A Hybrid RAG Pipeline for Linking Fraud News to Public Procurement Records

• Paulo Marcos de Assis UFSC
• Márcio Castro UFSC
• Jônata Tyska Carvalho UFSC

DOI: https://doi.org/10.5753/eniac.2026.30036

Resumo

Public procurement is a major target for fraud and mismanagement, and linking journalistic investigations to official procurement records can support automated auditing systems. However, news reports rarely mention identifiers such as tender numbers, making the matching task difficult in large databases with many near-identical descriptions. We present a hybrid Retrieval-Augmented Generation (RAG) pipeline combining structured attribute extraction, SQL pre-filtering, TF-IDF retrieval and Large Language Model (LLM) reranking to link fraud-related news to official procurement processes. Experiments were conducted on 315,052 real procurement records from Santa Catarina (Brazil), using 500 synthetic news-process pairs for development and 214 held-out pairs for final evaluation. Among thirteen retrieval methods, classical TF-IDF outperformed Portuguese neural encoders by more than 57 percentage points in top-50 retrieval coverage. Neural Portuguese encoders (BERTimbau, HeIBERT) were evaluated as zero-shot baselines; the 57-point gap versus TF-IDF reflects untrained embeddings and does not represent their fully optimized potential. The best configuration, TF-IDF top-10 retrieval followed by a single gpt-oss:20b reranking call, achieved 76.8% development accuracy and 71.5% held-out test accuracy.

Referências

Abbas, A., Mark, L., Shanavas, N., et al. (2026). Struct2unstruct: Creating tender ner datasets from structured procurement records using large language models. Preprint.

Assis, P. M. d., Castro, M., and Carvalho, J. T. (2025). Machine learning-based classification of portuguese news articles on public procurement fraud. In Anais da I Escola Regional de Aprendizado de Máquina e Inteligência Artificial da Região Sul (ERAMIA-RS 2025), Porto Alegre, RS. Sociedade Brasileira de Computação.

Assis, P. M. d., Castro, M., and Carvalho, J. T. (2026). Information extraction from brazilian news articles on public procurement fraud. In Anais do 41º Simpósio Brasileiro de Bancos de Dados (SBBD). Accepted for publication; awaiting publication.

Ather, H. (2026). Llm-assisted record linkage: A framework for official statistics. Statistical Journal of the IAOS.

Bonifacio, L. H., Abonizio, H., Fadaee, M., and Nogueira, R. (2022). Inpars: Data augmentation for information retrieval using large language models. ArXiv, abs/2202.05144.

Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. (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 (NAACL-HLT).

Halterman, A. (2025). Synthetically generated text for supervised text analysis. Political Analysis, 33(3):181–194.

Izacard, G. and Grave, E. (2021). Leveraging passage retrieval with generative models for open domain question answering. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics (EACL), pages 874–880.

Karpov, I., Kirillovich, A., et al. (2026). Synel: A synthetic benchmark for entity linking. Expert Systems with Applications.

Karpukhin, V., Oğuz, B., Min, S., Lewis, P., Wu, L., Edunov, S., Chen, D., and Yih, W.-t. (2020). Dense passage retrieval for open-domain question answering. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 6769–6781.

Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W.-t., Rocktäschel, T., Riedel, S., and Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive nlp tasks. In Advances in Neural Information Processing Systems (NeurIPS), pages 9459–9474.

Lima, W. E. M., da Silva, V. R., da Silva, J. C., Rabêlo, R. d. A. L., and de Paiva, A. C. (2026). Helbert: A bert-based pretraining model for public procurement tasks in portuguese. Journal of the Brazilian Computer Society.

Lyra, M. S., Damásio, B., Pinheiro, F. L., et al. (2022). Fraud, corruption, and collusion in public procurement activities, a systematic literature review on data-driven methods. Applied Network Science, 7:83.

Nogueira, R., Jiang, Z., Pradeep, R., and Lin, J. (2020). Document ranking with a pre-trained sequence-to-sequence model. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 708–718.

OECD (2018). The Role of the Media and Investigative Journalism in Combating Corruption. OECD Publishing, Paris.

Oliveira, E. T. L., Alves, P. M. H., Mendes, F. I. d. L., da Silva Júnior, E. F. P., Alves, A. L. F., and Baptista, C. d. S. (2026). Improving audits in public procurement: An approach to recognition of named entities for brazilian official gazettes. Procedia Computer Science, pages 141–148.

Premasiri, D., Ranasinghe, T., Mitkov, R., et al. (2025). Survey on legal information extraction: current status and open challenges. Knowledge and Information Systems, 67:1128–1135.

Qin, X., Chai, C., Tang, N., et al. (2022). Synthesizing privacy preserving entity resolution datasets. In 2022 IEEE 38th International Conference on Data Engineering.

Rahmani, H. A., Craswell, N., Yilmaz, E., et al. (2024). Synthetic test collections for retrieval evaluation. arXiv preprint arXiv:2405.07767.

Robertson, S. and Zaragoza, H. (2009). The probabilistic relevance framework: Bm25 and beyond. Foundations and Trends in Information Retrieval, 3(4):333–389.

Schneider dos Santos, E., Machado dos Santos, M., Castro, M., et al. (2025). Detection of fraud in public procurement using data-driven methods: A systematic mapping study. EPJ Data Science, 14(52). Published 22 July 2025.

Souza, F., Nogueira, R., and Lotufo, R. (2020). Bertimbau: Pretrained bert models for brazilian portuguese. In Proceedings of the 9th Brazilian Conference on Intelligent Systems (BRACIS), pages 403–417. Springer.

Sun, W., Yan, L., Ma, X., Wang, S., Ren, P., Chen, Z., Yin, D., and Ren, Z. (2023). Is chatgpt good at search? investigating large language models as re-ranking agents. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP).

Thakur, N., Reimers, N., Rücklé, A., Srivastava, A., and Gurevych, I. (2021). Beir: A heterogeneous benchmark for zero-shot evaluation of information retrieval models. Proceedings of the 35th Conference on Neural Information Processing Systems (NeurIPS), Datasets and Benchmarks Track.

UNODC (2014). Reporting on Corruption: A Resource Tool for Governments and Journalists. United Nations Office on Drugs and Crime, Vienna.

Wang, S., Liu, Y., Xu, Y., Zhu, C., and Zeng, M. (2021). Want to reduce labeling cost? gpt-3 can help. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 4195–4205.

Zhuang, H., Qin, Z., Jagerman, R., Hui, K., Ma, J., Lu, J., Ni, J., Wang, X., and Bendersky, M. (2023). Rankt5: Fine-tuning t5 for text ranking with ranking losses. Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval.