You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Justin Mason <jm...@jmason.org> on 2005/11/15 20:51:34 UTC
Re: apple mail better than SA?
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Matthew.van.Eerde@hbinc.com writes:
> Simon Butler wrote:
> > read this link for more info on apple mail spam filtering
> >
> > http://www.macdevcenter.com/pub/a/mac/2004/05/18/spam_pt2.html
>
> Sounds like per-user Bayes to me.
Pretty similar, alright. It's another probabilistic classifier, like our
BAYES_* rules. It uses a fancier algorithm (LSA), but research hasn't
really proven if that helps in real-world use; some reviews have said our
results are better, some say Apple's are. All I know is that SpamAssassin
requires no upfront hand-training ;)
To answer the original question, the easiest way to get SpamAssassin's
accuracy to the same levels is to chain the classifiers; keep copies of
all the mails Apple's app says are ham and spam, separately, then feed
those to sa-learn.
You may need to do some hand-classification if Apple's classifier makes
mistakes, of course. No point training SpamAssassin with incorrect
data.
- --j.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Exmh CVS
iD8DBQFDejxGMJF5cimLx9ARAm+uAKCbjGLKpXQjQYGka3W4wNfkhGitmACfWqjw
RIR8QAo/CbzkIBfat5XQhkw=
=ejY+
-----END PGP SIGNATURE-----
Re: apple mail better than SA?
Posted by Vivek Khera <vi...@khera.org>.
On Nov 15, 2005, at 2:51 PM, Justin Mason wrote:
> You may need to do some hand-classification if Apple's classifier
> makes
> mistakes, of course. No point training SpamAssassin with incorrect
> data.
The main problem with Apple's Mail.app is that you can't tell it NOT
to learn from some folders, such as my spamassin users list folder,
or my spam-l folder. It learns it all, which makes for difficult
filtering of real spam as opposed to discussion about spam.
I had to turn it off.
Re: apple mail better than SA?
Posted by Theo Van Dinter <fe...@apache.org>.
On Tue, Nov 15, 2005 at 11:51:34AM -0800, Justin Mason wrote:
> Pretty similar, alright. It's another probabilistic classifier, like our
> BAYES_* rules. It uses a fancier algorithm (LSA), but research hasn't
> really proven if that helps in real-world use; some reviews have said our
> results are better, some say Apple's are. All I know is that SpamAssassin
> requires no upfront hand-training ;)
It also depends on use I'd say. A previous mail was saying that the
MUA was catching everything that SA let through. So the MUA gets to
focus on only catching a certain class of mails. If it had to try
catching everything, it may not function so well. (my apple mail still
misses a bunch of spam that also misses SA, and I generally haven't
been impressed.)
--
Randomly Generated Tagline:
"Winnie the Pooh is the balls." - Matt