On 6/4/2014 10:09 AM, Daniele Paoni wrote:
On 06/04/2014 04:01 PM, Bowie Bailey wrote:
Your Bayes DB is messed up. Spam should never get BAYES_00. Check your
training procedures to make sure that spam is not being accidentally
trained as ham. The best way to fix the problem is to wipe the DB and
start over. If you have a saved corpus of ham and spam, you can use
that to start off the training of the new DB -- just make sure it's
sorted properly or you'll wind up with the same problem all over again.
Ok I will do it, but I don't think it will fix this particular message
as the BAYES_00 is only -1.9.
Is there some rule that recognize words in "obfuscated" html ? The
message is made with random letters and uses colors to show only
relevant characters.
Please keep list messages on the list so everyone can see them. I'm
just another user. Other people may be able to add additional suggestions.
The problem isn't that BAYES_00 subtracts 1.9 points. The problem is
that you DON'T get the 3.5 points added from BAYES_99, which is what
should have hit. Even if it got to BAYES_60, you still would have
gotten 1.5 points, which would have increased your score by 3.4 points.
There are some rules for obfuscated words both in html and plain text.
But the spammers are always finding new methods. Paste the original
message text to www.pastebin.com and give us the link. That way we can
see what the message looks like and give you better suggestions for how
to catch that type of message.
--
Bowie
