TSHR-MR/echr_facts

---
dataset_info:
features:
- name: id
dtype: int64
- name: labels
dtype: string
- name: judgment_facts
dtype: string
- name: replaced_countries
dtype: string
- name: one_hot_labels
sequence: int64
splits:
- name: train
num_bytes: 154260079
num_examples: 15689
download_size: 71010918
dataset_size: 154260079
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

# European Court of Human Rights Dataset

This dataset collects roughly 15,000 cases from the European Court of Human Rights' database - HUDOC.

It utilizes regex to scrape through judgment facts and then applies Large Language Models (LLMs) to obtain the relevant violation(s) of the European Convention on Human Rights.

The dataset contains any violation of the European Court of Human Rights that does not pertain to the Court's functioning. Instead, the violations included deal exclusively with substantive or procedural rights.

This dataset also attempts to remove paragraph numbering wherever possible to provide cleaner text for analysis.

## Columns

- **id**: Unique ID labels.
- **judgment_facts**: The text between the "Facts" heading and the next relevant heading that indicates the end of the facts section, which is variable.
- **replaced_countries**: A list of countries that have been replaced in the original text by `<COUNTRY>`.
- **one_hot_labels**: The one-hot encoded labels for the 30 violations`.

## Purpose

The purpose of this dataset is to assist in multilabel classification.

## Label to ECtHR Article Index

{'P1': 0, 'P4': 1, 'P6': 2, 'P7': 3, '1': 4, '2': 5, '3': 6, '4': 7, '5': 8, '6': 9, '7': 10, '8': 11, '9': 12, '10': 13, '11': 14, '12': 15, '13': 16, '14': 17, '15': 18, '18': 19, '25': 20, '34': 21, '37': 22, '38': 23, '39': 24, '41': 25, '44': 26, '46': 27, '50': 28, '': 29}