[Rspamd-Users] Per user bayes filtering

[Stefan Bauer]

Per user means per user. Our observations have shown, that you need a
decent amount of mails trained by each user to have a real base to make
decicions.

thats why we do not have per user. As users are unable to differentiate
between spam an unwanted mails, i do not advice you to let your users train
filters.

All the real spam coming in is either that obvious that other checks detect
it already (fuzzy, dcc, dnsbl) or that new and hand-made, that need to be
learned manually once.

For us, a combination of spamtraps, current av-software, fuzzy database and
manual training of user reports made our users happy.

Also greylisting is quite important to have some time so dnsbl and fuzzy
list can keep up. 90% of user reported mails are already detected by checks
at the time they forward us their mails.

If you need further help, just contact me offlist. I'm always happy to help
fighting spam ;)

Stefan

Am Freitag, 8. Mai 2020 schrieb Mark Constable via Users <
users at lists.rspamd.com>:

> On 7/5/20 10:55 am, Mark Constable via Users wrote:
>
>> Hi, I've set per_user=true but I'm struggling to find any
>> documentation about how to manage per user bayes filtering. I noticed
>> the docs said to be careful of misc users incorrectly classifying
>> spam so that comment seems to indicate any user driven bayes
>> retraining (via sieve scripts) will influence bayes spam detection
>> stats for all users.
>>
>
> Could anyone point to some docs or explain what is the effective
> difference between per_user=false and per_user=true?
> --
> Users mailing list
> Users at lists.rspamd.com
> https://lists.rspamd.com/mailman/listinfo/users
>