Banca de DEFESA: FERNANDA GUEDES QUEIROZ DE LIRA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : FERNANDA GUEDES QUEIROZ DE LIRA
DATE: 23/07/2025
TIME: 15:00
LOCAL: Ambiente virtual ( https://meet.google.com/pre-mbya-dvc )
TITLE:

ARTIFICIAL INTELLIGENCE FOR THE CLASSIFICATION OF SENTENCE ENFORCEMENT IN CLASS ACTIONS BEFORE THE FEDERAL REGIONAL COURT OF THE 5TH REGION

KEY WORDS:

Artificial Intelligence; Access to Justice; Reasonable Duration of Proceedings; Judgment Enforcement; Collective Actions.

PAGES: 142
BIG AREA: Ciências Sociais Aplicadas
AREA: Direito
SUMMARY:

Artificial intelligence (AI)-based solutions have become increasingly essential in the Brazilian Judiciary, particularly due to the challenges posed by procedural backlog, which undermines the effectiveness of fundamental principles such as legal certainty, access to justice, and reasonable duration of proceedings. This study focuses on the enforcement of collective action judgments at the Federal Regional Court of the 5th Region (TRF5).The selection of collective actions is justified by the increasing complexity of these demands, the large number of beneficiaries, the difficulty of identifying such cases within judicial systems, and their strategic importance in promoting procedural efficiency, equality among the parties, and democratic participation in public administration. Furthermore, ensuring correct classification and effective monitoring of these demands is fundamental to guaranteeing access to justice, as it allows for the effective recognition and enforcement of collective rights, thereby expanding legal protection for affected groups and communities.An inductive research approach was adopted, beginning with empirical observations of specific cases and progressing towards theoretical generalizations. This was complemented by an applied empirical methodology combining qualitative and quantitative analyses, supported by bibliographic review, documentary analysis, and practical application of the developed tool.The main objective of this study was to develop a machine learning model using the Positive-Unlabeled (PU) Learning technique to identify patterns in judicial decisions and enable automated classification of enforcement proceedings related to collective action judgments at TRF5. The guiding research question was: “How can a machine learning solution based on Positive-Unlabeled Learning assist in the automated identification and classification of enforcement proceedings of collective action judgments at TRF5, based on second-instance decisions?”Given the difficulty of precisely defining what constitutes enforcement of collective action judgments and the need for reliable data to assess their incidence at TRF5, a manual labeling process was conducted on a dataset of 3,000 cases. These cases were categorized into two classes: those that qualify as enforcement of collective action judgments and those that do not. This manual annotation was critical for constructing a reference dataset necessary for supervised training of the classification model. Data extraction was performed through the Júlia System API, which provides exclusively second-instance decisions, enabling both the training and evaluation of the classifier using real and representative data.Results demonstrated the technical feasibility of automating case triage and underscored the effectiveness of automated classification for enforcement proceedings in collective actions. This step is crucial both for identifying such demands and for verifying their incidence within TRF5, given the challenges in recognizing them through judicial information systems. The automation of classification represents a significant advancement in organizing and managing procedural information, yielding considerable improvements in processing speed, predictability, and quality of case management.In conclusion, the application of AI techniques in this context not only accelerates the processing of collective demands but also reinforces the pillars of access to justice and judicial effectiveness. Moreover, systematic classification of this specific case type provides practical insights for normative and managerial improvements at TRF5 while opening new avenues for future research on technological solutions within the Judiciary.

 

COMMITTEE MEMBERS:
Presidente - 2353000 - ELIAS JACOB DE MENEZES NETO
Interno - 1254860 - FABRICIO GERMANO ALVES
Externo à Instituição - LUCIANA GROSS SIQUEIRA CUNHA
Interno - 1358062 - MARCO BRUNO MIRANDA CLEMENTINO