Abordagens de processamento de linguagem natural e aprendizado profundo para classificação de atos administrativos de diário oficial

Abstract

This study investigated the application of Natural Language Processing (NLP) and Deep Learning (DL) in the automation of processes within public institutions, aiming to improve efficiency and data-driven decision-making. Information overload and the shift toward digital processes have posed challenges, among them the resistance of public officials to adopting digital technologies, as they continue relying on manual methods for repetitive tasks. The objective of this research was to implement models employing NLP and DL techniques capable of classifying relevant information found in administrative acts extracted from Portable Document Format (PDF) documents published in official gazettes. The theoretical framework drew upon Information Systems (IS) concepts to examine the challenges and opportunities in integrating these technologies into public administration. The methodology included a literature review, exploratory analysis, and experimental procedures. Supervised Deep Learning algorithms were used to develop models that classify textual information after a data collection and preprocessing phase from public portals. This descriptive and quantitative research approach enabled an assessment of the developed models’ efficiency. The results showed that the Bidirectional Encoder Representations from Transformers (BERT) model achieved a 99% accuracy rate, outperforming models previously described in the literature, and proving effective for extracting and classifying relevant information. The conclusion is that the application of NLP and DL technologies significantly contributes to process automation and improved decision-making, exerting a positive impact on both public administration and the academic community. This work emphasizes the importance of adopting digital technologies to enhance the efficiency and quality of public services.