HESEIA: A Community-Based Dataset for Evaluating Social Biases in Large Language Models, Co-Designed in Real School Settings in Latin America
Published in Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP 2025), 2025
Most resources for evaluating social bias in LLMs lack community co-design. HESEIA is a community-sourced dataset developed through participatory workshops in secondary schools in Argentina. We detail the participatory methodology, analyze the identified bias dimensions, and demonstrate how current models reproduce these harms.