Not Tagging As Well As It Did

[Mike Carlson <mi...@mcarlson.net>]

I have been running SA for a few months and initially it was tagging 90% of the incoming spam. Now it only gets about 50%. I have been training and training with sa-learn but the tagged spam count doesnt seem to be increasing.

Is there a way to check for a corrupt bayes DB? Where is the bayes DB located (FreeBSD)? It is the bayes_seen under the .spammassin dir in the users home directory? I was thinking I would try moving it and seeing where the counts go.

Also, should I be auto-learning? It seems that the more messages I learn the tagged count either goes nowhere or it gets lower.

Our users loved SA when I first set it up, but now they are getting a bit annoyed at the fact that their inboxes are getting filled with spam again even though the amount of mail that we are getting hasnt increased.

--Mike

Re: Not Tagging As Well As It Did

[Robert Menschel <Ro...@Menschel.net>]

Hello Mike,

Friday, March 19, 2004, 7:17:40 AM, you wrote:

MC> I have been running SA for a few months and initially it was
MC> tagging 90% of the incoming spam. Now it only gets about 50%. I have
MC> been training and training with sa-learn but the tagged spam count
MC> doesnt seem to be increasing

Then something is wrong.

What version of SA are you using?  If not 2.62 or 2.63, upgrade.

Have you checked to make sure you're getting reasonable Bayes scores? If
Spam is coming through with BAYES_00 to BAYES_30 you may have a problem
in your Bayes database.

Check the custom rule sets at
http://wiki.apache.org/spamassassin/CustomRulesets -- they're a big help
when used intelligently.

Bob Menschel

Re: Not Tagging As Well As It Did

[Bob George <ma...@ttlexceeded.com>]

Mike Carlson wrote:

> I have been running SA for a few months and initially it was tagging 90% 
> of the incoming spam. Now it only gets about 50%. I have been training 
> and training with sa-learn but the tagged spam count doesnt seem to be 
> increasing.

Have you checked the scoring of messages that are slipping through?

How have you set up bayes, site-wide or individual?

Are you sure bayes is even being used?

What rules are you using?

> Is there a way to check for a corrupt bayes DB? 

Check out 'man sa-learn', especially the dump and rebuild options.

> Where is the bayes DB 
> located (FreeBSD)? It is the bayes_seen under the .spammassin dir in the 
> users home directory? I was thinking I would try moving it and seeing 
> where the counts go.

Or perhaps just turn bayes off temporarily.

> Also, should I be auto-learning? It seems that the more messages I learn 
> the tagged count either goes nowhere or it gets lower.

Only use auto-learning if things are working to start with. If you're 
auto-learning and are getting LOTS of false negatives/positives, unless 
you're very diligent with the retraining on those messages, things will 
only get worse.

> Our users loved SA when I first set it up, but now they are getting a 
> bit annoyed at the fact that their inboxes are getting filled with 
> spam again even though the amount of mail that we are getting hasnt 
> increased.

Er, doesn't that mean they're getting less regular mail? Not sure 
spamassassin can help with THAT! :)

Good luck!

- Bob