RE: [Bmf-user] A few questions...

[Matt <ma...@lo...>]

> Tonight, at the request of a friend who has had been using bmf for a 
> while now, I integrated bmf into my custom mail server 
> software. I have 
> trained it with about 300 spams and about 100 hams. However, 
> the first 
> ten messages I've received with the system were all spam and 
> none got a 
> Spamicity score greater than 0.02 or such. Is this a result of not 
> enough training?
> 

I had a similar problem initially, however this was caused by me
training BMF using one account (root) and then receiving mail and
processing through BMF as another user.
Effectively BMF was not trained for the user, and thus nothing was spam.


> Also, it's not clear in the docs, but when training, am I to include 
> message headers? It's quite easy to just use a large mailbox 
> file as an 
> input when training. Perhaps this is where I've gone wrong?

I include everything, surely this is the way forward, doesn't this then
train BMF to use the sender and other information to help evaluate
spamness.

For instance the email I just replied to was classified...
'X-Spam-Status: No, hits=0.000000 required=0.900000'

It would be interesting to process the same body from an unknown source.

M