Paper Number
ICIS2026-1864
Paper Type
Completed
Abstract
Social media platforms are increasingly replacing professional fact-checking with crowdchecking to combat misinformation. Focusing on X's Community Notes program, we empirically investigate whether political bias exists in crowdchecking and whether algorithmic aggregation can mitigate it. We find that notes on politically divisive tweets receive 2.7% fewer helpfulness ratings and have 6.6% lower odds of receiving raters' support. A randomized online experiment confirms that this disparity stems from individual-level political bias among raters, who discount notes correcting tweets aligned with their own ideology. These individual biases further aggregate into politically biased outcomes, where notes on divisive tweets are 35.2% less likely to become public. The bridge-based aggregation algorithm not only fails to mitigate individual political bias but also exhibits excessive conservatism. We propose a discernment-based algorithm with an entropy-guided rotation, which substantially increases the misinformation labeling rate while maintaining precision, greatly improving the overall effectiveness of crowdchecking.
Recommended Citation
Li, Adam Shenxiong and Gao, Yang, "Political Bias in Crowdchecking and an Algorithmic Remedy" (2026). ICIS 2026 Proceedings. 6.
https://aisel.aisnet.org/icis2026/platforms/platforms/6
Political Bias in Crowdchecking and an Algorithmic Remedy
Social media platforms are increasingly replacing professional fact-checking with crowdchecking to combat misinformation. Focusing on X's Community Notes program, we empirically investigate whether political bias exists in crowdchecking and whether algorithmic aggregation can mitigate it. We find that notes on politically divisive tweets receive 2.7% fewer helpfulness ratings and have 6.6% lower odds of receiving raters' support. A randomized online experiment confirms that this disparity stems from individual-level political bias among raters, who discount notes correcting tweets aligned with their own ideology. These individual biases further aggregate into politically biased outcomes, where notes on divisive tweets are 35.2% less likely to become public. The bridge-based aggregation algorithm not only fails to mitigate individual political bias but also exhibits excessive conservatism. We propose a discernment-based algorithm with an entropy-guided rotation, which substantially increases the misinformation labeling rate while maintaining precision, greatly improving the overall effectiveness of crowdchecking.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
17_Platforms_sharing_crowds