Jump to content


Bayesian Question

  • This topic is locked This topic is locked
No replies to this topic

#1 Randy Myers

Randy Myers
  • Members
  • 8 posts

Posted 04 October 2005 - 11:52 AM

Good Morning,My company recently purchased a model 300 spam firewall. I am trying to understand the Bayesian filter and your input/experience would be greatly apprecaited.At this time we have our scoring set at:Tag = 2.5Quar = 7Blaock = 8We have listed 318 messages as spam and 324 as not spam.I have noticed a lot of spam coming in, as a matter of fact, most e-mails receiving a rating above 0.1 are spam. We get a ton of e-mails that get tagged receiving ratings of 4 or higher, but not high enouch to get quaranteined.My boss is afraid to tag messages with scoring above 0.1 and quar above 4 (which is what I was suggesting). He thinks I should score all of these messages that are receiving ratings just over 4 as spam and that these would then start getting higher scores. He thinks the machine will learn these. From everything I read I believe that this would "over train" the machine and really make it a convaluted mess.What are your experiences? What kind of scores and settings seem to work best for you? Is my fear of "over training " the firewall by tagging all of these types of messages justified? And lastly, what type of scoring settings seem to work best for you?Thanks for all of your repsonces. I do appreciate them...I would be glad to leave my e-mail address if you want to contact me directly (I do not know if that is allowed on this forum)...