Detecting Hate Speech in Portuguese Social Media: A Machine Learning Framework for Safer Online Communities

This study explores the development of a machine learning (ML) model for detecting offensive language in Portuguese — particularly misogyny, hate speech, and sexism — in social media content. Named Curupira after the Brazilian folklore guardian of nature, the project symbolically aligns with its goal of protecting digital spaces from harmful discourse. Using Natural Language Processing (NLP) techniques and a Python-based Jupyter Notebook prototype, the model analyzes textual patterns, contextual ambiguities, and linguistic nuances specific to Portuguese. While the current focus is on identification (not yet blocking), the long-term objective is to evolve into an AI-driven moderation system. Preliminary results highlight the challenges and potential of NLP in addressing regional language variations, offering insights for future improvements in accuracy and scalability. This work contributes to the broader discussion on AI-assisted tools for fostering safer online environments.