At 12:07 PM 12/29/2004, Rakesh wrote:
Can any body tell me when does Bayes00 gives the score. Is it

1) If a mail has a lot of tokens that Bayes has never seen before.
or
2) If the mail has a lot of tokens that Bayes has previously learnt has spam.

Neither.

BAYES_00 fires when there are a lot of tokens that bayes has previously learned as HAM.


BAYES_99 fires when there are a lot of tokens that bayes has previously learned as spam


BAYES_50 fires when there are few tokes learned before, or there's a roughly equal number of spam/ham tokens.

The reason of my weird question is that recently I have suddenly started recieving a huge chunk of Payroll Spams from Indian spammers and my Bayes always gives them a score of -4.9. And after individually giving feedback of every mail i manage to get some better score from bayes on these mails.

How are you "giving feedback".. are you forwarding messages to a script account, or are you feeding the real message to sa-learn directly.


I think my bayes is badly poisoned, however i need to give a good explaination to my Boss before i nuke my bayes and start all over again.

Sounds a bit like it. You can double-check by running one of those messages through spamassassin -D.. This will show you the tokens matched, and the score of each individual token (0.0000 to 0.99999) used in coming up with the overall probability.






Reply via email to