Date Posted: 8/06/2021

Justin Cheng, Ph.D. in Computer Science '17 and fellow researchers, including Cristian Danescu-Niculescu-Mizil, Associate Professor in the Department of Information Science in the Ann S. Bowers College of Computing and Information Science, have the most cited paper of the past five years from the ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW) entitled "Anyone Can Become a Troll: Causes of Trolling Behavior in Online Discussions."

At Cornell, Cheng studied with Jon Kleinberg, Tisch University Professor of Computer Science and Information Science, and Lillian Lee, Charles Roy Davis Professor of Computer Science and Information Science, and coauthored the paper "You Had Me at Hello: How Phrasing Affects Memorability" (2012).  

The research findings in "Anyone Can Become a Troll" has been covered widely in media, including:

  • PBS produced an animated educational video                                                
  • The Wall Street Journal: We're All Internet Trolls (Sometimes)
  • The Atlantic: Trolls Are Winning the Internet, Technologists Say
  • The Times: What does an internet troll look like? Just like you                                  
  • MIT Technology Review: There's a Troll Inside All of Us, Researchers Say

The authors, who also include Jure Leskovec and Michael Bernstein, both of Stanford University, spoke with The Conversation about this piece of research. Here are some excerpts from their remarks:

Many believe that trolling is done by a small, vocal minority of sociopathic individuals. This belief has been reinforced not only in the media, but also in past research on trolling, which focused on interviewing these individuals. Some studies even showed that trolls have predisposing personal and biological traits, such as sadism and a propensity to seek excessive stimulation.

But what if all trolls aren’t born trolls? What if they are ordinary people like you and me? In our research, we found that people can be influenced to troll others under the right circumstances in an online community. By analyzing 16 million comments made on CNN.com and conducting an online controlled experiment, we identified two key factors that can lead ordinary people to troll.

What makes a troll?

We recruited 667 participants through an online crowdsourcing platform and asked them to first take a quiz, then read an article and engage in discussion. Every participant saw the same article, but some were given a discussion that had started with comments by trolls, where others saw neutral comments instead. Here, trolling was defined using standard community guidelines – for example, name-calling, profanity, racism or harassment. The quiz given beforehand was also varied to either be easy or difficult. [...]

The first factor that seems to influence trolling is a person’s mood. In our experiment, people put into negative moods were much more likely to start trolling. We also discovered that trolling ebbs and flows with the time of day and day of week, in sync with natural human mood patterns. Trolling is most frequent late at night, and least frequent in the morning. Trolling also peaks on Monday, at the beginning of the work week.

Moreover, we discovered that a negative mood can persist beyond the events that brought about those feelings. Suppose that a person participates in a discussion where other people wrote troll comments. If that person goes on to participate in an unrelated discussion, they are more likely to troll in that discussion too.

In addition to the mood of the participant, there is also context to consider:

The second factor is the context of a discussion. If a discussion begins with a “troll comment,” then it is twice as likely to be trolled by other participants later on, compared to a discussion that does not start with a troll comment.

In fact, these troll comments can add up. The more troll comments in a discussion, the more likely that future participants will also troll the discussion. Altogether, these results show how the initial comments in a discussion set a strong, lasting precedent for later trolling.

Cheng, et al. conclude:

Trolling also can differ in severity, from swearing to targeted bullying, which necessitates different responses. 

It’s also important to differentiate the impact of a troll comment from the author’s intent: Did the troll mean to hurt others, or was he or she just trying to express a different viewpoint? This can help separate undesirable individuals from those who just need help communicating their ideas.

When online discussions break down, it’s not just sociopaths who are to blame. We are also at fault. Many “trolls” are just people like ourselves who are having a bad day. Understanding that we’re responsible for both the inspiring and depressing conversations we have online is key to having more productive online discussions.

Read more from The Conversation, and see also other collaborative work between Cheng and Cristian Danescu-Niculescu-Mizil, including: