HESEIA: A community-based dataset for evaluating social biases in large language models, co-designed in real school settings in Latin America

2025-05-30

AI Summary

Existing social bias evaluation datasets suffer from insufficient involvement of impacted communities and thus fail to capture intersectional biases in localized, educational contexts. To address this, we propose an educator-led participatory co-construction paradigm: 370 teachers and 5,370 students from 189 primary and secondary schools across Latin America collaboratively designed 46,499 annotated sentences for bias assessment—spanning multiple school subjects, multidimensional demographic attributes (e.g., gender, ethnicity, socioeconomic status), and authentic pedagogical scenarios. Methodologically, we integrate minimal-pair construction, a fine-grained multidimensional labeling schema, and an education-context-driven data collection and validation pipeline. Empirical evaluation demonstrates that our dataset uncovers latent biases systematically missed by current large language models. The dataset is publicly released to enable community-driven, context-sensitive fairness auditing.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Ethics, Bias, and Fairness
• Philosophy and Ethics of AI: Bias, Fairness & Equity

Application Category

• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Most resources for evaluating social biases in Large Language Models are developed without co-design from the communities affected by these biases, and rarely involve participatory approaches. We introduce HESEIA, a dataset of 46,499 sentences created in a professional development course. The course involved 370 high-school teachers and 5,370 students from 189 Latin-American schools. Unlike existing benchmarks, HESEIA captures intersectional biases across multiple demographic axes and school subjects. It reflects local contexts through the lived experience and pedagogical expertise of educators. Teachers used minimal pairs to create sentences that express stereotypes relevant to their school subjects and communities. We show the dataset diversity in term of demographic axes represented and also in terms of the knowledge areas included. We demonstrate that the dataset contains more stereotypes unrecognized by current LLMs than previous datasets. HESEIA is available to support bias assessments grounded in educational communities.

Problem

• Evaluating social biases in LLMs with community co-design
• Capturing intersectional biases across demographic and school contexts
• Addressing stereotypes unrecognized by current LLMs in local settings

Innovation

• Community co-designed dataset for bias evaluation
• Intersectional biases across multiple demographic axes
• Minimal pairs method for stereotype identification