>I'm wondering if a simple Bayesian filter might do the trick for the vast majority of Internet trollery. Not actually possible because "trolling" so often isn

>I'm wondering if a simple Bayesian filter might do the trick for the vast majority of Internet trollery. Not actually possible because "trolling" so often isn't in the actual words as it is the context. For example, presenting an intentionally poorly-reasoned argument designed to infuriate people. Or for instance, it is possible to appropriate the rhetoric of "social justice" and construct a troll that is indistinguishable from a legitimate statement, the only difference is the context, e.g. used by an oppressor rather than the oppressed. Finally, in many communities it is acceptable as part of the in-group to flame the out-group, but not vice-versa. One person's "eat shit and die" is acceptable, the other person's is not. I suppose this could be mitigated by training your own bayesian filter based on your ingroup and your outgroup separately, based on who you designate your ingroup via social network or something. Sort of like "if I flame an outgroup, I won't be punished for doing so by people who consider me part of their ingroup."