At 10:59 AM 9/4/2003 -0500, Thomas Cameron wrote:
All -

I have a client with a Red hat 9 + SA 2.55 + spamass-milter server in
front of a Lotus Notes server.  The client has hired a Notes developer
to give their users an extra button in Notes to forward false negatives
to a spam account the Linux server so I can run sa-learn --spam on the
messages.  The thing is, the messages get sent as base64 encoded
attachments (see example below).

Can sa-learn use this format to learn from?

Well, bayes can learn from base 64 messages, but you absolutely should not use forwarded messages for bayes training.


The problem is that the bayes engine winds up learning "anything that looks like it was forwarded via lotus notes is spam", which is clearly not the desired effect.

For bayes to work the message fed to bayes must not be modified in any way from what it looks like as it comes in from the network. The bayes engine does also examine some headers, so those need to be the same as the originals as well, possibly with the exception of added Received: lines.

Anything else is a recipe for a badly trained bayes database.

See the spamassassin FAQ as well:

http://spamassassin.taint.org/faq/index.cgi?req=show&file=faq05.003.htp






------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk

Reply via email to