Estudo de aplicabilidade de técnicas de processamento de linguagem natural em petições iniciais

Abstract

This study aimed to investigate the application of natural language processing (NLP) algorithms in the classification of initial petitions in the Brazilian legal system. The Doc2Vec and BERT algorithms were explored to analyze their effectiveness and compare their performances. The methodology involved data preprocessing, model training, and evaluation stages. The results were assessed using metrics such as precision, accuracy, and confusion matrix. The BERT model demonstrated an 85% precision in document classification, while the Doc2Vec model showed moderate results. The importance of data preprocessing and the influence of dimensionality were also discussed. Additionally, the need for adequate computational resources for analyzing large language models was emphasized. These findings contribute to understanding the performance of NLP algorithms in classifying initial petitions and can assist in selecting and implementing suitable models in various legal contexts.