Título do projeto: Automatic identification of bias in Large Language Models

Nível: Mestrado

Resumo: Large Language Models (LLMs) have demonstrated remarkable capabilities across various domains, from legal reasoning to clinical decision support. As these models become increasingly integrated into real-world applications, concerns about their reliability, fairness, and ethical implications have emerged. Studies have shown that LLMs can produce biased outputs, reinforcing harmful stereotypes and discriminating against marginalized groups. This work proposes a systematic and scalable framework for evaluating and ranking LLMs based on stereotype generation in Brazilian Portuguese. The framework combines template-based sentence generation, human annotation, and supervised classification into a unified pipeline. A set of 164 sentence templates, covering gender, race, and their intersections, was used to elicit completions from 37 LLMs from multiple providers. The resulting sentences were annotated by human annotators along two dimensions: alignment with social stereotypes and potential harm. The stereotype alignment labels served as the foundation for training a BERTimbau-based classifier, selected via nested cross-validation, which achieved a macro averaged F1 of 0.665. Classifier predictions were then used to construct pairwise match tables, feeding to an Elo rating system, that generated two complementary rankings: a model ranking and a social marker ranking. The results reveal that smaller open-source models tend to generate less stereotyped content than larger commercial ones, and that social markers combining race and gender consistently elicit the most stereotyped outputs across all models. The framework is made available as an interactive interface that supports the incremental addition of new models.

Orientadora: Profa. Dra. Helena de Medeiros Caseli

Vídeo da defesa 

Publicações decorrentes deste trabalho:

• Fernanda Malheiros Assi and Helena Caseli. 2024. Biases in GPT-3.5 Turbo model: a case study regarding gender and language. In Proceedings of the 15th Brazilian Symposium in Information and Human Language Technology, pages 32–43, Belém do Pará, Brazil. Association for Computational Linguistics.

• ASSI, Fernanda Malheiros. Automatic identification of bias in Large Language Models. 2026. Dissertação (Mestrado em Ciência da Computação) – Universidade Federal de São Carlos, Campus São Carlos, 2026. Disponível em: https://repositorio.ufscar.br/handle/20.500.14289/24224.

Título do projeto: Inferência automática da carga afetiva em postagens de redes sociais

Nível: Iniciação Científica

Breve descrição: Este projeto visa criar um modelo computacional capaz de medir a carga afetiva através do tempo, em postagens de redes sociais, atreladas a um tópico ou evento específico por meio de séries temporais de domínios emocionais. A valência e a excitação média estimadas a partir de postagens realizadas em uma mesma janela curta de tempo representarão uma observação dessa série temporal, que descreve a variação da carga afetiva no espaço de Scherer. Para a previsão de carga afetiva, pretende-se utilizar como base a arquitetura do modelo LSTM e células GRU.

Agência financiadora: 

• FAPESP (2021/07067-0)

Orientadora: Profa. Dra. Helena de Medeiros Caseli

Publicação decorrente deste projeto:

• ASSI, F. M.; CANDIDO, G. B.; SILVA, L. N. S.; SILVA, D. F.; CASELI, H. M. UFSCar’s Team at ABSAPT 2022: Using Syntax, Semantics and Context for Solving the Tasks