Elamur Research⌕
☰
home / papers / 2026 / Gender Identification in Brazilian Portuguese Product Reviews: A Comparative Study of Classical Models, BERT, and LLMs

Gender Identification in Brazilian Portuguese Product Reviews: A Comparative Study of Classical Models, BERT, and LLMs

Paper · 2026 · Tiago de Melo and Carlos M. S. Figueiredo

Summary

The paper addresses inferring a review author's gender from the text alone in Brazilian Portuguese. Profiles on e-commerce platforms are often anonymous, and findings from English author-profiling research may not carry over because Portuguese has its own gender morphology. The authors built a new corpus of Amazon Brazil reviews (2021–2024) covering ten product categories. Gender labels came from reviewer names: two annotators judged each profile independently, and nicknames, unisex names and disputed cases were discarded.

They then benchmarked ten systems:

  • three TF-IDF-based classical classifiers;
  • mBERT and BERTimbau, used without fine-tuning;
  • five LLMs queried zero-shot through their APIs.

mBERT scored highest, but ChatGPT 4o and Logistic Regression trailed it by less than one point. This suggests the task remains difficult for every model family. Reviews written by women were classified more reliably than those by men, and difficulty depended on the product domain. Very short, generic reviews caused most errors. The released dataset fills a resource gap for PT-BR author profiling, and the authors also discuss the ethical risks of binary gender inference.

Key points

  • Corpus: 4,000 reviews (2,000 female, 2,000 male), with an equal share drawn from each of ten product areas: pets, fashion, books, laptops, games, food, cell phones, toys, baby products and automobiles. It is published on Zenodo (DOI 10.5281/zenodo.18856652).
  • Models and setup:
    • LR, RF and SVM used TF-IDF unigrams and bigrams capped at 3,000 terms, tuned with GridSearchCV (3-fold, macro-F1).
    • mBERT and BERTimbau were not fine-tuned.
    • ChatGPT 4o, ChatGPT 3.5, DeepSeek, Sabia3 and Sabiazinho ran zero-shot at temperature 0.
  • Best result: mBERT reached macro-F1 = 0.634. It edged out ChatGPT 4o and Logistic Regression (by under 1 point) and ranked above the Portuguese-specific BERTimbau.
  • By gender: F1 was 0.654 for female-authored texts versus 0.614 for male-authored ones. The authors credit richer morphological markers and affective vocabulary in women's reviews.
  • By domain and errors:
    • Book, automobile, laptop and game reached F1 ≥ 0.65, while baby and pets fell below 0.60.
    • Recurrent mBERT errors involved very short reviews (about 13–16 words on average in the hardest categories) and generic phrases such as "gostei muito".

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Tiago de Melo and Carlos M. S. Figueiredo. 2026. Gender Identification in Brazilian Portuguese Product Reviews: A Comparative Study of Classical Models, BERT, and LLMs. Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{melo2026gender,
  title = {{Gender Identification in Brazilian Portuguese Product Reviews: A Comparative Study of Classical Models, BERT, and LLMs}},
  author = {Tiago {de Melo} and Carlos M. S. Figueiredo},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-1.2/}
}

References

28 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Basem H Ahmed and Ayman S Ghabayen. 2022. Review rating prediction framework using deep learning. Journal of Ambient Intelligence and Humanized Computing, 13(7):3423–3432. find
  2. José Almeida Neto and Tiago de Melo. 2023. Exploring supervised learning models for multi-label text classification in brazilian restaurant reviews. Anais do Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), pages 126–140. find
  3. Leo Breiman. 2001. Random forests. Machine learning, 45:5–32. find
  4. Bassem Bsir and Mounir Zrigui. 2018. Bidirectional lstm for author gender identification. In International Conference on Computational Collective Intelligence, pages 393–402. Springer. find
  5. Alberto F Cabrera and 1 others. 1994. Logistic regression analysis in higher education: An applied perspective. Higher education: Handbook of theory and research, 10:225–256. find
  6. Corinna Cortes and Vladimir Vapnik. 1995. Support vector machine. Machine learning, 20(3):273–297. find
  7. Washington Cunha, Leonardo Rocha, and Marcos André Gonçalves. 2025. A thorough benchmark of automatic text classification: From traditional approaches to large language models. arXiv preprint arXiv:2504.01930. — [arXiv:2504.01930](https://arxiv.org/abs/2504.01930) find
  8. Paula Luegi, Márcio Leitão, Daniela Avila-Varela, Jéssica Gomes, and Armanda Costa. 2024. Reflexive pronoun resolution in portuguese: testing similarity-based interference. Frontiers in Language Sciences, 3:1473948. find
  9. Miguel de Oliveira and Tiago de Melo. 2021. An empirical study of text features for identifying subjective sentences in portuguese. In Intelligent Systems: 10th Brazilian Conference, BRACIS 2021, Virtual Event, November 29–December 3, 2021, Proceedings, Part II 10, pages 374–388. Springer. find
  10. Abdul Majeed and Sungchang Lee. 2021. Anonymization techniques for privacy preserving data publishing: A comprehensive survey. IEEE Access, 9:8512– 8545. find
  11. Oluwakemi Onikoyi and Amina Adamu. 2023. Improving gender classification in low-resource hausa with profile metadata. In AfriNLP. find
  12. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies,…
  13. Ramon Pires, Hugo Abonizio, Thales Sales Almeida, and Rodrigo Nogueira. 2023. Sabiá: Portuguese large language models. In Intelligent Systems, pages 226–240, Cham. Springer Nature Switzerland. find
  14. Francisco Rangel, Anastasia Giachanou, Bilal Hisham Hasan Ghanem, and Paolo Rosso. 2020. Overview of the 8th author profiling task at pan 2020: Profiling fake news spreaders on twitter. In CEUR workshop proceedings, volume 2696, pages 1–18. Sun SITE Central Europe. find
  15. Rafael Dias and Ivandré Paraboni. 2020. Cross-domain author gender classification in brazilian portuguese. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 1227–1234. find
  16. José Antonio García-Díaz, Ghassan Beydoun, and Rafel Valencia-García. 2024. Evaluating transformers and linguistic features integration for author profiling tasks in spanish. Data & Knowledge Engineering, 151:102307. find
  17. Fábio Souza, Rodrigo Nogueira, and Roberto Lotufo. 2019. Bertimbau: Pretrained bert models for brazilian portuguese. In ICAI. find
  18. Chanchal Suman, Rohit Shyamkant Chaudhary, Sriparna Saha, and Pushpak Bhattacharyya. 2022. An attention based multi-modal gender identification system for social media users. Multimedia Tools and Applications, 81(19):27033–27055. find
  19. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, and 1 others. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  20. Pradeep Vashisth and Kevin Meehan. 2020. Gender classification using twitter text data. In 2020 31st Irish Signals and Systems Conference (ISSC), pages 1–6. find
  21. Sanja Hanić, Marina Bagić Babac, Gordan Gledec, and Marko Horvat. 2024. Comparing machine learning models for sentiment analysis and rating prediction of vegan and vegetarian restaurant reviews. Computers, 13(10):248. find
  22. Claudimar Pereira da Veiga, Cássia Rita Pereira da Veiga, Júlia de Souza Silva Michel, Leandro Ferreira Di Iorio, and Zhaohui Su. 2024. E-commerce in brazil: An in-depth analysis of digital growth and strategic approaches for online retail. Journal of Theoretical and Applied Electronic Commerce Rese… find
  23. Supanat Jintawatsakoon and Ekkapob Poonsawat. 2023. Gender classification of social network text using natural language processing and machine learning approaches. In 2023 20th International Joint Conference on Computer Science and Software Engineering (JCSSE), pages 415–420. IEEE. find
  24. Shikha Verma. 2019. Weapons of math destruction: how big data increases inequality and threatens democracy. Vikalpa, 44(2):97–98. find
  25. Katikapalli Subramanyam Kalyan. 2024. A survey of gpt-3 family large language models including chatgpt and gpt-4. Natural Language Processing Journal, 6:100048. find
  26. Svitlana Volkova, Yoram Bachrach, Michael Armstrong, and Vijay Sharma. 2015. Inferring latent user properties from texts published in social media. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 29. find
  27. Os Keyes. 2018. The misgendering machines: Trans/hci implications of automatic gender recognition. Proceedings of the ACM on human-computer interaction, 2(CSCW):1–22. find
  28. Gustavo Lopes and Helena Oliveira. 2023. A cascade approach for gender prediction from brazilian portuguese tweets. In RecSys. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/melo-2026-gender-identification-brazilian-portuguese-product/