I suggested something in a conversation about here. Basically, should we just farm "hot words" from things that are flagged, then if they're in below .01 of questions just add them to smoke detector? It might make it a little less obvious and a little more false-positive. Making it just learn though would be nice.

  • 1
    Perhaps someone could program in the ability to respond to Smoke Detector with a certain phrase that he could then parse and store - be it a False Positive, spam, offensive, etc. Obviously, that would be quite time-consuming, seeing as you're writing a program to find patterns in scattered responses-but not totally impossible. It could compare body text & title for key words. Feb 18, 2015 at 13:40
  • 1
    @RPiAwesomeness We already have that, restricted to a certain set of trusted users. We're working on a Bayesian algorithm for this, but it's not really ready.
    – Undo
    Feb 18, 2015 at 17:54
  • @Undo Ah. Fair enough :) Feb 18, 2015 at 20:24


You must log in to answer this question.

Browse other questions tagged .