Bayesian Options
From ASSPSMTP
On This Page |
How It Works
Details
Bayesian FAQs
Bayesian Configuration
based upon version 1.3.5(9.18) Bayesian Check Theory of Operation (DoBayesian)
If activated, the message is checked based on Bayesian factors in Spamdb. This needs a fully functional spamdb built by rebuildspamdb.pl. For starters it is best practice to put this inactiv and built the spamdb collection with the help of DSNBL and URIBL. BlackRE and WhiteRE checks will be performed even if inactive 0 = disabled, 1 = block, 2 = monitor, 3 = score
Skip Bayesian Check* (noBayesian)
Mail from/to any of these addresses are ignored by Bayesian check, mails will not be stored in spam/notspam collection. Accepts specific addresses (user@domain.com), user parts (user) or entire domains (@domain.com)
Bayesian Probability Threshold (baysProbability)
Messages with spam-probability below this threshold are considered Ham. Recommended '0.6'.
Bayesian Confidence Threshold (experimental) (baysConf)
Spam-Mails having a confidence below this threshold are passed in TestMode . Spam-Mails having a confidence above this threshold are blocked. Set this only above 0 if you are familiar with the bayesian statistics used in ASSP. Messages that are processed by the bayesian check get a spam-probability score and a confidence score. The confidence score in assp is a quality indicator. A confidence near 0 would mean the probability score is like a wild guess. A confidence score near 1 would mean that it's pretty sure that the bayesian analysis result is correct. The confidence threshold is an allowance to process a Bayesian Spam as-if in Bayesian TestMode, if the message's *confidence* score is lower than the confidence threshold. Set this level to a specfic value, let's say .001, then: - messages with spam-probability higher than 0.6 and a confidence of less than .001 would come through as in test mode - messages with spam-probability higher than 0.6 and a confidence of more than .001 would be blocked - messages with spam-probability less than 0.6 would pass The 0.6 threshold can be set in baysProbability. Setting the confidence threshold to 0.001 will result (in our experience) in 1/3 passing in testmode because there is too much doubt about the correctness of the result of the bayesian check. empty = disabled.
Reduce Scoring for Low Confidence (baysConfidenceHalfScore) Spam-Mails having a confidence below the threshold and get half of the normal penalty score for bayesian hits.
No Subject Tag in Testmode (NoTagInTestmode) In Bayesian TestMode "/Prepend Spam Subject/" is only added if the message is Spam and confidence is higher than Bayesian Confidence Threshold" .
Add Bayes Confidence Header (AddConfidenceHeader) Adds a line to the email header "X-Assp-Bayes-Confidence: 0.0123".
