Lightweight Adherence Prediction and Justification for Government Strategic Priorities

• Matheus Utino USP
• Marcos Gôlo USP
• Marcelo Turine USP / UFMS
• Ricardo Marcacini USP

DOI: https://doi.org/10.5753/lasdigov.2026.23817

Resumo

The Brazilian National Postgraduate System plays a strategic role in scientific development and policymaking. The CAPES National Agenda organizes 470 strategic themes into 20 macro-themes, reflecting priorities from Brazil’s 27 Federative Units. Given the large annual volume of academic outputs, identifying their alignment with these priorities is essential. This study evaluates lightweight machine learning models as scalable alternatives to LLMs for adherence prediction and justification. Using 42,026 academic outputs and nine classifiers, Bag-of-Words with Random Forest achieved competitive performance (F1 of 0.755), outperforming semantic embeddings. A fine-tuned PTT5-base produced explanations highly similar to human references.

Referências

Annepaka, Y. and Pakray, P. (2025). Large language models: a survey of their development, capabilities, and applications. Knowledge and Information Systems, 67(3):2967–3022.

Awasthy, P., Trivedi, A., Li, Y., Bornea, M., Cox, D., Daniels, A., Franz, M., Goodhart, G., Iyer, B., Kumar, V., et al. (2025). Granite embedding models. arXiv preprint arXiv:2502.20204.

Barbier, V. and Jeangirard, E. (2025). Mapping scientific communities at scale. arXiv preprint arXiv:2501.10035.

Brasil (2021). Lei nº 14.129, de 29 de março de 2021: Princípios, regras e instrumentos para o governo digital e para o aumento da eficiência pública. Diário Oficial da União. Accessed: 2025.

CAPES (2021). Evolução do SNPG no decênio do PNPG 2011–2020. Technical report, Comissão Especial de Acompanhamento do PNPG 2011–2020 / CAPES, Brasília. Accessed: 30 mar. 2026.

CAPES (2023). Plano nacional de pós-graduação – PNPG 2024–2028: versão preliminar para consulta pública. Technical report, Ministério da Educação / CAPES, Brasília. Accessed: 30 mar. 2026.

CAPES (2026). Portal do observatório da pós-graduação. Accessed: March 30, 2026.

Carmo, D., Piau, M., Campiotti, I., Nogueira, R., and Lotufo, R. (2020). Ptt5: Pretraining and validating the t5 model on brazilian portuguese data. arXiv preprint arXiv:2008.09144.

Gôlo, M. P. S. and Utino, M. Y. R. (2025). One-class lightweight interpretable filtering for academic profiles and strategic themes affinity. In SBAC-PADW, pages 147–154. IEEE.

ICMC (2025). Parceria entre ufms e usp desenvolve projeto inédito da capes que mapeia teses e dissertações em temas estratégicos da pós-graduação nos estados brasileiros. Accessed: 29 mar. 2026.

Jiang, Y., Pang, P. C.-I., Wong, D., and Kan, H. Y. (2023). Natural language processing adoption in governments and future research directions: A systematic review. Applied Sciences, 13(22):12346.

Jin, Z. and Mihalcea, R. (2022). Natural language processing for policymaking. In Handbook of computational social science for policy, pages 141–162. Springer.

Landhuis, E. (2016). Scientific literature: Information overload. Nature, 535(7612):457–458.

Miranda, A. C. D., Utino, M. Y. R., Gôlo, M. P. S., Dos Santos, M. A. A., and de Souza, M. C. (2025). Reducing costs in large-scale classification: A hybrid bert–llm strategy. In SBAC-PADW, pages 131–138. IEEE.

Rigo, S. and Assunção, M., editors (2025). International Symposium on Computer Architecture and High Performance Computing Workshops SBAC-PADW.

Sakiyama, K. M., Fujimoto, M. L., Santin, R. V., and Rezende, S. O. (2025). Leandl hpc challenge 2025: Applying large-scale model adaptation techniques. In SBAC-PADW, pages 139–146. IEEE.

Utino, M. Y. R. and Gôlo, M. P. S. (2025). Efficient filtering with bert embeddings for researcher–topic affinity prediction in hpc pipelines. In SBAC-PADW, pages 155–162. IEEE.