One can't also tell the difference between subterfuge trolling and plain up trolling. It would be nice to use machine learning in a multidimensional conversation scoring system so that tired hollow techniques at conversation manipulation would fade in to the background noise.