Am 20.05.2016 um 16:20 schrieb Kris Deugau:
Emin Akbulut wrote:
I tried to train SA with tons of spam messages which contains zip file
(includes .js)
The max spam score was lesser than 5 so I did set 4 to delete messsages.

Then same kind of spam messages appear with the score of lesser than 2.

In short; training the SA seems not helpful.

What do you suggest to fight these spams?

I've had some luck doing that, but it takes a while

make 10 copies of such a message and change date/message-id header

in fact we have a "spamfilter-retrain /path/to/sample.eml" which creates 5 copies per call in the corpus folder and when something i train not get's BAYES_99 it's called as long as it hits BAYES_99 (except rare caeses which you need to ignore and can't tran that way)

why should i wait until i get the same crap 10 times fro outside?

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to