Plugging anti-spam software: SpamProbe

This isn't my own software, but I have talked a lot with the developer over
the sourceforge message forums. For server side spam filtering, this really
is still the best overall bayesian-type filter I have seen to date

http://spamprobe.sourceforge.net/

Try out the latest development versions. He now has a hash based local
database format (just build without extra parameters and it will become the
default db format). The only weakness of this software in PAST versions was
that the db would continually grow as the software re-trained itself on new
mail. With the latest database format, the hash database has a fixed size.

For instance, I am using a 16 MB hash database per user account. The filter
is pipe based and sits in procmail; incoming emails run "spamprobe train".
Software is VERY fast (C++), far better than resource impact of spam
assassin or spambayes. The filter just generates a score and it's up to you
what you want to do with it (more procmail filtering maybe).

What's nice is that each email that is received also automatically retrains
the filter. If you receive spam containing terms that aren't already known,
they will be associated with spam. So the filter follows morphing spam
techniques. Train it well to begin with, as with any bayesian.

To give you an idea of how well it can perform when well trained, I get
several hundred emails daily on different accounts. The accuracy rates on
each account are 98.9%, 99.9%, 99.9%

This is why I post on USENET unmunged :)

--
Jem Berkes
Software design for Windows and Linux/Unix-like systems
http://www.sysdesign.ca/