This repository contains the code to implement the paper Pseudonymization in Legal Texts according to the LGPD: A Named Entity Recognition Approach.
The code implemented in the paper used Llama version 3. The code in this repository uses Llama version 3.1 and has some minors changes.
@INPROCEEDINGS{242811,
AUTHOR="Marcelo Filho and Bruno Cesar Ribas",
TITLE="Pseudonymization in Legal Texts according to the LGPD: A Named Entity Recognition Approach",
BOOKTITLE="BRACIS 2024 () ",
ADDRESS="",
DAYS="23-21",
MONTH="may",
YEAR="2024",
ABSTRACT="This study explores the application of Named Entity Recognition (NER) for the pseudonymization of data in legal texts, aiming to protect Personally Identifiable Information (PII) in compliance with Brazil's General Data Protection Law (LGPD). The research highlights the challenge of balancing data privacy and utility, presenting a methodology that uses artificial intelligence technologies to effectively identify and obscure PII in legal documents. In this study, we propose a Transformer model along with Regex techniques to identify entities in a text. To test the model, we created a new dataset from the existing LenerBR. We also used a function to generate synthetic data and prompt engineering applied to the Llama 8B version 3 model. Tests showed the need for further adjustments in the proposed new model. Future work will focus on improving the model's accuracy and efficiency, as well as enhancing the identification of sensitive data and learning from user interactions.",
KEYWORDS="- Natural Language Processing; - Pattern Recognition and Cluster Analysis; - Large Language Models; - Generative AI",
URL="http://XXXXX/242811.pdf"
}