[Paper Review] Anyone Can Become a Troll: Causes of Trolling Behavior in Online Discussions

Justin Cheng, Michael S. Bernstein+2 more|arXiv (Cornell University)|Feb 3, 2017

Social Media and Politics108 citations

TL;DR

This paper shows that negative mood and exposure to prior troll posts increase trolling propensity, using a controlled online experiment and a large-scale CNN.com dataset, and presents a predictive model (AUC 0.78) suggesting trolling is largely situational rather than innate.

ABSTRACT

In online communities, antisocial behavior such as trolling disrupts constructive discussion. While prior work suggests that trolling behavior is confined to a vocal and antisocial minority, we demonstrate that ordinary people can engage in such behavior as well. We propose two primary trigger mechanisms: the individual's mood, and the surrounding context of a discussion (e.g., exposure to prior trolling behavior). Through an experiment simulating an online discussion, we find that both negative mood and seeing troll posts by others significantly increases the probability of a user trolling, and together double this probability. To support and extend these results, we study how these same mechanisms play out in the wild via a data-driven, longitudinal analysis of a large online news discussion community. This analysis reveals temporal mood effects, and explores long range patterns of repeated exposure to trolling. A predictive model of trolling behavior shows that mood and discussion context together can explain trolling behavior better than an individual's history of trolling. These results combine to suggest that ordinary people can, under the right circumstances, behave like trolls.

Motivation & Objective

• Motivate understanding of why ordinary users engage in trolling in online discussions.
• Identify two trigger mechanisms—user mood and discussion context—that influence trolling propensity.
• Experimentally establish causal effects of mood and context on trolling behavior.
• Validate findings with large-scale observational data from a major news discussion site.
• Develop a predictive model to assess trolling likelihood based on mood and context.

Proposed method

• Conduct a two-by-two online experiment varying participant mood (positive vs negative) and discussion context (initial troll vs benign posts).
• Measure trolling behavior and sentiment using expert labeling and LIWC-based negative affect.
• Use a randomized, multi-universe design to control for path dependence and ensure independent discussion environments.
• Perform a mixed effects logistic regression to assess the effects of NegMood and NegContext on trolling likelihood.
• Augment experimental results with a large-scale, longitudinal analysis of CNN.com discussions (16.5M posts) to study mood proxies and exposure effects.
• Build and evaluate a logistic regression model to predict trolling (AUC = 0.78) using mood and discussion context as predictors.

Experimental results

Research questions

• RQ1 Can negative mood increase the likelihood of a user posting trolling content?
• RQ2 Does exposure to prior troll posts in a discussion context elevate trolling propensity?
• RQ3 Do mood and discussion context interact to influence trolling behavior?
• RQ4 Is trolling contagious, spreading across users and discussions in real-world data?
• RQ5 How well can mood and context predict trolling behavior compared to a user’s history of trolling?

Key findings

• Negative mood increases the odds of trolling by 89%.
• Presence of prior troll posts in the discussion increases the odds of trolling by 68%.
• The combination of negative mood and negative context doubles baseline trolling rates in the experiment.
• In CNN.com data, 1 in 4 flagged posts come from users with no prior flag history, indicating ordinary users contribute to trolling.
• Trolling propensity tracks population mood fluctuations over time and shows persistence and decay patterns across discussions.
• A logistic regression model using mood and discussion context achieves AUC = 0.78 in predicting trolling behavior.