[Rspamd-Users] Autolearn rejected messages as spam

[Carsten Rosenberg]

Hey,

I would also suggest autolearn, but have a look to your medium ham
scores. To have a good bayes database you should have learned nearly the
same amount of ham and spam messages.

On my systems, I currently set autolearn to [2,17]. Maybe this is not
sufficient for you ;)

Just 2 points:

Set autoleam spam to 15 will also trigger learning bayes on every mail
rejected by a force_action. Just keep that in mind, don't know if you
want that.

Set autolearn ham to -2 will probably only learn mails from senders
present in any of the dkim/dmarc ... etc whitelists. These are mainly
mails from Paypal, Google and all the other bis ones. The question is,
what kind of text bayes will learn from Paypal, Facebook or Ebay mails.

I would suggest set autolearn to the medium ham score of "normal" emails
you receive and spam to reject score or higher.

--

Carsten


On 05.12.18 17:50, Tim Harman via Users wrote:
> On 06/12/2018 3:18 am, J. Fahrner wrote:
> 
>> Hi,
>> I have a sieve rule in roundcube to learn messages as spam that go into the spam folder.
>> My problem is, that messages which are rejected by rspamd (score > 15) don't go into the spam folder and therefore are not learned.Send [1]
>> Is it possible to learn rejected messages by a lua rule or some config setting?
>>
>> Jochen
> 
> You can use the autolearn command. 
> 
> I have 
> 
> autolearn = [-2, 15]; 
> 
> in my local.d/classifier-bayes.conf configuration. 
> 
> This learns all messages with a score greater than 15, including those
> that are rejected.  You might want to tune the -2 to be a lower score
> (-5 for example) before a message is learned as ham. 
> 
> Links:
> ------
> [1]
> https://muppetz.com/rc/?_task=mail&_action=compose&_id=7040046785c080142999d3&_extwin=1#
>