Re: Really good spam filtering (Bayesian)

"Jem Berkes" <jb@users.pc9.org> wrote in message
news:Xns93BC90BBFF2CDjbuserspc9org@205.200.16.73.. .
| I don't know why this project hasn't received more attention. I am
| currently using SpamProbe, a Bayesian filter by Brian Burton
| http://spamprobe.sourceforge.net/
|
| This filter is invoked by procmail. spamprobe assigns a label and scores
| each incoming email. Unlike spamassasin and homegrown techniques, the spam
| score is NOT determined by arbitrary thresholds. With Bayesian filtering,
| each word and phrase in the email is looked up in a database to calculate
| probability of spam based on prior experience with the word combinations.
|
| Bayesian filtering has only two downsides as far as I can tell: first, you
| have to "train" the system to teach what is spam and what is good, along
| with occasional corrections. Second, a rather large database of words is
| required. In my experience this uses around 20 to 30 MB of storage.
|
| On my most "hopeless" email address (the one I post to USENET with ;) I
| have been using spamprobe for many months, and achieving about 98% filter
| accuracy. I'm laughing, seriously.
|
| I have also now released a group of scripts that add a web interface to
| spamprobe, so that you can reclassify mail without having to type in
| console stuff. If you want to check out these bash scripts, visit:
| http://www.pc-tools.net/unix/webfilt/
|
| --
| Jem Berkes
| http://www.pc-tools.net/
| Windows, Linux & UNIX software
|

vipuls razor is utilized by spamassassin if you install it first. the
combination is quite effective with minimal training.

ken k