Automated Audit of Judicial Metadata with Natural Language Processing: A Multi-Level Cascade Approach

• Manuel Lucala Zengo UTFPR
• Marlon Marcon TJGO
• Fernando Ribeiro Trindade UFG
• Antonio Pires de Castro Júnior UFG
• Willgnner Ferreira Santo UFG
• Manuel Findas UTFPR

DOI: https://doi.org/10.5753/eniac.2026.31570

Resumo

In the Brazilian judiciary, metadata such as case value, parties’ names, and identification documents (CPF/CNPJ) are manually registered in the PROJUDI system alongside each filed petition, and clerks then manually verify this data against the petition—a process averaging 3 minutes per case. We propose a multi-level cascade combining high-precision regex, trained NER models, zero-shot extraction via GLiNER as a semantic judge, and a reverse search technique seeking expected metadata values within the petition text across multiple Brazilian formatting variants. Experiments on two real-case batches from the Court of Justice of Goiás (TJGO) show that, among validatable cases, extracted values agree with PROJUDI’s registered metadata in 96% of cases in a noisier homologation batch (67 of 70 validatable, out of 185 total) and in 95.58% of a cleaner production batch (173 of 181 validatable, out of 182 total), at zero operational cost.

Referências

Brasil (2006). Lei nº 11.419, de 19 de dezembro de 2006. Dispõe sobre a informatização do processo judicial. Diário Oficial da União, Brasília, DF. Disponível em: [link].

Chiticariu, L., Li, Y., and Reiss, F. R. (2013). Rule-based information extraction is dead! Long live rule-based information extraction systems! Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 827–832.

Conselho Nacional de Justiça (2020). Resolução nº 332, de 21 de agosto de 2020. Dispõe sobre a ética, a transparência e a governança na produção e no uso de Inteligência Artificial no Poder Judiciário. Diário da Justiça Eletrônico, Brasília. Disponível em: [link].

Dale, R. (2019). Law and word order: NLP in legal tech. Natural Language Engineering, 25(1):211–217.

de Araujo, P. H. L., de Campos, T. E., de Oliveira, R. R. R., Stauffer, M., Couto, S., and Bermejo, P. (2018). LeNER-Br: a dataset for named entity recognition in Brazilian legal text. In Proceedings of the International Conference on Computational Processing of the Portuguese Language (PROPOR), pages 313–323. Springer.

Honnibal, M., Montani, I., Van Landeghem, S., and Boyd, A. (2020). spaCy: Industrial-strength natural language processing in Python. In Zenodo. DOI: 10.5281/zen-, odo.1212303.

Souza, F., Nogueira, R., and Lotufo, R. (2020). BERTimbau: Pretrained BERT models for Brazilian Portuguese. In Proceedings of the Brazilian Conference on Intelligent Systems (BRACIS), pages 403–417. Springer.

Zaratiana, U., Ngouda, N., Ayach, A., Tomeh, N., and Charnois, T. (2023). GLiNER: Generalist model for named entity recognition using bidirectional transformer. arXiv preprint arXiv:2311.08526.