Bad to *over* train Spamsieve?

[jmchavez]

Hi Michael, I had a problem with a client and her Spamsieve (in use for over 4 years) not doing a good job of catching spams, etc., and as I reviewed with her how she uses Spamsieve, she said that she would periodically go into her spam folder, look for any false positives, and once that was done select the entire message list and *train as spam*. I was taken aback because I had told her to only train as spam in the spam folder the messages colored yellow as those were the ones that Spamsieve was uncertain about. I saw that her corpus and lists were quite huge and I just figured lets reset and start over and see what happens.

So I was curious if there was anything in the manual that specifically said there's a danger to marking spam as spam again but I couldn't find anything along those lines. The manual does say to 'correct mistakes' and promptly, etc., and in going over some of your posts here you do imply that the inverse is frowned upon, but I wanted to point-blank ask you if excessive OVER training of Spamsieve is liable to backfire, etc. If there is such a thing, as a suggestion that might be good to spell out in the next manual revision, as there are people out there that, using this one person as a sample, might need it spelled out for them.

Thanks much!

[Michael Tsai]

The manual does say to 'correct mistakes' and promptly, etc., and in going over some of your posts here you do imply that the inverse is frowned upon, but I wanted to point-blank ask you if excessive OVER training of Spamsieve is liable to backfire, etc.

I’ve updated the manual. After the initial training, it is not necessary (or recommended) to train SpamSieve with messages that are not mistakes. Yellow spam messages do not count as mistakes.

[jmchavez]

Thanks so much for that. So I'm clear: I thought the yellow spams are what trigger Growl to say 'this is an uncertain spam' (or words to that effect) - Am I mistaken? And so regardless, even if they are 'uncertain' we are better off *not* 'affirming' they are spam?

[Michael Tsai]

Thanks so much for that. So I'm clear: I thought the yellow spams are what trigger Growl to say 'this is an uncertain spam' (or words to that effect) - Am I mistaken? And so regardless, even if they are 'uncertain' we are better off *not* 'affirming' they are spam?

The Uncertain Spam Growl notifications are controlled by the A spam message is uncertain if preference. Depending on the number that you’ve selected, this may or may not correspond to yellow. In any case, “uncertain” means that SpamSieve thought it was spam, just not as much so as other messages. You are better off not affirming, only correcting. (The exception to this rule would be if you receive an atypical good message and want to be sure SpamSieve knows that type of message is good.)

[jmchavez]

Many thanks Michael!