Hello..
I am currently in the process of implementing a site-wide Bayes database.
I understand the difficulty in processing forwarded mails (explained in the Wiki).
However, i was wondering what are the implications of just using the bodies of forwarded mails.. i.e. not worrying about the headers... In the organisation i work at, setting up everyeone to forward things as attachments is prohibitively difficult...
So, what if i strip ALL headers and just feed the body of the mail to sa-learn. Is this better than nothing, or will my Bayes db end up skewed as it has been header-deprived?
Has anyone any experience with this?
I'm fairly certain sa-learn won't even be able to properly parse a message with no headers at all. (it starts off assuming a message which starts with headers, since a message without headers isn't vaild)
So, I'd strongly recommend not feeding invalid input to sa-learn.
