UC Davis

Online deviant behaviors in social media have become one of the major concerns of the public, with its prevalence and rapid and widely spread. Though massive research has been done to understand the antecedents and consequences of online deviant behaviors, academic scholarship is still far from a definitive explanation about how the dynamic of the behavior is embedded in its context. In this dissertation, for two forms of online deviant behaviors, incivility and trolling, three studies were conducted to a) examine the overall dynamic of incivility on the macro level, b) understand the mechanisms of trolls and victims on the meso level, and c) identify the underlying mathematical architecture of trolling behavior on the micro level. Study 1 provides a descriptive account of the incivility dynamic over the past eleven years by examining the trends of incivility in three major categories: political, non-political, and mixed. Using longitudinal data from Reddit that accounts for 95% of the entire Reddit universe across eleven years and relying on the combination of supervised machine learning models and traditional statistical inference, the study found that incivility consistently represents around 10% of total Reddit comments, with fluctuations that correspond to offline socio-political events and platform-specific policies. We also found that political groups tend to be more uncivil, and discussions in mixed groups that are not overtly political but nevertheless discuss politics are less uncivil than that in political groups. Study 2 investigated the aftermath of trolling on community dynamics by examining the likelihood and conditions in which individual users react toward trolls. Using a longitudinal behavioral dataset collected from popular video communities on YouTube, the study found that the valence of the trolling message, characteristics of the individual member, as well as the patterns of past engagement with trolls from other community members all played a role in predicting how an individual would react to trolls. In other words, well-connected users situated in densely connected communities with a prior pattern of engaging trolls are more likely to respond to trolls, especially when the trolling messages convey negative sentiments. Study 3 employed an information theoretical approach to deconstruct trolling behavior as a dynamic process to understand how people engage in such behavior. Using longitudinal data from Reddit’s active users and applied stochastic process and statistical inference, the study found that individuals engaging in trolling behaviors is a complex process that involves both internal influences and external influences from others they interact with. In addition, the hidden pattern of mathematical architecture shows a contagion effect of the behavior. Overall, the three studies demonstrated that online deviant behavior is nested in individual processing, community norms, and platform contexts.