Re: FW: spam scores being neutralized by bayes

Thank you all,

Joshua, I think this is the route I'll have to take, bayes just appears 
to be problematic at this stage.  Any spam actually being caught as 
spam, removing bayes is not going to do much score wise

I have seen a massive increase in spam getting through, all coming from 
.eu and .xyz addresses and all being allowed through thanks to Bayes  
doing a "-1.900 BAYES_00"


Although maybe just as worryingly, I've had spam hit the bayes rule 
*and*  "-2.200 RCVD_IN_IADB_VOUCHED".   Looks like the spammers are 
getting better...

1.347 	RCVD_IN_BL_SPAMCOP_NET 	Received via a relay in bl.spamcop.net
0.100 	DKIM_SIGNED 	Domain Keys Identified Mail: message has a signature
0.010 	T_REMOTE_IMAGE 	(No description provided)
0.001 	HTML_MESSAGE 	HTML included in message
-0.001 	SPF_PASS 	SPF: sender matches SPF record
-0.001 	RCVD_IN_IADB_LISTED 	Participates in the IADB system
-0.001 	RCVD_IN_IADB_SENDERID 	IADB: Sender publishes Sender ID record
-0.010 	T_RP_MATCHES_RCVD 	Envelope sender domain matches handover relay 
domain
-0.059 	RCVD_IN_IADB_SPF 	IADB: Sender publishes SPF record
-0.100 	DKIM_VALID 	Message has at least one valid DKIM or DK signature
-0.235 	RCVD_IN_IADB_RDNS 	IADB: Sender has reverse DNS record
-1.470 	RCVD_IN_IADB_OPTIN 	IADB: All mailing list mail is opt-in
-1.900 	BAYES_00 	Bayesian spam probability is 0 to 1%
-2.200 	RCVD_IN_IADB_VOUCHED 	ISIPP IADB lists as vouched-for sender



¬¬JS


On 2015-07-08 23:13, Joshua Small wrote:
> Your point below was always my issue with Bayes. The whole "you just
> need to train it" argument might be fine for users of this list, but
> the customers running these products are comparing them to everything
> else on the market, which "just works".
> 
> Given I pretty much never saw Bayes trained correctly, I've only ever
> seen bayes fall into these categories: Spam flagged as spam that would
> have been captured by other rules anyway, spam that would have been
> flagged and quarantined if not for an incorrect negative bayes score,
> and mail flagged as spam by Bayes that was a false positive. In short,
> I disabled bayes on all my environments.
> 
> -----Original Message-----
> From: Maia-users [mailto:[email protected]] On Behalf
> Of Jason Street
> Sent: Wednesday, 8 July 2015 9:13 PM
> To: Blake Hudson
> Cc: [email protected]
> Subject: Re: [Maia-users] spam scores being neutralized by bayes
> 
> On 2015-07-07 15:23, Blake Hudson wrote:
>> Jason, it sounds like your Bayes database has learned SPAM. This could
>> be from incorrect user reports (messages were reported as NOT SPAM
>> when they were SPAM) or from auto learning.
> 
> Potentially but the users are pretty hands off and ignore their cache
> contents in the web interface, except for checking the spam cache for
> anything valid caught.
> 
> For the most occasion "non-spam cache" is ignored and left to be auto 
> cleared.
> 
>> My suggestion is to turn off auto learning (bayes_auto_learn 0 in
>> local.cf) and use Maia's digest function to confirm HAM/SPAM daily in
>> order to train bayes. To help speed up the learning you may want to
>> clear out your current bayes db. You can do this with the
>> maiadbtool.pl script and the --clear-bayes option.
> 
> Forgive me ignorance  but how do I do the digest function you suggest?
> 
> _______________________________________________
> Maia-users mailing list
> [email protected]
> http://www.renaissoft.com/cgi-bin/mailman/listinfo/maia-users
> _______________________________________________
> Maia-users mailing list
> [email protected]
> http://www.renaissoft.com/cgi-bin/mailman/listinfo/maia-users

_______________________________________________
Maia-users mailing list
[email protected]
http://www.renaissoft.com/cgi-bin/mailman/listinfo/maia-users