By Adrienne LaFrance @ The Atlantic, March 29

[.....] Researchers have already used technology to begin to understand what they’re up against. Earlier this month, a team of computer scientists from Stanford University and Cornell University wrote about how they used machine-learning algorithms to forecast whether a person was likely to start trolling. Using their algorithm to analyze a person’s mood and the context of the discussion they were in, the researchers got it right 80 percent of the time. 

They learned that being in a bad mood makes a person more likely to troll, and that trolling is most frequent late at night (and least frequent in the morning). They also tracked the propensity for trolling behavior to spread. When the first comment in a thread is written by a troll—a nebulous term, but let’s go with it—then it’s twice as likely that additional trolls will chime in compared with a conversation that’s not led by a troll to start, the researchers found. On top of that, the more troll comments there are in a discussion, the more likely it is that participants will start trolling in other, unrelated threads [....]