Mailinglist Archive: opensuse (1702 mails)

[John Andersen <jsamyth@xxxxxxxxx>]

Anton Aylward wrote:
> Roger Oberholtzer said the following on 09/07/2009 10:04 AM:
> > [...]
> >
> > All works great. Spam Assassin detects lots of messages as being spam.
> >
> > My question is how Spam Assassin can learn about messages that were
> > missed as being SPAM? 
>
> RTFM.  See 'sa-learn'
>
>
> > Since this is done outside of my mail reader
> > (evolution), I can't just tell the reader to tell that it is SPAM and
> > that the filter should learn. At least I would not imagine this would be
> > the case.
>
> Yes you can.
>
> 1. Create a folder SPAM, and subfolders "isSpam", "false-positive' and
>    'false-negative'.
>
> 2. Have procmail put the spam it detects in "isSpam".
>
> 3. Manually put items that go into "isSpam" that aren't spam into
>    "false-positive"
>
> 4. Manually put items that are in your inbox that *ARE* into
>    "false-negative".
>
> 5. Have CRON run 'sa-learn' with '-spam' on 'false-negative'
>    and then empty it
>
> 6. Have CRON run 'sa-learn' with '-ham' on 'false-positive'
>    and then empty it.
>
> 6. Clean out "isSpam" periodically, either by CRON or manually.
>
> You are telling the automatically run learning mechaism of SpamAssassin
> ('sa-learn') what are examples of spam that it missed ('false-negative')
> and what it thought was spam that wasn't ('false-positive').
>
>
> In practice, I would not rely 100% on SpamAssasin.
> I run blacklist and whitelist filters and sanitizer from procmail
> *before* SpamAssassin:
>
>
> INCLUDERC=$LIB/whitelist.rc
> INCLUDERC=$LIB/mailinglists.rc
>         # Another kind of whitelist
> # INCLUDERC=$LIB/blacklist.rc
> INCLUDERC=$LIB/attachment.rc
> INCLUDERC=$LIB/likelyspam.rc
> INCLUDERC=$LIB/fonts.rc
>
> This reduces the load on SpamAssassin.
>
> You can find examples of the white/black/mailing/font filters in the
> many examples of how to use procmail by googling.

These are all very good instructions.

But watch out:  If AMAVIS is involved it will all be for not.

AMAVIS uses its own site-wide spamassassin Bayes database, rather
than user databases, and the above sa-learn have to be pointed to that
database instead of the users.

Secondly, if you use imap on your machine (if not, WHY not?) the
directories you need to scan are not in the user's home directory
but rather in the Imap areas, usually /var/spool/imap/user if using
cyrus.

I plan to drop AMAVIS all together in my next major install, as this
site wide spamassassin nonsense is totally wrong.
-- 
To unsubscribe, e-mail: opensuse+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: opensuse+help@xxxxxxxxxxxx