flag is either set or cleared by the user. Would need mods to Cyrus, but could be done I guess. > What happens to messages which the user decides are spam? Presumably one > of: > > - Delivered to a spam folder which the user may never (or rarely) visit > (which is vaguely equivalent to:) > - Blackholed silently with no feedback to the sender > - A bounce is generated to the apparent sender, thus potentially > spamming an innocent bystander with abusive bounces. > > Note that if the first and second options above _aren't_ at least > vaguely equivalent, there's little point in the exercise anyway, because > it means the 'per-user' tuning of spam filtering hasn't allowed your > user to have any form of confidence in the accuracy anyway, and they're > still checking their spam folder regularly and in detail. I check my SPAM folder every few days. I can spot 99.9% of genuine spam just by the From: and Subject: so I have the preview pane off. Plus I haven't had a false positive in over six months because my Bayesian engine is well-trained by now, so it doesn't take me more than 30 seconds to mark and delete everything. "So why doesn't the system just dump all the spam without you needing to check it?" I hear you cry. Answer: politics. There's a huge difference between "the system deleted my valuable false positive" and "I deleted my valuable false positive". > I'd quite like to be able to do per-user filtering, but I'd need to set > things up so I can do that at SMTP time, because I'm really not willing > to do any of the three options above, and can't see any alternative. > > > I guess it depends on your environment. I can't tell our community > > (we're a university) "OK, I'm going to decide what incoming mail is > > spam and you can all take it or leave it". > > Well, you _can_ do that, and some universities do. You just can't be > _too_ strict about what you call spam, that's all. In fact I have just a > few tens of users, and those are mostly using my machines _because_ of > the spam filtering that's done there. Believe me, there's a difference between a few tens of users who use your machines because they want to, and a few thousand users on a central institutional mail hub on which they pretty much *have* to have an account (though it needn't be their only one). poc _______________________________________________ evolution maillist - [email protected] http://lists.ximian.com/mailman/listinfo/evolution
