[Prev] Thread [Next]  |  [Prev] Date [Next]

Re: Yet another thread about AWL Antonio Gutiérrez Mayoral Tue Feb 21 09:00:27 2012

So, if I have a lot of spam getting negative score from AWL, the problem was
that in previous moment these spam was not classified as "spam" (getting
high score) and in conclusion AWL understand that is a legal sender?

Sorry if this cuestion is trivial :-( I am not pretty sure
if I am understanding the behaviour...

Thank you.


2012/2/21 Bowie Bailey <[EMAIL PROTECTED]>

> On 2/21/2012 10:42 AM, Antonio Gutiérrez Mayoral wrote:
> > I know that Bayes and AWL are different thinks, but I think that if
> > your non-spam
> > thresold is low (-0.001 in my case) mails classified are non spam
> > could be false positives. [1]
> > It doesnt affect to the AWL table? I think that AWL has auto-learn...
> >
> > [8409] dbg: auto-whitelist: sql-based get_addr_entry: found existing
> > entry for [EMAIL PROTECTED]|ip=XX.YY
> > [8409] dbg: auto-whitelist: sql-based [EMAIL PROTECTED]|ip=XX.YY
> > scores 13/11.213
> > [8409] dbg: auto-whitelist: AWL active, pre-score: 8.001, autolearn
> > score: 8.001, mean: 0.862538461538461, IP: X.Y.W.Z
> >
> > So, its AWL deprecated? Is better solution remove the addresses from
> > the DB? Im confused...
> The threshold only affects the Bayes learning.  AWL always learns.
> There is really no such thing as a false positive with AWL.  AWL is just
> a score averager.  It keeps track of the average score for each sender
> and adjusts the score for new messages to prevent them from being too
> much higher or lower than the average.  A sender who usually sends
> low-scoring emails will get a negative AWL score when he suddenly sends
> a high-scoring email.  A spammer getting a negative AWL score simply
> indicates that SA did not do a good job of catching his previous spams.
> I don't remember if AWL is still on by default or not.  I'm still
> running it since it seems to work fine for me.
> --
> Bowie

Antonio Gutiérrez Mayoral <[EMAIL PROTECTED]>