[spambayes-dev] Small Company Problem
Neil Schemenauer
nas at arctrix.com
Mon Dec 6 21:38:18 CET 2004
On Mon, Dec 06, 2004 at 01:07:17PM -0500, Peter Fraser wrote:
> To add interest to the spam, the spammer often use
> employee's names and email address. The net result
> is that spambayes decides that these words are bad!
>
> So short email sent to employees in the company
> end up being classed as Spam or Maybe Spam. Training
> doesn't help. The true spam always overwhelms.
Hi Peter,
Those words should appear in both ham and spam messages and
therefore should have a neutral score. It sounds like you may have
a ham/spam imbalance in your training set (e.g. more spam messages).
For best results, the number of spam messages in the training set
should be equal to the number of ham messages.
Neil
More information about the spambayes-dev
mailing list