Brazilian Portuguese Product Reviews Dataset for Gender Identification

This dataset contains 4,000 product reviews written in Brazilian Portuguese (PT-BR) collected from Amazon Brazil across ten product categories: automobile, baby, toy, cellphone, food, game, laptop, book, fashion, and pets. Each review was manually annotated with a binary gender label (male or female) inferred from the author's public profile name and contextual information available on the platform. The annotation process involved two independent annotators, and cases with disagreement were removed to ensure higher label reliability. User names were anonymized to preserve privacy. The dataset is balanced by construction, containing 40,000 reviews, with 20,000 labeled as female and 20,000 labeled as male, evenly distributed across the ten product categories. This balanced design aims to support fair comparisons between machine learning models and avoid distortions caused by class imbalance. This resource was created to support research on: Author profiling Gender identification from text Natural Language Processing for Portuguese Evaluation of large language models (LLMs) Cross-domain text classification The dataset was used in the study: Gender Identification in Brazilian Portuguese Product Reviews: A Comparative Study of Classical Models, mBERT, and LLMs. In that study, ten models were evaluated, including classical machine learning algorithms, BERT-based models, and modern large language models such as ChatGPT and DeepSeek, showing that multilingual transformer models can achieve competitive performance in gender prediction tasks using only textual information. Dataset Statistics Total reviews: 40,000 Languages: Brazilian Portuguese Categories: 10 Reviews per gender:Female: 20,000 Male: 20,000 Reviews per category:4,000 reviews per category