A classification approach for estimating subjects of bills in the Brazilian Chamber of Deputies

• Autor(a) principal: Nunes, Rafael Oleques
• Data de Publicação: 2023
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: eng
• Título da fonte: Repositório Institucional da UFRGS
• Texto Completo: http://hdl.handle.net/10183/267612
• topic: Mineração de texto; Análise de dados; Inteligência artificial; Multi-label classification; Legislative documents classification; Language models

A political-legal environment usually involves many documents and stages regarding laws and their processing route. Due to this large volume of data, a considerable amount of essential data, such as subject classification, keywords, and summary, is often missing for bills that are proposed. This issue increases the gap between citizens and politics, negatively affecting society. Considering the Brazilian Chamber of Deputies from 1991 to 2022, around 75% of the bills do not have subject classification included in their associated metadata. However, thanks to many bills in the corpus, this scenario suits machine learning and natural language processing approaches. This study proposes a new method for estimating subjects for the Brazilian Chamber of Deputies’ bills. Our solution presents and compares two BERT models adapted for the Portuguese language using the summary information, referring to a brief description or overview of the main points of a political document. We obtained our best results using the BERTimbau model variation, achieving 78.94% of the weighted F1 score and 72.78% of the macro F1 score. To the best of our knowledge, this is the first work to propose a model for predicting the subjects of the Brazilian Chamber of Deputies’ bills. Our approach encourages researchers to explore similar techniques for other legal documents. Our findings help political scientists perform a more robust data analysis, which was not possible with the previous data, directly impacting society.