Are you sure? Measuring models bias in content moderation through uncertainty

A recent study highlights the ongoing challenges in measuring bias in content moderation systems that use language models. While these models are essential for maintaining safety on social media platforms, they often reflect and perpetuate existing racial and social biases. Despite the development of various resources aimed at addressing these issues, the fairness of these models remains a significant concern. This research introduces an unsupervised method to better assess model bias, which is crucial for improving content moderation practices and ensuring equitable treatment across diverse user groups.