https://www.logsat.com/spamfilter/forums/ Mon, 07 Sep 2026 02:53:11 +0000 Sat, 24 Jun 2017 22:07:07 +0000 http://blogs.law.harvard.edu/tech/rss Web Wiz Forums 11.04 360 https://www.logsat.com/spamfilter/forums/RSS_post_feed.asp?TID=7151 https://www.logsat.com/spamfilter/forums/forum_images/web_wiz_forums.png https://www.logsat.com/spamfilter/forums/ https://www.logsat.com/spamfilter/forums/forum_posts.asp?TID=7151&PID=14875&title=bayesian-filter-still-operational#14875 Author: LogSat
Subject: 7151
Posted: 24 June 2017 at 10:07pm

The Bayesian filter is very selective, and most of the emails it classifies will have probabilities that will be very close to either the upper or lower extremes, meaning very close to either 0% or 100%. That is normal, and is just the nature of the Bayesian filtering.
 
SpamFilter applies several different filters to incoming emails in a specific order to optimize performance. The Bayesian filter is one of the last ones to be used by SpamFilter, and thus will catch a very small percentage of spam compared to the other filters. In our own ISP for example, the Bayesian filter (when activated) catches only about 0.1% of spam, compared to 99.9% of the other filters (we disabled this filter a few years ago on our own live server). Adding to this, the Bayesian filters were "the thing" 9-10 years ago, and for a while this was the "star" filter in our SpamFilter. However the spammers have since learned how to easily bypass them, making the Bayesian filter even less effective. We often suggest disabling this filter for companies that receive large amounts of emails (~250,000 or more per day) as it does use a lot of resources with only minor gains.]]> Sat, 24 Jun 2017 22:07:07 +0000 https://www.logsat.com/spamfilter/forums/forum_posts.asp?TID=7151&PID=14875&title=bayesian-filter-still-operational#14875 https://www.logsat.com/spamfilter/forums/forum_posts.asp?TID=7151&PID=14874&title=bayesian-filter-still-operational#14874 Author: yapadu
Subject: 7151
Posted: 24 June 2017 at 2:42am

I know Bayesian filtering was popular with LogSat several years ago, but we ended up disabling it (can't really remember why).

Today we wanted to turn it back on and see how it performed.

We have it set to activate at 5000 messages ham/spam.

Currently the corpus file says we have much more than that processed:

Spam=15689
Good=12866

We have yet to see it catch anything, our logs indicate stuff like this:

06/24/17 06:13:43:788 -- (2664055264)     Token    Good    Spam    Prob is Spam
06/24/17 06:13:43:788 -- (2664055264) EMail from g-21254616771-21161-2102972555-1498284821079@bounce.news.crumble.com to m_w@example.com passes Bayesian filter - 0% spam  (172ms)

Should the token  good spam prob spam say anything?

All messages in the logs say 0% spam, I would have assumed messages have at least some 'spam' component so they should not all be 0%.


]]> Sat, 24 Jun 2017 02:42:33 +0000 https://www.logsat.com/spamfilter/forums/forum_posts.asp?TID=7151&PID=14874&title=bayesian-filter-still-operational#14874