I'm looking at creating an email address to capture spam and only be used for spam. Since I can be guaranteed that all email I received to this address is spam, is there a way to weigh this higher in the sa-learn?
No. Although exhaustive or weighted training are both interesting in some situations, SA currently uses a pure single-weight training method.
Really the only reason you should want to weigh them higher is if you're training less spam than ham, something which is very uncommon.
"middle of the road" tokens that exist equally in spam and ham will be biased by your training ratios. The use of Chi-squared combining covers up most problems from this, but if your effective training ratio gets way off (ie: greater than 1:20) then you might start having some FP problems on ham mail as common words start weighing in with a high probability of spam..
