CAREER: Models and Methods to Help Online Communities Support Behaviors Aligned with Community Norms

About This Grant

This project aims to develop new methods to support community-determined behavior norms in online communities consistent with the goal of helping communities reward behaviors that serve the community's goals. Current approaches to online community management typically focus on detecting and discouraging undesirable behavior with sanctions like bans, quarantines, and content removals. These punitive approaches do not directly foster an environment for community members to engage in positive ways around the group's norms. This project will look at ways to use positive reinforcement to proactively incentivize community-determined behavior. The work will advance knowledge about how communities recognize and encourage behavior while developing novel technologies to promote community norms in online spaces. The education and outreach plan is deeply tied to the research activities, focusing on scaling up the broader impacts of the research. A public application programming interface will enable developers and moderators around the world to integrate the computational approaches developed by this research into their own communities. The project will be carried out in three phases. Phase 1 will involve the development of new computational models to detect community-determined behavior across different types of online communities. This phase will uncover the structure of encouraged behavior within online conversations and develop community- and context-sensitive computational models to identify behavior. Phase 2 will employ causal inference frameworks to evaluate the efficacy of current feedback mechanisms in terms of their effects on positive reinforcement and promoting desirable behavior. This phase will draw on theories from human-computer interaction and psychology to examine the causal effects of various forms of positive feedback in motivating users and reinforcing normative behavior. Phase 3 will involve the building of a new AI-backed system that will allow moderators to easily identify instances of valuable, normative behavior within their communities and respond with timely feedback that serves as positive reinforcement. Easy access to pre-trained models and carefully designed human-centered tools developed as part of this research will allow moderators to incorporate high quality inputs suggested by state-of-the-art AI methods into their workflows to help achieve their community's goals. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.