hi Marco -- there's lots of room to investigate. That blog post details the fundamentals of the JM_SOUGHT ruleset, which is "live" and in production and has been for several years -- so that works ;)
But association rules -- automatically-generated meta rules -- still have plenty of room for improvement. My attempts have largely resulted in a lot of self-reinforcement -- ie. they push existing spam determinations higher, rather than fix FPs or false negatives. this isn't particularly useful, so I stopped investigating that approach, but I think with a bit of work (and research into the state of the art) it should be possible to come up with something better. --j. On Wed, Oct 13, 2010 at 02:35, Marco Ribeiro <[email protected]> wrote: > Hello guys, > I've read some of the material on the wiki about the "RulesProjectPlan". > Taking a quick look at the proposed Solutions, I don't see any automatic or > semi-automatic approach. Is anyone working on such a solution? I was > directed to this blog post , but it's from 2007. I've thought before of > trying to use association rules ,as they are highly readable and can be > filtered on confidence to avoid false positives. > Thanks, > Marco Ribeiro
