Empirical Analysis of Binding Precedent Efficiency in Brazilian Supreme Court via Case Classification

Strategic Insights -

Researchers at the Getulio Vargas Foundation conducted a study on the effectiveness of binding precedents in the Brazilian Supreme Court, highlighting concerns about their ability to reduce repetitive demands. The team used a unique framework to analyze five binding precedents and found that they tend to fail in this direction, often creating new demands instead. This research has significant implications for the Brazilian legal system and has been recognized by Springer, a leading publisher in the field of Artificial Intelligence and Law.

Key Takeaways:

  • The study focused on five binding precedents (11, 14, 17, 26, and 37) at the highest Court level, analyzing their effects on the legal subjects they address.
  • The analysis involved techniques of Similar Case Retrieval, which are only possible through Case Classification.
  • The team used different methods of Natural Language Processing (NLP) for Case Classification, comparing TF-IDF, LSTM, Longformer, and regex models.
  • TF-IDF models performed slightly better than LSTM and Longformer when compared through common metrics, but deep learning models were able to detect certain important legal events that TF-IDF missed.
  • The researchers identified five main hypotheses that explain the inefficiency of binding precedents in responding to repetitive demand, which are found in different combinations in each of the precedents studied.
  • The study concluded that the reasons for binding precedents to fail in responding to repetitive demand are heterogeneous and case-dependent, making it impossible to single out a specific cause.
  • The research was supported by Conselho Nacional de Desenvolvimento Cientifico e Tecnologico (CNPQ), Fundacao Carlos Chagas Filho de Amparo a Pesquisa do Estado do Rio De Janeiro (FAPERJ), Fundacao Getulio Vargas, University of Sao Paulo (PRPI), and Coordenacao de Aperfeicoamento de Pessoal de Nivel Superior (CAPES).
  • The study has been peer-reviewed and published in the journal Artificial Intelligence and Law.

Statistics:

  • 5 binding precedents were studied at the highest Court level (11, 14, 17, 26, and 37).
  • 4 different methods of NLP were used for Case Classification: TF-IDF, LSTM, Longformer, and regex.
  • 5 main hypotheses were identified that explain the inefficiency of binding precedents in responding to repetitive demand.
  • 6 researchers contributed to the study, including Raphael Tinarrage, Henrique Ennes, Lucas Resck, Jorge Poco, Lucas T. Gomes, and Jean R. Ponciano.

Sources:

  • NewsRx. Studies from Getulio Vargas Foundation in the Area of Artificial Intelligence and Law Reported (Empirical Analysis of Binding Precedent Efficiency In Brazilian Supreme Court Via Case Classification). Robotics & Machine Learning. June 16, 2025; p 453.
  • Getulio Vargas Foundation. Empirical Analysis of Binding Precedent Efficiency In Brazilian Supreme Court Via Case Classification. Artificial Intelligence and Law, 2025.
  • Springer. Artificial Intelligence and Law. www.springer.com; www.springerlink.com/content/0924-8463/