Research coauthored by Info Sci Assistant Professor Cristian Danescu-Niculescu-Mizil on internet trolling is part of a recent BBC feature exploring efforts to curb online abuse.
Highlighting a good mix of past studies and initiatives to cut down on trolling, the BBC looks to Danescu-Niculescu-Mizil who explains how a troll’s online behavior – not just the choice of words, but length of posts in a limited number of threads, for instance – is a fairly accurate indicator of whether or not a verbally abusive user will eventually get the boot from a particular website or forum.
“Anti-social users tend to focus on a few threads," Danescu-Niculescu-Mizil tells the BBC. "They write a lot but only on a small number of threads instead of spreading out like other users would."
By examining 40 million online comments on cnn.com, breitbart.com, and ign.com, the Stanford and Cornell research team created an algorithm to help identify anti-social behavior. Findings were published in their paper, "Antisocial Behavior in Online Discussion Communities".
“It turns out that there’s enough information in the first five or 10 posts to actually predict whether they’re going to be banned in the future with an accuracy of about 80%,” Danescu-Niculescu-Mizil tells the BBC.
The hope, he adds, is to help human moderators identify potential trolls early on so as to prevent online comments from becoming increasingly abusive or threatening.