A newbie question about training

[David Relson]

At 04:42 PM 7/23/03, Rob Tanner wrote:
>Hi,
>
>Got bogofilter up and running and gave it some training.  Already it's at
>99%!!  We are currently delivering whether it's labelled as SPAM or not,
>and prepending the flag "[SPAM]" to the subject line by setting the
>spam_subject_tag in bogofilter.cf.
>
>My question is whether bogofilter also examines the subject line when
>analysing the message when training, and if so, do I need to remove that
>"[SPAM]" flag prior to feeding the message to bogofilter, or does it
>sutomatically discount it?

Rob,

Glad you've got bogofilter running.  You can safely leave the "[SPAM]" tag 
in place when you train.  The result will be that token subj:SPAM will have 
a spam score of 1.0 and any incoming messages that have "Subject: ... SPAM 
..." will have a token with a high score.  Since bogofilter's score is 
based on _all_ the tokens and the highness/lowness of each token's score, 
having that one token won't have much effect.  You should be fine.

Note also that bogofilter doesn't require any particular token be present, 
e.g. subj:SPAM, in order to score a message as spam.  It scores what _is_ 
present in the message.

David