[Spambayes] Outlook 2003 is eating our lunch ?
tim.peters at gmail.com
Mon Nov 22 01:48:26 CET 2004
> I saw a presentation at Microsoft TechEd a couple of years back about
> the spam filter technology going in to Outlook 2003 / Exchange 2003. It
> did not impress me that much. ...
There used to be great info here:
They reverse-engineered the code and data files, and learned a lot
about 2003's filtering internals. Alas, it's all been removed now
("due to legal issue").
> The reason it works reasonably well for a lot of people is because of
> the huge Hotmail corpus that they use to statistically determine the
> point values for each token. They have also released a few updates to
> the rule sets since Outlook 2003 RTM, these perform marginally better
> against current spam.
The biggest problem is that it doesn't learn, and everyone's 2003
comes with the same scoring. That includes copies of 2003 that
spammers buy, so they can easily fine-tune their campaign to evade
2003's filtering just by sending msgs to themselves to see how they
score. Of course they score the same for everyone else too.
> That said, I get something near 98% capture rate with zero false
> positives using SpamBayes, so I won't be ditching it anytime soon in
> favor of Outlook 2003's filter or the Exchange 2003 Intelligent Message
Me neither. 2003's filter is fast, fixed-size, and does a very good
job on the spam it was trained against. That's about it. If the
spammers would just agree to send nothing except messages 2003 was
trained it, it would be even better than SB <wink>.
More information about the Spambayes