Re: Feeding bayes outbounds

[mouss <us...@free.fr>]

Logan Shaw wrote:
> On Mon, 21 Aug 2006, jdow wrote:
>> From: "Joe Zitnik" <JZ...@hfcc.net>
>
>>> Our scanning program has the ability to archive all e-mail, both 
>>> inbound
>>> and outbound, which we have been doing for months now.  Given that your
>>> outbound mail is almost certainly ham, the majority of it's content is
>>> going to be specific to our business sector, wouldn't feeding outbounds
>>> through bayes manually be a win win situation?  Am I oversimplifying
>>> things, or am I missing something with that logic?
>
>> If the terms in the outbound mail are likely to be the same as
>> acceptable terms on the inbound mail that may be true.
>
> There will be a strong correlation, because most users tend to
> quote the entire e-mail when they reply.  Granted, this only
> affects replies, but if someone is quoting a message you sent
> to them, Bayes is probably right to score that as ham.

sure, but
- people don't reply to newsletters.
- don't reply to all ML mail
...

and

- virus may send outbound mail
- autoreply robots may replay to spam
...


so don't use your outbound mail as ham feed unless you know what you're 
doing.

More importantly, Bayes is a classification method that has been shown 
to work in some situations, without a mathemetical proof. getting beyond 
these "situations" require serious experimentation. or you run the risk 
to join a cargo cult.