On 2/19/2011 6:43 AM, John Hardin wrote:
On Sat, 19 Feb 2011, Justin Mason wrote:
On Friday, February 18, 2011, Warren Togami Jr. <[email protected]>
wrote:
Is there any way our corpora can be part of SOUGHT's safety net
without giving up our privacy?
Unfortunately not --- the generation process makes the entire mail
contents fully visible.
Perhaps yes, though. The generation process could check the masscheck
results of the SOUGHT subrules and permanently suppress any subrules
that hit (a certain threshold of) ham.
It's feedback, just not as immediate as the dedicated SOUGHT ham corpus,
and it doesn't require exposing the masscheck ham corpora. I think
that's what Warren had in mind.
Aren't the nightly masscheck subrules combinations of many patterns, so
you aren't sure exactly which pattern is bad?
What I'm suggesting is a mini-masscheck for SOUGHT only. Put each
pattern into thousands of individual sub-rules. Participants run this
mini-masscheck against their ham only, then upload the logs directly to JM.
Warren
