Antisocial Behavior in Online Discussion Communities

Abstract: User contributions in the form of posts, comments, and votes are essential to the success of online communities. However, allowing user participation also invites undesirable behavior such as trolling. In this paper, we characterize antisocial behavior in three large online discussion communities by analyzing users who were banned from these communities. We find that such users tend to concentrate their efforts in a small number of threads, are more likely to post irrelevantly, and are more successful at garnering responses from other users. Studying the evolution of these users from the moment they join a community up to when they get banned, we find that not only do they write worse than other users over time, but they also become increasingly less tolerated by the community. Further, we discover that antisocial behavior is exacerbated when community feedback is overly harsh. Our analysis also reveals distinct groups of users with different levels of antisocial behavior that can change over time. We use these insights to identify antisocial users early on, a task of high practical importance to community maintainers.

• The paper develops a predictive framework using post content analysis and longitudinal metrics to achieve over 80% AUC for early detection of antisocial users.
• It employs quantitative measures such as readability and text similarity to differentiate Future-Banned Users from Never-Banned Users with distinct behavioral trajectories.
• The study indicates that early heavy censorship accelerates user behavior degradation, providing actionable insights for improved moderation policies.

Antisocial Behavior in Online Discussion Communities

This paper explores the identification and characterization of antisocial behavior in online discussion forums, using large-scale, longitudinal analyses. By examining behavioral data from several prominent online communities, the analysis characterizes users who exhibit antisocial behaviors and consequently are banned. The study provides quantitative insights enabling early identification of such users, which holds significant implications for maintaining the health of online communities.

Characterizing Antisocial Behavior

The research provides a detailed analysis of antisocial behavior based on user-generated content on platforms such as CNN.com, Breitbart.com, and IGN.com. Two primary user groups are identified: Future-Banned Users (FBUs) and Never-Banned Users (NBUs). FBUs are characterized by their propensity to post less readable content, often using inflammatory language and being less integrated into existing discussion threads. These users receive more replies, indicating successful engagment in discussions, albeit in a likely negative manner.

Quantitative metrics such as readability scores and text similarity measures are employed to differentiate FBUs from NBUs. These analysis methods help ascertain the off-topic nature and inflammatory tendency of FBUs’ contributions.

Evolution of Antisocial Behavior

The study examines how FBUs’ behavior degrades over their tenure in a community. Using metrics like post deletion rates and readability, it is evident that FBUs not only start out with lower quality contributions but further degrade over time. This degradation is exacerbated by community responses, wherein increased post deletion rates over time suggest reduced community tolerance towards these users.

A crucial insight is the potential role of excessive censorship in accelerating the decline of user behavior. Statistical matching techniques reveal that users who experience heavier censorship early in their online presence are more likely to worsen over time.

Typology of Antisocial Users

A key finding of the study is the categorization of antisocial users into two distinct classes: those with consistently high deletion rates (Hi-FBUs) and those with initially low but increasing deletion rates (Lo-FBUs). Hi-FBUs exhibit a higher propensity for antisocial behavior early in their tenure and receive quick and consistent punitive actions from moderators. On the other hand, Lo-FBUs demonstrate a gradual increase in undesirable behavior, indicating a potential window for intervention before their behavior escalates substantially.

A two-phase model of behavior is proposed, which examines changes in post deletion rates over time. This model highlights user subpopulations that exhibit varying trajectories of behavioral degradation or improvement, even among those not ultimately banned, emphasizing the complex dynamics of antisocial behavior evolution.

Predictive Modeling for Early Detection

A significant practical contribution of the paper is the development of a predictive framework capable of identifying potential FBUs early in their engagement history. The framework employs features drawn from post content, user activity, community responses, and moderator actions, achieving a prediction accuracy exceeding 80% AUC with only the first few posts.

The paper highlights the robustness of the model, capable of generalizing across different community contexts, demonstrating its potential for scalable deployment in diverse online environments. Such early detection systems could dramatically reduce the resource burden on human moderators and help preserve the community’s overall health.

Conclusion

This study provides a comprehensive examination of antisocial behavior in online communities, presenting critical insights into the behavioral dynamics and strategies for early detection. The findings have significant implications for designing moderation policies and ensuring a healthy digital discourse environment. Future research directions could focus on refining textual analysis for better behavioral segmentation and exploring real-time intervention strategies for identified users. By understanding and curbing antisocial behavior, online platforms can foster more inclusive and constructive communities.