Towards Multi-Agent-Simulation-Based Community Note Evaluation
Researchers have developed a new framework called MultiCom to address the challenges of timely and accurate community-based fact-checking on social media. This system utilizes a persona-guided multi-agent approach to simulate diverse rater populations and generate structured assessments of community notes. By clustering contributors and prompting agents with specific rating schemas, MultiCom produces explainable judgments, including confidence levels and reasons. An aggregation algorithm then combines these signals with raw votes to achieve reliable predictions, outperforming alternative methods with an average accuracy of 84.7% on a large dataset derived from X. AI
IMPACT This research could lead to more efficient and reliable automated fact-checking systems on social media platforms.