On Fri, 2006-01-27 at 19:27 +1100, Rod Butcher wrote:
Ok, but still, won't marking "new" messages as not-Junk also make SA do the analysis?I'm not 100% sure. But remember, filters err on the side of caution - if it's not sure it won't flag it as junk. Hence telling it something isn't junk when it already has decided it isn't is unlikely to teach it much, if anything you're just confirming what it already believes. On the other hand, telling it something is junk when it has not yet decided it is, or telling it someting isn't when it thinks it is, presents it with A1 opportunities to learn from - learning from mistakes. From my own experience of training evolution, I needed to flag +- 100 examples as junk, and only 1 or 2 corrections (flagging junk as not junk). But this last action will depend on the nature of your "legitimate" email - if you get a lot of key spamlike words in your regular email you may indeed have to correct quite a few bad spam decisions.
This was discussed very recently. The "Non-junk" button does not do what most people (including me) would expect. The thing it does is cancels out a previously messages marked as "spam". So if you first mark a message as "junk" and then apply "not junk" to it, the netto result is zero. On the other hand, applying "not junk" on a message that was not previously marked as "junk" does absolutly *nothing* (which is very bad imho).
Attachment:
smime.p7s
Description: S/MIME cryptographic signature