Classificação de processos judiciais da Procuradoria Geral da União utilizando algoritmos de aprendizado de máquinauma análise comparativa da abordagem clássica e GPT-4

Resumo

Purpose: The study aims to investigate ways to improve the distribution and classification of judicial cases and initial petitions at the Attorney General's Office (AGU), leveraging machine learning (ML) techniques to increase the accuracy in allocating these cases to the corresponding judicial sectors. Methodology: A quantitative approach is adopted, using computational and experimental methods to test and explore a variety of ML and natural language processing (NLP) techniques in legal documents. The research focuses on conducting a comparative analysis between different artificial intelligence pipelines, from traditional ML models to recent innovations such as the Generative Pre-trained Transformer (GPT). Findings: Significant limitations were identified in the GPT-4 model with Zero-Shot Learning in terms of the volume of data that can be processed compared to traditional classification methods. This highlights a barrier in the effectiveness of this particular approach. Resarch Limitations: The study acknowledges operational restrictions, including the limited capacity of the GPT-4 model to process large volumes of data, which suggests the need for future research to expand the training scope to data from other regions of the AGU. An expansion of the training scope to data from other AGU regions and the exploration of fine-tuning strategies in pre-trained models are suggested, aiming to enhance the applicability and effectiveness of ML solutions in the legal domain. Practical implications: The research aspires to apply machine learning techniques to increase the effectiveness in the automatic distribution of cases at PRU1. This research problem is relevant from both a technological and operational perspective, as it involves integrating technological solutions into the processing of judicial data, with the goal of improving management and the quality of workflow at the AGU, with the potential to significantly transform current practices and contribute to digital transformation within the agency. Social implications: This study contributes to the digital transformation of the justice system, providing valuable insights for the continuous improvement of case management and reinforcing the commitment to modernization and efficiency of judicial services. Originality: The research introduces an innovative approach by combining ML techniques with the analysis of legal documents to enhance the distribution and classification of cases at the AGU, marking an original contribution to the field of legal technology. Keywords: Judicial Process Distribution, Traditional Machine Learning Models, Natural Language Processing, GPT-4, Zero-Shot Learning, Digital Transformation

Ficha do documento

Tipo

Dissertação

Ano

2024

Instituição

Fundação Getulio Vargas

Fonte

Repositório da FGV

Idioma

Português

Acesso

Acesso aberto

Identificador

oai:repositorio.fgv.br:10438/35220

Temas

Administração PúblicaTecnologia

Palavras-chave

Distribuição de Processos JudiciaisModelos Tradicionais de Aprendizado de MáquinaProcessamento de Linguagem NaturalGPT-4Zero-Shot LearningTransformação Digital no JudiciárioAprendizado do computadorInteligência artificialProcessamento da linguagem natural (Computação)Administração pública - Automação - Brasil