[Spambayes] spam forwarded through the list

[Simone Piunno]

Alle 23:47, lunedì 26 gennaio 2004, Tim Peters ha scritto:

> >> I'm running Mailman+spambayes integration since last April, 
> >> and I'm very happy with it.
>
> You should write up what you did on the spambayes wiki!  I bet many would
> be interested in the details.

Oh well, one year ago Barry Warsaw wrote a prototype patch to integrate 
SpamBayes into Mailman.  I took that patch, did some changes and ported it to 
new Mailman and SpamBayes releases.  It's patch SF #713522
http://sourceforge.net/tracker/index.php?func=detail&aid=713522&group_id=103&atid=300103
This patch implements a per-list classifier and wordinfo database, integrated 
with the Mailman UI (e.g. you can train from the list moderation queue, TTW).

> The most common reaction so far seems to be to switch lists so that only
> members can post to them, with other msgs held for moderator review.  

In my experience:
 - this a pain for the list admin, because many people subscribe with an
   address and send messages with another.
 - this starts to become ineffective.  In a desperate attempt at passing
   through filters, more and more spammers are correlating email addresses
   they collect, and try to send spam using a sender they know you have in
   your address book.  Things get much worse for worms.
Therefore it's not worth the pain.

> > One of the biggest problems with filtering this list is that people
> > can quite validly post entire spam messages ("why did this score
> Yup.  When python.org lists were filtered by SpamAssassin, SA was
> effectively disabled for this list, because it rejected too much legitimate
> traffic.

IMHO the biggest win in a per list classifier (integrating with the list 
manager) is that each list has a specific topic and therefore a specific set 
of clearly-non-spam clearly-on-topic words.  Filtering on this smaller domain 
it's easier to recognize abnormal text (spam or just off-topic messages).
When you move the filter at the MTA level, you lose the focus and filtering 
becomes harded.
Of course the same holds true moving the filter from the personal MUA to the 
shared MTA, albeit normally personal email is not so focused (or maybe it is, 
for non-english-speking people).
As Tony Meyer pointed out, on this list is perfectly normal to receive 
forwarded spam... to address the needs of this list you're pretty forced to 
weaken the spamicity score of those tokes for all other lists on python.org.
What would happen if python.org hosted a list where doctors discuss how that 
well known chemical works at enhancing you performance with women?  (I'm not 
using the real words here to avoid triggering the antispam, but you know what 
meant).

Cheers
  Simone
-- 
This signature intentionally left blank