[Asrg] differential confidence

[J.D. Falk]

On 05/12/2008 02:11, "Chris Lewis" <clewis at nortel.com> wrote:

> I could replace $prod with $traffic in all the denominators, but you
> can't get content or complaint hits on the traps.  So that just dilutes
> the score for very high trap hits.
> 
> Remember, I'm not going for pedantic purity.  I'm going for a function
> that has lots of head room and catches the worst offenders.

Yep, and that's all anyone can do, given the ever-changing nature of spam.

Other reputation systems I've worked on or learned about also mix in other
factors, such as:

- "not spam" votes (which aren't weighted 1:1 with spam votes)
- reporter reputation (not all users' votes get the same weight)
- past behavior of the IP (change thresholds based on recent behavior)
- type of mailstream (different thresholds for gmail vs. sears)