I understand that bayesian poison at the end of a spam will eventually
be recognised and further spams rejected.

However, there seems to me a good way of speeding up the process - to :

1. Delete the spam poision from the end of a spam when doing the Bayesian 
   aggregation (OK, manual entry)

2. Change the bayesian analyser to operate on a "line by line" basis. This
   means that there will be a "spike" at the last line of the spam, which 
   will "dilute" as it goes into the bayesian poison. However, the "spike"
   will be prominent enough that its spam status will be clear.

Yes, this involves manual manipulation, but I would find the results - less
spam sooner - worth the manual intervention.

I'm willing to poke around with implementing this approach if nobody else is
willing, with some suggestions on how the bayesian analyser would be changed.
I don't want to start from a cold start.

I think I did post on this a year or so ago, but my appreciation is a bit more
subtle and I' again fired up about it.

Thanks,

-- 
John August, convenor The Sydney Shove - www.sydneyshove.org

Reply via email to