Lots of spam gets through because of BAYES_00 -2.60

[Hugo van der Kooij]

On Thu, 13 Sep 2007, Greg Matthews wrote:

> Chris W. Parker wrote:
>>  On Wednesday, September 12, 2007 5:38 AM Greg Matthews said:
>> 
>> >  In summary, if Bayes is not working for you, its worth taking the time
>> >  to get it right rather than simply skewing the scores.
>>
>>  Would you mind giving more details on how I can take the time to "get it
>>  right"?
>
> theres no substitute for reading the docs! SA is a complex piece of software 
> and you need to understand at least how it works. but...

Take a word of advice from the competition: 
http://www.barracudanetworks.com/ns/downloads/Barracuda_Bayes.pdf

The core advice to be carefull what message you feed to your bayesian 
database has proven to be sound on MailScanner installations as well.

In short:
  1. Feed it ~250 HAM and ~250 SPAM messages to start with.
  2. Now sparsely feed it SPAM or HAM messages.
  3. Feed SPAM if it classified too poorly by other means.
  4. Feed HAM messages if they get tagged.

Try to keep the numbers relative low to make it work more accurate.

Hugo.

-- 
 	hvdkooij at vanderkooij.org	http://hugo.vanderkooij.org/
 	    This message is using 100% recycled electrons.

 	Some men see computers as they are and say "Windows"
 	I use computers with Linux and say "Why Windows?"
 	(Thanks JFK, for this quote of George Bernard Shaw.)