On 01/20/16 14:28, John Hardin wrote:
On Wed, 20 Jan 2016, Marc Perkel wrote:
Here's another way to use my evolution filtering idea with SA.
Get rid of all the rule scores and just make a list of the rule
names. From the rule names generate all combinations of those rule
names up to 4 rule names in a fingerprint and learn those
fingerprints as either ham or spam. sort of like this:
“A” “AB” “B” “C” “AC” “ABC” “BC” “D” “AD” “ABD” “BD” “CD” “ACD”
“ABCD” “BCD” “E” “AE” “BE” “CE” “ACE” “BCE” “DE” “ADE” “ABDE” “BDE”
“CDE” “ACDE” “ABCDE” “BCDE”
Then - when a new message comes in you make the same combo of
fingerprints from the rule names and then use my formula.
card(Test intersect Spam diff Ham) - card(Test Intersect Ham diff Spam)
Positive result = spam
Negative result = ham
Unfortunately this also requires training. It would render SA a
product that does not work out-of-the-box.
Actually it could include a pretrained corpus on the rules at least to
get people started. Could also have someone (like me?) provide it as a
service that SA would talk to. SA would send the tokens to the service
and the service would return a score.
--
Marc Perkel - Sales/Support
supp...@junkemailfilter.com
http://www.junkemailfilter.com
Junk Email Filter dot com
415-992-3400
