Elamur Research⌕
☰
home / papers / 2026 / Field of Science and Technology Classification of Academic Documents in Portuguese

Field of Science and Technology Classification of Academic Documents in Portuguese

Paper · 2026 · Ivo Simões, Hugo Gonçalo Oliveira and João Correia

Summary

Academic repositories often have missing or wrong metadata. In Estudo Geral (EG), the University of Coimbra's repository, many theses have no OECD Field of Science and Technology (FOS) label. The authors treat FOS assignment for Portuguese theses as a five-class text classification problem. Agricultural Sciences is left out because the university has no unit in that area. They built a dataset of 9,696 theses, each described by title, keywords and abstract. Records that were incomplete or had several FOS labels were removed, and the data was split 80/20 with stratification. Four encoders were fine-tuned and compared against five decoder LLMs prompted zero-shot in Portuguese with structured output. The fine-tuned encoders clearly performed better. BERTimbau reached a macro F1 of about 0.88 using all three fields, roughly 17 points ahead of the best LLM. The authors describe this as the first study of the task for Portuguese. It suggests that small task-tuned models are still preferable when labelled data exist, with prompted LLMs as a fallback. The dataset and fine-tuned models are shared on HuggingFace.

Key points

  • Encoders tested: multilingual cased BERT BASE, mE5 small, BERTimbau BASE and Albertina-100M-PTPT. Each was fine-tuned for 10 epochs with AdamW (lr 3×10⁻⁵), and results average four runs.
  • Decoders tested zero-shot: GPT-5 Nano, gpt-oss-20b, Llama3.1-8B Instruct, Llama3.3-70B Instruct and Qwen3-8B. The best was Llama3.3 70B (F1 ≈ 0.71), and the weakest was Llama3.1 8B (≈ 0.64).
  • Using title, keywords and abstract together gave the best results, with BERTimbau at F1 0.881. A title-only input, averaging about 25 tokens, still gave encoders F1 ≈ 0.78–0.79.
  • The label distribution is skewed toward Medical and Health Sciences (34.0%) and Social Sciences (32.3%). Natural Sciences was the hardest class for both the best encoder and the best decoder.
  • Qwen3 8B outperformed Llama3.1 8B on every field combination. The authors attribute this to Qwen3's stronger multilingual pretraining.

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

Cite

Cite as Ivo Simões, Hugo Gonçalo Oliveira and João Correia. 2026. Field of Science and Technology Classification of Academic Documents in Portuguese. 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{simoes2026field,
  title = {{Field of Science and Technology Classification of Academic Documents in Portuguese}},
  author = {Ivo Simões, Hugo Gonçalo Oliveira and João Correia},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-1.104/}
}

References

28 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Hugo Abonizio, Thales Sales Almeida, Thiago Laitz, Roseval Malaquias Junior, Giovana Kerche Bonás, Rodrigo Nogueira, and Ramon Pires. 2024. Sabi\\’a-3 technical report. arXiv preprint arXiv:2410.12049. — [arXiv:2410.12049](https://arxiv.org/abs/2410.12049) find
  2. Cristina Arhiliuc, Raf Guns, Walter Daelemans, and Tim CE Engels. 2025. Journal article classification using abstracts: a comparison of classical and transformer-based machine learning methods. Scientometrics, 130(1):313–342. find
  3. Akari Asai, Jacqueline He, Rulin Shao, Weijia Shi, Amanpreet Singh, Joseph Chee Chang, Kyle Lo, Luca Soldaini, Sergey Feldman, Mike D’arcy, and 1 others. 2024. OpenScholar: Synthesizing scientific literature with Retrieval-Augmented LLMs. arXiv preprint arXiv:2411.14199. find
  4. Iz Beltagy, Kyle Lo, and Arman Cohan. 2019. SciB-ERT: A pretrained language model for scientific text. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 4 https://huggingface.co/collections/ivosimoes/ propor-fos-classification 5 https://amaliallm.pt/9t… ↗
  5. Amanda da Silva Oliveira, Thiago de Carvalho Ce-cote, João Paulo Reis Alvarenga, Vander Luis de Souza Freitas, and Eduardo José da Silva Luz. 2024. Toxic speech detection in Portuguese: A comparative study of large language models. In Proceedings of the 16th International Conference on Computational… find
  6. 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,…
  7. Nikolaos Gialitsis, Sotiris Kotitsas, and Haris Papageorgiou. 2022. Scinobo: A hierarchical multi-label classifier of scientific publications. In Companion Proceedings of the Web Conference 2022, WWW ’22, page 800–809, New York, NY, USA. Association for Computing Machinery. find
  8. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, and 542 o… — The Llama 3 Herd of Models
  9. Michael Gusenbauer, Jochen Endermann, Harald Huber, Simon Strasser, Andreas-Nizar Granitzer, and Thomas Ströhle. 2025. Fine-tuning scibert to enable asjc-based assessments of the disciplinary orientation of research collections. Scientometrics, pages 1–38. find
  10. Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jaewoo Kang. 2020. BioBERT: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics, 36(4):1234–1240.
  11. Kyle Lo, Lucy Lu Wang, Mark Neumann, Rodney Kinney, and Daniel Weld. 2020. S2ORC: The semantic scholar open research corpus. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4969–4983, Online. ACL.
  12. Óscar E. Mendoza, Wojciech Kusa, Alaa El-Ebshihy, Ronin Wu, David Pride, Petr Knoth, Drahomira Herrmannova, Florina Piroi, Gabriella Pasi, and Allan Hanbury. 2022. Benchmark for research theme classification of scholarly documents. In Proceedings of the Third Workshop on Scholarly Document Processin… find
  13. Ana Eva Miguéis and Bruno Neves. 2021. A visão dos gestores de repositórios. o caso da universidade de coimbra. Sob a lente da Ciência Aberta: Olhares de Portugal, Espanha e Brasil, pages 273–294. find
  14. OpenAI, :, Sandhini Agarwal, Lama Ahmad, Jason Ai, Sam Altman, Andy Applebaum, Edwin Arbus, Rahul K. Arora, Yu Bai, Bowen Baker, Haiming Bao, Boaz Barak, Ally Bennett, Tyler Bertao, Nivedita Brett, Eugene Brevdo, Greg Brockman, Sebastien Bubeck, and 108 others. 2025. gpt-oss-120b & gpt-oss-20b model…
  15. Yifan Peng, Shankai Yan, and Zhiyong Lu. 2019. Transfer learning in biomedical natural language processing: An evaluation of BERT and ELMo on ten benchmarking datasets.
  16. Tomás Pinto, Bruno Ferreira, Catarina Silva, and Hugo Gonçalo Oliveira. 2025. Prompting LLMs for relation classification in Portuguese: Is it worth it? In Progress in Artificial Intelligence, pages 249–261, Cham. Springer Nature Switzerland. find
  17. Susie Xi Rao, Peter H Egger, and Ce Zhang. 2025. Hierarchical classification of research fields in the mag science network using deep learning. Quantitative Science Studies, pages 1–60. find
  18. Zhyar Rzgar K Rostam and Gábor Kertész. 2024. Fine-tuning large language models for scientific text classification: A comparative study. find
  19. Mobashir Sadat and Cornelia Caragea. 2022. Hierarchical multi-label classification of scientific documents. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 8923–8937, Abu Dhabi, United Arab Emirates. ACL. overview of microsoft academic service (mas) a… find
  20. Fábio Souza, Rodrigo Nogueira, and Roberto Lotufo. 2020. BERTimbau: Pretrained BERT models for Brazilian Portuguese. In Brazilian Conference on Intelligent Systems (BRACIS), pages 403–417. find
  21. Ross Taylor, Marcin Kardas, Guillem Cucurull, Thomas Scialom, Anthony Hartshorn, Elvis Saravia, Andrew Poulton, Viktor Kerkez, and Robert Stojnic. 2022. Galactica: A large language model for science. arXiv preprint arXiv:2211.09085.
  22. Autumn Toney and James Dunham. 2022. Multi-label classification of scientific research documents across domains and languages. In Proceedings of the third workshop on scholarly document processing, pages 105–114. find
  23. Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. 2024. Multilingual e5 text embeddings: A technical report. Preprint, arXiv:2402.05672.
  24. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, and 41 others. 2025. Qwen3 technical report. Preprint, arXiv:2505.09388.
  25. Tarek Saier and Michael Färber. 2019. Bibliometric-enhanced arxiv: A data set for paper-based and citation-based tasks. In Proceedings of the 8th International Workshop on Bibliometric-enhanced Information Retrieval (BIR) co-located with the 41st European Conference on Information Retrieval (ECIR 20… find
  26. Rodrigo Santos, João Rodrigues, Luís Gomes, João Silva, António Branco, Henrique Lopes Cardoso, Tomás Freitas Osório, and Bernardo Leite. 2024a. Fostering the ecosystem of open neural encoders for portuguese with Albertina PT-* family. In Proceedings of the 3rd Annual Meeting of the Special Interest… find
  27. Rodrigo Santos, João Ricardo Silva, Luís Gomes, João Rodrigues, and António Branco. 2024b. Advancing generative AI for Portuguese with open decoder Gervásio PT*. In Proceedings of the 3rd Annual Meeting of the Special Interest Group on Under-resourced Languages @ LREC-COLING 2024, pages 16–26, Torin… find
  28. Arnab Sinha, Zhihong Shen, Yang Song, Hao Ma, Darrin Eide, Bo-June Hsu, and Kuansan Wang. 2015. An find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/simoes-2026-field-science-technology-classification-academic/