Print Page | Close Window

Bayesian Filtering Unsatisfactory

Printed From: LogSat Software
Category: Spam Filter ISP
Forum Name: Spam Filter ISP Support
Forum Description: General support for Spam Filter ISP
Printed Date: 21 August 2018 at 6:14pm

Topic: Bayesian Filtering Unsatisfactory
Posted By: Guests
Subject: Bayesian Filtering Unsatisfactory
Date Posted: 11 May 2004 at 9:18pm

Is there going to be any way to get false negative emails back into the corpus DB? I've reduces the Bayesian threshold to 2% but it still only catches a few a day. Currently my Spam to Good email ratio is about 4:1 but I suspect it is more like 6:1. Can you update us on the status of the capability to reverse email falsely tagged as good?

One thought I have is to have an "optional" Good Email DB similar to the Quarantine DB. If one chose to use this optional DB, they could go through and send false negatives (spam) to the corpus and reverse the good score they got before.  

I think getting false negatives out of the DB is crucial to get any kind of accurate filtering. At a probablitiy of 95% setting, it would take 20 spam catches to reverse one false negative.




Posted By: LogSat
Date Posted: 13 May 2004 at 12:15am


Have you tried, just for the initial training period, to implement more aggressive filtering that will catch more spam? Rejecting  "no reverse DNS", adding more powerful keyword searhes, and more MAPS servers could increase your spam/good ratio to the more realistic 6:1 - 8:1 as you correctly pointed out.

We will provide for the ability to add "false negatives" soon, but it will still be very helpful to have strong training during the initial deployment.

Roberto F.
LogSat Software

Print Page | Close Window