Moderator Perspectives on Encouraging Desirable Behavior
This article summarizes my paper exploring Reddit moderators’ perspectives on positive reinforcement. The paper will be presented at CSCW 2024.
In 2023, an estimated 5.7 billion people used social media worldwide. Research has shown that engaging with these platforms can affect users’ mental health, especially for those who are exposed to harmful or offensive content. The people who are tasked with protecting users from seeing such content are the moderators. Moderating online communities can be a huge burden that often falls to volunteers and because they are on the front lines of sifting through user contributions, moderator mental health is most at risk. The tools given to moderators often come in the form of punishment, such as removing offending content or banning the author of such content from participating in the future. While these methods have been shown to be effective at curbing the spread of bad behavior, carrying out these reactive, punitive actions inherently exposes moderators to potentially offensive and triggering content.
One way to address this issue is to establish better practices for encouraging users to acquire the norms of their communities. Moderators are also responsible for norm-setting in their communities. However, on platforms like Reddit, understanding community norms is a non-trivial task. Across the millions of sub-communities on Reddit, there are nearly as many distinct community norms that implicitly dictate what makes a successful contribution. These norms are often not stated explicitly, and those that are do not necessarily get followed.
HCI and behavioral psychology theory present us with alternative ways to proactively facilitate norm-setting in online communities. More specifically, positive reinforcement is a long-studied behavior modification principle that involves introducing a positive stimulus in response to desired behaviors to encourage those behaviors in the future. HCI theory similarly stresses the importance of highlighting positive examples of user contributions in helping enhance community health and encourage more desired behaviors.
Though positive reinforcement is effective at encouraging desired behaviors in offline settings such as education, workplaces, and parenting, there is a gap in our understanding of its role in the moderation of online spaces. Our paper explores whether positive reinforcement currently has a role in moderation alongside popular punitive approaches. Specifically, we conducted a survey of moderators focused on understanding whether positive reinforcement is currently practiced in moderation.
In our survey of 115 Reddit moderators, we find that more than 95% of our respondents are already taking some action to encourage desirable behavior they see in their communities (See Fig. 2). Specifically, we constructed a taxonomy with the responses to the following question:
“What kinds of content and behavior would you want to encourage as a moderator?”
This revealed that prosocial behavior is the most common attribute moderators want to encourage with specific emphasis on civility, politeness, and respectfulness. Other moderators valued contributions that are original, useful, and/or creative.
To accompany our first taxonomy of desirable behavior, we constructed another taxonomy based on moderator responses to the question:
“What do you do when you see content you want to encourage?”
Interestingly, the responses to this question revealed that more than half the participants respond to contributions they want to encourage through anonymous interface signals such as upvotes and awards. These methods, however, are available to all users (not just moderators) and recipients are unable to distinguish between feedback from moderators and feedback from their fellow users. This is key evidence that valuable moderator effort is not always recognizable as such. Only about 21% of respondents reported taking advantage of moderator-specific interface signals, such as pinning and flaring, to encourage desirable content. Participants in our survey also explained that they were motivated to provide positive feedback to contributions in order to encourage similar behavior from the authors, essentially positive reinforcement. Many other moderators simply wanted to highlight exemplary behavior, something HCI theory dictates can help users better learn community norms.
Despite the fact that positive reinforcement is clearly on the minds of Reddit moderators, consistently employing positive reinforcement techniques is not feasible for every moderation team. Moderators of larger subreddits mentioned having to manually inspect large volumes of content if they want to find content they would want to encourage. For those who are able to identify content deserving of rewards, the lack of tools to help them actually provide those rewards is discouraging. Additionally, the moderators who indicated not engaging with positive reinforcement whatsoever cited the lack of available tools to explain why they only use punitive moderation strategies.
Overall, this study highlights the need for research into the effectiveness of positive reinforcement as a moderation strategy. We also contribute taxonomies that can support the development of community-specific methods of detecting desirable behavior. Finally, our survey provides motivation for platforms to support moderators’ interest in engaging with positive reinforcement with explicit tools for locating and rewarding desirable content.
This paper will be presented at the 27th ACM Conference on Computer-Supported Cooperative Work and Social Computing.
Charlotte Lambert, Fred Choi, and Eshwar Chandrasekharan. 2024. “Positive reinforcement helps breed positive behavior”: Moderator Perspectives on Encouraging Desirable Behavior. In Proc. ACM Hum.-Comput. Interact. 8, CSCW2, Article 390 (November 2024), 33 pages. https://doi.org/10.1145/3686929 (currently available here)
