On 01/12/15 18:59, RW wrote:
On Mon, 30 Nov 2015 20:45:25 +0000
Sebastian Arcus wrote:
After
setting up a site-wide bayes database as per the wiki instructions
and fixing file permissions etc., and feeding it about 300 spam
messages (I don't get a lot of spam in general) and 12,000 ham
messages of my own hand sorted email, the score for the same sample
spam message I mentioned in my original post jumped from 1.4 to
104.5 !!
I had no idea that bayes filtering can have such a dramatic effect on
a message with only a small amount of text in it!
It doesn't. At most it adds 3.7 points with default scores. You've
probably picked-up more points from net tests, but it's still a huge
change.
It seems to me that the huge score comes from a shortcircuit rule -
which adds or subtracts, depending on email, 100 points. However, this
is now also causing problems with proper spam not being recognised - but
I'm going to start a different thread about it - as it is a different
problem from my original post.
