On 10 Nov 2003, at 07:33, Terry Milnes wrote:
The typical user is capable of making toast in his electric toaster, but when it comes to the overwhelming complexities involved in operating a computer he is totally lost. He will become extremely agitated when he looses the *REALLY IMPORTANT* email that was tagged as spam, put into his spam mailbox and which he subsequently deleted because he didn't pay close enough attention to what *HE* was doing, that instantly becomes our fault!
Simple, I give users a choice, not SA or SA. If they choose to NOT use SA, I ignore complaints about spam. If they chose to use SA, they are told to check before they delete, if they don't I remind them I told them to check before they delete and suggest they get the mail resent. If they continue to have problems we give them the option of not using SpamAssassin.
These users have the same choice.....
Blaming SA for user's deleting suspicious mail is a bit like blaming the hammer for you hitting your thumb.
So, what's your point, you know this I know this but the typical computer user these days blames it all on his computer, its never the user himself that made a poor decision, or did something he shouldn't. And I bet when he uses the hammer and clobbers his thumb he really does balme it on the hammer......<g>
When someone posts to this list asking how he can improve the hit ratio for his customers/users, cites examples or ideas that may improve the success ratio for his situation, (multiple user, many morons) the last thing he wants to hear about is how good spamassassin is without any of his kind of modification and that if he use bayes or spends a little time tweaking he can see results like these.... He is probably already aware of that
"Training Bayes" is not something that users need to do. It simply happens as the Bayes filters learn their email.
This isn't what I have seen, unless there's a corpus of 200 + spam, ham bayes does not work. (Lets not discuss this here, the next response has this covered)
If you don't want false positives set your default threshold at 9.0 for users new to SA and lower it to 5.0 after a few months, once bayes has a chance to learn. My server _discards_ mail scoring 9+ unless a user specifically asks us not to. mail that scores 5.0-8.99 is marked as possible spam (Spam? 5.60), and users are told to sort their Spam folders (if they have them) by subject, so the lowest scoring potential spams are listed first. Makes it easy for them to check.
And that is exactly what we have done, as my mail explained, I was responding to the posters who questioned why I claimed the success ratio was only at 50 -60 %, its because the threshold is set at 8.
And I haven't seen a single "important" mail get tagged as Spam since the SA 2.3 days. Usually the false positives are mails that are indistinguishable spam, just spam that user happens to want (like my ads from ezydvd).
Most email the users get isn't what I would consider important either, its the users perception of what is or isn't important that has to be considered, if he says its important, its important.
tm.
------------------------------------------------------- This SF.Net email sponsored by: ApacheCon 2003, 16-19 November in Las Vegas. Learn firsthand the latest developments in Apache, PHP, Perl, XML, Java, MySQL, WebDAV, and more! http://www.apachecon.com/ _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
