PREVE_pet - Automatização da Classificação de Textos Jurídicos: Desenvolvimento de um Sistema Baseado em Modelos de Inteligência Artificial para a PGFN

• Bruno Justino Garcia Praciano ^[1] ; Fábio L. L. de Mendonça ^[1] ; Thiago Leite de Sousa ^[1] ; Flávio Garcia Praciano ^[1] ; Geraldo P. Rocha Filho ^[2] ; Laerte Peota de Melo ^[1]
  1. [1] Universidade de Brasília

     Universidade de Brasília

     Brasil

     
  2. [2] Universidad Estatal del Sudoeste de Bahía

     Universidad Estatal del Sudoeste de Bahía

     Brasil

     
• Localización: RISTI: Revista Ibérica de Sistemas e Tecnologias de Informação, ISSN-e 1646-9895, Nº. Extra 77, 2025, págs. 117-131
• Idioma: portugués
• Títulos paralelos:
  • PREVE_pet – Automation of Legal Text Classification: Development of an Artificial Intelligence-Based System for the PGFN
• Enlaces
  • Texto completo (pdf)
• Resumen
  • English

    The increasing demand for efficiency in document management at Brazil’s Office of the Attorney General of the National Treasury (PGFN) underscores the challenges posed by the high volume of complex legal texts.

    Manual classification is time-intensive, error-prone, and lacks scalability. This study presents PREVE_pet, an automated legal document classification system built upon advanced natural language processing models, including BERT and LLAMA. The system performs preprocessing, vectorization, categorization, and generates legal summaries with a high degree of accuracy. Security guidelines from the Open Worldwide Application Security Project (OWASP) were implemented to address potential risks in AI models, such as data leakage and adversarial attacks.

    Experimental results demonstrate that PREVE_pet achieves classification accuracy exceeding 90%, significantly reducing human error and processing time. These findings suggest that PREVE_pet is an effective, secure, and scalable solution for automating workflows in legal document management.

  • português

    A crescente demanda por eficiência na gestão documental da Procuradoria-Geral da Fazenda Nacional (PGFN) evidencia os desafios impostos pelo grande volume de textos jurídicos complexos. A classificação manual é demorada, propensa a erros e difícil de escalar. Para enfrentar esse cenário, propõe-se o PREVE_pet, sistema automatizado de classificação jurídica baseado em modelos de linguagem natural, como BERT e LLAMA. A solução realiza préprocessamento, vetorização, categorização e geração de resumos jurídicos com alta precisão. Foram aplicadas diretrizes do Open Worldwide Application Security Project (OWASP) para mitigar riscos de segurança em modelos de IA, como vazamento de dados e ataques adversariais. O PREVE_pet obteve acurácia superior a 90% na classificação de matérias, reduzindo erros humanos e tempo de análise.

    Os resultados indicam que o sistema é eficaz, seguro e escalável para a automação de fluxos documentais jurídicos.

• Referencias bibliográficas
  • Ahmed, S., & O’Connor, L. (2022). Optimizing legal document classification using data augmentation and transformers. Legal Analytics Quarterly,...
  • Brown, N., & Singh, P. (2023). Evaluating the impact of pre-processing on legal text analysis with transformers. Legal Data Science Journal,...
  • Conneau, A., Kiela, D., Schwenk, H., Barrault, L., & Bordes, A. (2017). Supervised learning of universal sentence representations from...
  • Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). Bert: Pre-training of deep bidirectional transformers for language understanding....
  • Fernandez, C., & Huang, L. (2021). Cosine similarity in legal nlp applications: From concept to practice. AI for Legal Practice, 11(4),...
  • Foundation, O. (2025). Owasp machine learning security top 10. Retrieved from https://owasp.org/ www-project-machine-learning-security-top-10/
  • Garcia, L., & Martin, E. (2023). Automated classification of legal documents using finetuned language models. International Journal of...
  • Goldberg, Y. (2017). Neural network methods for natural language processing. Morgan Claypool Publishers.
  • Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT Press. Green, I., & Cooper, H. (2021). Scalable legal text...
  • Howard, J., & Ruder, S. (2018). Universal language model fine-tuning for text classification. Proceedings of the 56th Annual Meeting of...
  • Jaccard, P. (1901). Étude comparative de la distribution florale dans une portion des alpes et des jura. Bulletin de la Société Vaudoise des...
  • Johnson, E., & Carter, S. (2023). Deep learning approaches for semantic similarity in legal case matching.
  • Kim, D., & Zhang, W. (2021). Enhancing legal text summarization using transformers. Proceedings of LegalNLP, 45–52. DOI: 10.1145/1234567.1234568
  • Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., ... Stoyanov, V. (2019). Roberta: A robustly optimized bert pretraining approach....
  • Manning, C. D., Raghavan, P., & Schütze, H. (2008). Introduction to information retrieval. Cambridge University Press.
  • Manzini, S. (2009). Classification of legal documents using machine learning techniques. Journal of Legal Analysis.
  • Martin, L., & Rodrigues, F. (2017). Text classification in legal applications: A comprehensive review. In Proceedings of the international...
  • Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., & Dean, J. (2013). Distributed representations of words and phrases and their compositionality....
  • Miller, A., & Brown, L. (2022). Legal-bert: Pre-trained transformers for legal nlp tasks. AI Law, 30(4), 123–145. DOI: 10.1007/s10506-022-09332-y
  • Open Web Application Security Project (OWASP). (2023). Owasp machine learning security top 10 [Computer software manual].
  • Patel, R., & Nakamura, H. (2022). Leveraging keyword extraction for enhanced legal document categorization. Law and AI Journal, 5(3),...
  • Radford, A., Narasimhan, K., Salimans, T., & Sutskever, I. (2018). Improving language understanding by generative pre-training. In Openai...
  • Ruder, S., Peters, M. E., Swayamdipta, S., & Wolf, T. (2019). Transfer learning in natural language processing. Proceedings of the 2020...
  • Shavlik, J. W. (1994). Combining symbolic and neural learning to refine certainty-factor rules. Machine Learning, 11.
  • Smith, J., & Doe, J. (2023). Transformers in legal text analysis: A survey. Journal of Legal AI Research, 10(3), 234–256. DOI: 10.1016/j.legalai.2023.02.003
  • Taylor, M., & Roberts, E. (2022). Ai-driven automation in legal workflows: Challenges and opportunities. Journal of Legal Innovation,...
  • Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... Polosukhin, I. (2017). Attention is all you need. Advances...
  • Walker, R., & Gomez, V. (2022). Multi-task learning for legal text classification and summarization. Legal Artificial Intelligence Review,...
  • Xu, J., Wagner, L., et al. (2020). Legalbert: Pretrained transformers for legal text understanding. In Proceedings of the 2020 international...
  • Zhang, X., Zhao, J., & LeCun, Y. (2015). Character-level convolutional networks for text classification. Advances in Neural Information...
  • Zhu, X., & Li, H. (2018). Bert-based document classification in legal contexts. Journal of Legal Informatics.