Inteligência artificial para classificação dos cumprimentos de sentenças em ações coletivas no Tribunal Regional Federal da 5ª Região

Abstract

Artificial intelligence-based solutions have become increasingly essential within the Brazilian Judiciary, particularly in light of the challenges posed by procedural congestion, which undermines the effectiveness of the fundamental principles of legal certainty, access to justice, and the reasonable duration of proceedings. In this context, the present research focused on the enforcement of judgments in class actions before the Federal Regional Court of the 5th Region (TRF5). The choice to focus on class actions is justified by the growing complexity of these cases, the large number of beneficiaries involved, the difficulty in identifying such actions within judiciais systems, and their strategic importance in promoting procedural economy, equality between the parties, and democratic participation in public administration. Moreover, there is both a pressing need and significant difficulty in monitoring their specific performance within the TRF5. Moreover, it is important to emphasize that ensuring the classification and monitoring of these claims is essential to guaranteeing access to justice, as it enables the effective recognition and enforcement of collective rights, thereby expanding legal protection for affected groups and communities. An inductive method was adopted, starting from the empirical observation of concrete cases to formulate theoretical generalizations, in line with an applied empirical approach that combined both qualitative and quantitative analysis. This approach was grounded in a literature review, document analysis, and the practical application of the tool developed. The general objective of this study is to verify how a machine learning model, using the Positive-Unlabeled (PU) Learning technique to detect patterns in appellate court decisions, could enable the automated classification of judgment enforcement proceedings in class actions within the Federal Regional Court of the 5th Region (TRF5), thereby contributing to the realization of the principle of access to justice. The research question guiding this study is: “How can a machine learning model, using the Positive-Unlabeled (PU) Learning technique, be applied to the identification of patterns in second-instance judicial decisions to enable the automated classification of judgment enforcement proceedings in class actions before the TRF5, contributing to the realization of the principle of access to justice?” Given the difficulty in defining what constitutes a judgment enforcement proceeding in class actions, and the need to obtain data to assess the incidence of such cases within the TRF5, it became necessary to manually label a dataset of 3,000 cases. These cases were categorized into two classes: those that qualify as judgment enforcement proceedings in class actions and those that do not fall under this definition. This manual labeling process was essential for constructing a reference dataset required for the supervised training of the classification model. The data were extracted via the Júlia System API, which exclusively provides access to second-instance decisions, thereby enabling the training and evaluation of the classifier. The results demonstrated the technical feasibility of automating the screening of cases, yielding promising outcomes for the automated classification of judgment enforcement proceedings in class actions. This stage is therefore crucial both for identifying such claims and for verifying their incidence within the TRF5, especially considering the challenges of recognition within judicial systems. The automation of classification represents a significant advancement in the organization and management of procedural information, offering the potential to improve the speed of judicial proceedings when implemented. It is concluded that the application of artificial intelligence techniques in this context not only has the capacity to accelerate the screening of cases, thereby directly contributing to the processing of collective claims, but also to strengthen the fundamental principle of access to justice. Furthermore, the systematic classification of this specific type of case provides valuable inputs for both regulatory and managerial improvements within the TRF5, while also opening new avenues for future research on the use of technological solutions in the Judiciary.