Elamur Research⌕
☰
home / papers / 2026 / Aspect-based sentiment analysis in the accommodation domain with the BERTimbau model

Aspect-based sentiment analysis in the accommodation domain with the BERTimbau model

Original (Portuguese): Análise de Sentimento Baseada em Aspectos no Domínio de Acomodações Utilizando o modelo BERTimbau

Paper · 2026 · Franco Noronha Pereira, Larissa Astrogildo de Freitas and Ulisses Brisolara Corrêa

Summary

The paper addresses Aspect-Based Sentiment Analysis (ABSA) for Portuguese hotel reviews. The task has two parts: locating the aspect terms in a review and assigning each one a polarity. The authors aim to set a clear non-generative baseline using the monolingual encoder BERTimbau (Large variant) and compare two fine-tuning designs.

The first design is a pipeline: one model extracts aspects (AE) and a second classifies their sentiment (ASC). The second is end-to-end: a single token-classification model handles both subtasks through collapsed BIO tags that carry polarity, such as B-POS and I-NEG. The data come from the ABSAPT 2024 shared task, which uses TripAdvisor accommodation reviews annotated at the aspect level. The authors regrouped the one-aspect-per-row training file so that each review appears once with all its aspects.

The two designs behave differently. The pipeline is stronger at recall in extraction, while the end-to-end model is more precise but suffers because its label space is fragmented. Overall, both reach a combined F-measure of about 0.72. This is competitive with ABSAPT 2024 entries and gives a reference point for future hybrid or generative work.

Key points

  • Corpus: ABSAPT 2024 (derived from TripAdvisorBR-2020).
    • Training set: 4,828 samples from 1,322 reviews.
    • AE test set: 283 samples.
    • ASC test set: 1,176 samples from 282 reviews.
    • Positive polarity dominates; neutral and negative classes are scarce.
  • Results: the pipeline reaches an aspect-extraction F1 of 0.840. Both approaches score roughly 0.72 F-measure on the composite evaluation.
  • Labeling procedure:
    • Tokens are aligned to annotated spans using offset mappings.
    • Special tokens and non-initial subword pieces are masked with the label -100.
    • Text keeps its original casing because BERTimbau is a cased model.
  • Limitations: the study covers only one domain (lodging) and only explicit aspects with direct opinions. Implicit aspects and comparative opinions are left out.
  • Proposed future work:
    • LLM-based data augmentation to address class imbalance.
    • Testing other monolingual and multilingual encoders.
    • Additional domain-specific pre-training before fine-tuning.

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.Sentiment & emotion · 972 works

Cite

Cite as Franco Noronha Pereira, Larissa Astrogildo de Freitas and Ulisses Brisolara Corrêa. 2026. Análise de Sentimento Baseada em Aspectos no Domínio de Acomodações Utilizando o modelo BERTimbau. 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{pereira2026analise,
  title = {{Análise de Sentimento Baseada em Aspectos no Domínio de Acomodações Utilizando o modelo BERTimbau}},
  author = {Franco Noronha Pereira, Larissa Astrogildo {de Freitas} and Ulisses Brisolara Corrêa},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-1.82/}
}

References

16 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Alexandre Thurow Bender, Gabriel A Gomes, Emerson P Lopes, Ricardo M Araujo, Larissa A de Freitas, e Ulisses B Corrêa. 2024. Overview of absapt at iberlef 2024: Overview of the task on aspect-based sentiment analysis in portuguese. Procesamiento del Lenguaje Natural, 73:315–322. find
  2. Ulisses Brisolara Corrêa. 2021. Análise de Sentimento Baseada em Aspectos Usando Aprendizado Profundo: uma Proposta Aplicada à Língua Portuguesa. Tese (doutorado em ciência da computação), Universidade Federal de Pelotas, Porto Alegre. Disponível em http://guaiaca.ufpel.edu.br/handle/prefix/7433. A… ↗
  3. Júlia da Rocha Junqueira, Claudio Luis Junior, Félix Leonel V Silva, Ulisses Brisolara Corrêa, e Larissa A de Freitas. 2023. Albertina in action: an investigation of its abilities in aspect extraction, hate speech detection, irony detection, and question-answering. Em Simpósio Brasileiro de Tecnolog… find
  4. Gabriel Almeida Gomes. 2024. Avaliando a utilização de redes de grafos com atenção para a tarefa de análise de sentimentos em nível de aspecto em línguas de baixo recurso. Tese (mestrado em ciência da computação), Universidade Federal de Pelotas, Pelotas, Novembro. Disponível em https://guaiaca.ufpe… ↗
  5. Juliana Resplande Santanna Gomes, Eduardo Augusto Santos Garcia, Adalberto Ferreira Barbosa Junior, Ruan Chaves Rodrigues, Diogo Fernandes Costa Silva, Dyonnatan Ferreira Maia, Nádia Félix Felipe da Silva, Anderson da Silva Soares, and 1 others. 2023. Deep learning brasil at absapt 2022: Portuguese… — [arXiv:2311.05051](https://arxiv.org/abs/2311.05051) find
  6. Bing Liu. 2012. Sentiment Analysis and Opinion Mining. Morgan & Claypool Publishers, San Rafael. find
  7. Émerson Philippe Lopes, Larissa Freitas, Gabriel Gomes, Gerônimo Lemos, Luiz Otávio Hammes, e Ulisses B Corrêa. 2022. Exploring bert for aspect-based sentiment analysis in portuguese language. Em The International FLAIRS Conference Proceedings, volume 35, Daytona Beach, FL. Florida Online Journals. find
  8. Huaishao Luo, Tianrui Li, Bing Liu, e Junbo Zhang. 2019. DOER: Dual cross-shared RNN for aspect term-polarity co-extraction. Em Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, páginas 591–601, Florence, Italy. Association for Computational Linguistics. find
  9. Felix LV da Silva, Guilherme da S Xavier, Heliks M Mensenburg, Rodrigo F Rodrigues, Leonardo P dos Santos, Ricardo M Araújo, Ulisses B Corrêa, e Larissa A de Freitas. 2022. find
  10. Absapt 2022 at iberlef: Overview of the task on aspect-based sentiment analysis in portuguese. find
  11. Fábio Souza, Rodrigo Nogueira, e Roberto Lotufo. 2020. Bertimbau: Pretrained bert models for brazilian portuguese. Em Intelligent Systems: 9th Brazilian Conference, BRACIS 2020, Rio Grande, Brazil, October 20–23, 2020, Proceedings, Part I, página 403–417, Berlin, Heidelberg. Springer-Verlag. find
  12. Bruno Dal Pontte, Anderson Aparecido do Carmo Frasão, Marcus Vinícius Reisdoefer Pereira, Tiago Heinrich, e Vinicius Fulber-Garcia. 2024. Teamufpr at absapt 2024: Improving sentiment polarity classification with data augmentation using chatgpt. Em Proceedings of the 6th Iberian Languages Evaluation… find
  13. Dang Van Thin, Nguyen Tuan Kiet, Duong Ngoc Hao, e Ngan Luu-Thuy Nguyen. 2024. Abcd team at absapt 2024: Classification-based versus generation-based approach for aspect-based sentiment analysis in portuguese. Em Proceedings of the 6th Iberian Languages Evaluation Forum (IberLEF 2024), CEUR Workshop… find
  14. Larissa Astrogildo de Freitas. 2015. Feature-level sentiment analysis applied to brazilian portuguese reviews. Tese (doutorado em ciência da computação), Pontifícia Universidade Católica do Rio Grande do Sul, Porto Alegre. Disponível em https://tede2.pucrs.br/tede2/handle/tede/6031. Acesso em: 2025… ↗
  15. Qihuang Zhong, Haiyun Li, Luyao Zhuang, Juhua Liu, e Bo Du. 2024. Iterative data generation with large language models for aspect-based sentiment analysis. arXiv preprint arXiv:2407.00341. — [arXiv:2407.00341](https://arxiv.org/abs/2407.00341) find
  16. Jacob Devlin, Ming-Wei Chang, Kenton Lee, e Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. Preprint, arXiv:1810.04805.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/pereira-2026-analise-de-sentimento-baseada-em/