Elamur Research⌕
☰
home / papers / 2026 / AspectRAG: a retrieval and generation architecture for aspect-based sentiment analysis

AspectRAG: a retrieval and generation architecture for aspect-based sentiment analysis

Original (Portuguese): AspectRAG: Uma Arquitetura de Recuperação e Geração para Análise de Sentimentos Baseada em Aspectos

Paper · 2026 · Erick R. Ribeiro, André Carvalho and Rhedson Esashika

Summary

Aspect-based sentiment analysis (ABSA) goes beyond labelling a whole text as positive or negative. It identifies which aspects are discussed, which opinion words describe them, and with what polarity. For Portuguese, the hardest variant, aspect sentiment triplet extraction (ASTE), has mostly relied on supervised or rule-based systems that need annotated data and costly training. The authors propose AspectRAG, a retrieval-augmented pipeline that requires no fine-tuning. An LLM first extracts aspect terms using few-shot prompting. Each aspect, rather than the full sentence, is embedded and used to query a vector index of annotated training sentences through approximate nearest-neighbour search. When a sentence has several aspects, the result lists are merged, and the top examples are placed in the prompt of a generator (GPT-4o-mini or Qwen-3-8B) that outputs (aspect, opinion, polarity) triplets. On the ReLi and ReHol datasets, AspectRAG beats supervised baselines, including the Portuguese state of the art. The work suggests that targeted example retrieval can replace supervised training for low-resource ABSA.

Key points

  • Tasks covered: ATE (aspect term extraction), OTE (opinion term extraction) and ASTE (aspect sentiment triplet extraction). Evaluation uses the Portuguese datasets ReLi and ReHol.
  • Best reported scores are 93.47% for ATE, 80.68% for OTE and 69.83% for ASTE. These exceed the OTE-MTL, CMLA-MTL and BOTE baselines; BOTE combines BERTimbau, BiLSTM, graph attention and a biaffine scorer.
  • Retrieval details:
    • Queries are built only from aspect embeddings, compared by cosine similarity.
    • Each aspect retrieves its own top-N list.
    • Multiple lists are combined with Reciprocal Rank Fusion.
    • The retrieved "documents" are annotated training sentences containing gold triplets.
  • The ablation study indicates that aspect-guided semantic retrieval drives most of the gains. The parameter count of the generator LLM matters much less.
  • The authors present it as the first RAG architecture tailored to ABSA subtasks in Portuguese. Gains held with both GPT-4o-mini and Qwen-3-8B as generators.

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 · 18,614 worksArtificial Intelligence · 17,965 works

🧭 Topics 2Research subjects and languages the work is about.Sentiment & emotion · 1,035 worksInformation retrieval & search · 552 works

Cite

Cite as Erick R. Ribeiro, André Carvalho and Rhedson Esashika. 2026. AspectRAG: Uma Arquitetura de Recuperação e Geração para Análise de Sentimentos Baseada em Aspectos. 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{ribeiro2026aspectrag,
  title = {{AspectRAG: Uma Arquitetura de Recuperação e Geração para Análise de Sentimentos Baseada em Aspectos}},
  author = {Erick R. Ribeiro, André Carvalho and Rhedson Esashika},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-1.19/}
}

References

29 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. José Meléndez Barros e Glauber De Bona. 2021. A deep learning approach for aspect sentiment triplet extraction in portuguese. Em Intelligent Systems, páginas 343–358, Cham. Springer International Publishing. find
  2. V. Câmara e T. Melo. 2022. Estudo de método de extração de aspectos para português do brasil baseado em regras. Anais do XI Brazilian Workshop on Social Network Analysis and Mining, páginas 192–203. find
  3. Abir Chakraborty. 2024. End-to-end aspect based sentiment analysis using graph attention network. Em Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM ’24, página 3663–3668, New York, NY, USA. Association for Computing Machinery. find
  4. Z. Chen e T. Qian. 2020. Enhancing aspect term extraction with soft prototypes. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, páginas 2107–2117. find
  5. Judith A. Chevalier e Dina Mayzlin. 2006. The effect of word of mouth on sales: Online book reviews. Journal of Marketing Research, 43(3):345–354. find
  6. Keith Cortis e Brian Davis. 2021. Over a decade of social opinion mining: a systematic review. Artificial intelligence review, 54(7):4873–4965. find
  7. Vanessa Câmara e Tiago Melo. 2022. Estudo de método de extração de aspectos para português do brasil baseado em regras. Em Anais do XI Brazilian Workshop on Social Network Analysis and Mining, páginas 192–203, Porto Alegre, RS, Brasil. SBC. find
  8. Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, e Ming-Wei Chang. 2020. Realm: Retrieval-augmented language model pre-training. Em Proceedings of the 37th International Conference on Machine Learning (ICML) / PMLR 119, páginas 3929– 3938. Preprint disponível em arXiv:2002.08909.
  9. Gautier Izacard e Édouard Grave. 2020. Leveraging passage retrieval with generative models for open domain question answering. arXiv preprint, arXiv:2007.01282. Versão publicada e discussões em conferências posteriores (ex.: EACL/ACL).
  10. S. Y. B. Jian, T. Nayak, N. Majumder, e S. Poria. 2021. Aspect sentiment triplet extraction using reinforcement learning. Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, páginas 1–10. find
  11. Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, e Wen-tau Yih. 2020. Dense passage retrieval for open-domain question answering. Em Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), páginas 6769–6781. Prep…
  12. P. Lewis, E. Perez, A. Piktus, F. Petroni, and 1 others. 2020a. Retrieval-augmented generation for knowledge-intensive nlp. Em Advances in Neural Information Processing Systems (NeurIPS), páginas 9459–9474. find
  13. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, e Douwe Kiela. 2020b. Retrieval-augmented generation for knowledge-intensive nlp. Em Advances in Neural Information Processing…
  14. X. Li, L. Wang, P. Li, e Z. Yang. 2018. Aspect term extraction with history attention and selective fusion. Proceedings of the 27th International Joint Conference on Artificial Intelligence. find
  15. Bing Liu e Lei Zhang. 2012. A Survey of Opinion Mining and Sentiment Analysis, páginas 415–463. Springer US, Boston, MA. find
  16. M. T. Machado, T. A. S. Pardo, E. E. S. Ruiz, e A. Di Felippo. 2021a. Learning rules for automatic identification of implicit aspects in portuguese. Em Simpósio Brasileiro de Tecnologia da Informação e da Linguagem Humana (STIL). find
  17. Mateus Tarcinalli Machado, Thiago Alexandre Salgueiro Pardo, Evandro Eduardo Seron Ruiz, e Ariani Di Felippo. 2021b. Learning rules for automatic identification of implicit aspects in portuguese. Em Simpósio Brasileiro de Tecnologia da Informação e da Linguagem Humana (STIL). Disponível também em Re… find
  18. I. Naglik, L. Zhang, e B. Liu. 2024. Aste-transformer: Modelling dependencies in aspect-sentiment-opinion triplet extraction. Findings of the Association for Computational Linguistics: EMNLP 2024, páginas 129–139. find
  19. Gabriel Pereira, Luciano Barbosa, Johny Moreira, Tiago Melo, e Altigran Silva. 2024. Enhancing aspect-based sentiment analysis for portuguese using instruction tuning. Em Anais do XXI Encontro Nacional de Inteligência Artificial e Computacional, páginas 990–1001, Porto Alegre, RS, Brasil. SBC. find
  20. Maria Pontiki, Dimitrios Galanis, Haris Papageorgiou, Ion Androutsopoulos, Suresh Manandhar, Mohammad Al-Smadi, Mahmoud Al-Ayyoub, Yanyan Zhao, Bing Qin, Orphée De Clercq, and 1 others. 2016. Semeval-2016 task 5: Aspect based sentiment analysis. Em International workshop on semantic evaluation, pági…
  21. Eloize Seno, Lucas Silva, Fábio Anno, Fabiano Rocha, e Helena Caseli. 2024. Aspect-based sentiment analysis in comments on political debates in Portuguese: evaluating the potential of ChatGPT. Em Proceedings of the 16th International Conference on Computational Processing of Portuguese - Vol. 1, pág… find
  22. Tun Thura Thet, Jin-Cheon Na, e Christopher SG Khoo. 2010. Aspect-based sentiment analysis of movie reviews on discussion boards. Journal of information science, 36(6):823–848. find
  23. Qi Wang e et al. 2024. In-context example retrieval from multi-perspectives for few-shot aspect-based sentiment analysis. Em Proceedings of LREC-COLING 2024 (LREC Main Conference), páginas —. Recuperação de exemplos para prompting / ICL no contexto de ABSA. find
  24. W. Wang, S. J. Pan, D. Dahlmeier, e X. Xiao. 2017. Coupled multi-layer attentions for co-extraction of aspect and opinion terms. Em Proceedings of the AAAI Conference on Artificial Intelligence. find
  25. Kyung Hyan Yoo e Ulrike Gretzel. 2008. What motivates consumers to write online travel reviews? Information Technology & Tourism, 10(4):283–295. find
  26. Guoxin Yu, Redmond Liu, Haiyun Jiang, Shuming Shi, e et al. 2023. Retrieval-based aspect sentiment triplet extraction via label interpolation. Em Findings of the Association for Computational Linguistics: ACL 2023, páginas —. Código/ PDF: https://aclanthology.org/2023.findings-acl.303.pdf. ↗
  27. C. Zhang, Q. Li, D. Song, e B. Wang. 2020. A multi-task learning framework for opinion triplet extraction. Em Findings of the Association for Computational Linguistics: EMNLP 2020, páginas 819–828. find
  28. Wenxuan Zhang, Xin Li, Yang Deng, Lidong Bing, e Wai Lam. 2022. A survey on aspect-based sentiment analysis: Tasks, methods, and challenges. IEEE Transactions on Knowledge and Data Engineering, 35(11):11019–11038. find
  29. Daniel A. Pereira. 2020. A survey of sentiment analysis in the portuguese language. Artificial Intelligence Review, 54(2):1087–1115. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ribeiro-2026-aspectrag-uma-arquitetura-de-recuperacao/