PAVE-BR Golden Dataset is a manually curated benchmark dataset for Product Attribute Value Extraction (PAVE) in the context of Brazilian e-commerce. It is designed to support the evaluation of Large Language Model (LLM)-based extraction systems for Portuguese, and serves as a foundation for the AI-PAVE-Br project. This dataset aims to foster reproducible research in Natural Language Processing (NLP), with a particular focus on product classification and attribute extraction.

The PAVE-BR Golden Dataset serves as a gold-standard benchmark for evaluating the accuracy and performance of automatic systems in product classification and attribute extraction. By comparing model outputs against expert-annotated ground truth labels, researchers can assess the effectiveness of their approaches using consistent and objective criteria. The dataset has also been used to benchmark the proposed AI-PAVE-Br model against traditional and LLM-based baselines.

The dataset is available in two formats:

• Excel format (Dataset_in_Excel)

  • Each sheet corresponds to a specific product category.
  • May include visual annotations (e.g., color coding) to support review and validation.

• CSV format (Dataset_in_CSVs)

  • Each product category is represented by a separate CSV file.
  • This structure avoids excessive sparsity, as each category has a distinct attribute schema that would otherwise result in large numbers of missing values in a unified format.

All annotations were produced manually by a trained annotation team, with close attention to the specific linguistic and contextual nuances of Brazilian e-commerce. The dataset supports evaluation for both product classification (entity-level) and attribute extraction tasks.

The dataset includes representative and high-impact product types from Brazilian e-commerce, such as:

Air Conditioner, TV, Cell Phone, Refrigerator, Notebook, Tire, Wardrobe, Bed, Sneaker, Stove, Table and Chair Set, Backpack, Faucet, Headphone, Perfume, Doll, Motorcyclist Helmet, Pot, Lamp, Cell Phone Case.

Each annotated product includes:

• Entity (Tipo de Produto): The most specific product type (e.g., 'Ar Condicionado', 'Perfume').
• Category (Categoria): A broader classification group (e.g., 'AR' for air conditioners).
• Subcategories (Subcategoria): A list of relevant domain-specific tags or subcategories (e.g., ['ARCA', 'ACIV', 'ARAR'] for an air conditioner).

The annotation schema was defined in advance, and the list of attributes per product type was determined based on domain expertise and platform conventions.

Refer to Table 1. Product Entities and Their Associated Attribute Lists in the accompanying paper for a complete specification of attributes per product type.

To ensure statistical robustness and reduce selection bias, products were sampled randomly from each product type, considering their platform-level classification. Sample sizes were determined using Cochran’s formula for large populations, targeting a 95% confidence level with a 5% margin of error. This led to approximately 385 annotated items per product type, balancing statistical validity with the practical limitations of manual annotation effort.

This dataset is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.

You are free to:

• Share — copy and redistribute the material in any medium or format.
• Adapt — remix, transform, and build upon the material.

Under the following terms:

• Attribution (BY) — You must give appropriate credit to Magazine Luiza, provide a link to the license, and indicate if changes were made.
• NonCommercial (NC) — You may not use the material for commercial purposes.
• ShareAlike (SA) — If you remix, transform, or build upon the material, you must distribute your contributions under the same license.

For the full legal terms, please refer to the LICENSE file.

If you use the PAVE-BR Golden Dataset in your research, please cite it as follows (or use the CITATION.cff file included in the repository):

Gazzola, M., Souto, H. G., Silva, S., Peixoto, J. S., Siqueira, F., Morais, A. L. P., & Gomes, C. (2025). AI-PAVE-Br: Leveraging Large Language Models for Enhanced Product Attribute Value Extraction through a Golden Set Approach. In Proceedings of the Symposium in Information and Human Language Technology (STIL 2025).

@inproceedings{gazzola2025aipavebr,
  author    = {Murilo Gazzola and Hugo Gobato Souto and Samuel Silva and Júlia Schubert Peixoto and Felipe Siqueira and André Luis Pedroso de Morais and Caio Gomes},
  title     = {AI-PAVE-Br: Leveraging Large Language Models for Enhanced Product Attribute Value Extraction through a Golden Set Approach},
  booktitle = {Proceedings of the Symposium in Information and Human Language Technology (STIL)},
  year      = {2025},
  url       = {https://github.com/ai-luizalabs/AI-PAVE-Br}
}