README.md

metadata

configs:
  - config_name: default
    data_files:
      - split: train
        path:
          - first section/first_section_train.csv
          - second section/second_section_train.csv
          - third section/third_section_train.csv
          - forth section/forth_section_train.csv
          - fifth section/fifth_section_train.csv
          - sixth section/sixth_section_train.csv
          - seventh section/seventh_section_train.csv
          - contentious/contencioso_section_train.csv
      - split: test
        path:
          - first section/first_section_test.csv
          - second section/second_section_test.csv
          - third section/third_section_test.csv
          - forth section/forth_section_test.csv
          - fifth section/fifth_section_test.csv
          - sixth section/sixth_section_test.csv
          - seventh section/seventh_section_test.csv
          - contentious/contencioso_section_test.csv
license: mit
task_categories:
  - token-classification
language:
  - pt
tags:
  - legal
size_categories:
  - 1K<n<10K

Work developed as part of [IRIS] (https://www.inesc-id.pt/projects/PR07005/)

Extreme Multi-Label Classification of Descriptors

The goal of this dataset is to train an Extreme Multi-Label classifier that, given a judgment from the Supreme Court of Justice of Portugal (STJ), can associate relevant descriptors to the judgment.

Dataset Contents:

• Judgment ID: Unique identifier for each judgment.
• STJ Section: The section of the STJ to which the judgment belongs.
• Judgment Text: Full text of the judgment.
• Descriptors List: A list of binary values (0's and 1's) where 1's indicate the presence of active descriptors.

The dataset is organized by the sections of the STJ, and each section is further divided into training and testing subsets.

Additional Files:

• label.py: A Python file containing lists of descriptor names for each section. The order of these lists corresponds to the order of 0's and 1's in the dataset.

In each section, except for the contentious section, there is an additional file containing judgments excluded from the dataset. In these files, the judgments after the removal of descriptors appearing only once or twice were left with no descriptors.

Code: https://github.com/MartimZanatti/Extreme_multi_label_Portuguese_Supreme_Judgments

Contributions

Paper: Sleec Assignment of Descriptors to Judgments of the Supreme Court of Justice of Portugal

link: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5017021