🇪🇺 🏷️ EuroVoc dataset

This dataset contains 5,820,634 documents with their associated EuroVoc labels.

EuroVoc is a large multidisciplinary hierarchical thesaurus of 7,049 classes covering the activities of EU institutions.

Why was this dataset created ?

"Extreme classification come with challenges of scalability due to large label spaces, data sparsity issues due to insufficient training samples."

https://medium.com/datapy-ai/extreme-multi-label-classification-for-eurovoc-b51d74623820

How was dataset this created ?

The source code is available, check update.py

When this dataset was created ?

20 November 2025

What are the main characteristics of this dataset ?

There are a total of 83 different languages present in this dataset, of which some are EU languages and some are not. As the following graph illustrates, most of the documents of the dataset are written in EU languages (English being the most present language in the dataset), and the non-EU languages are very poorly represented (for example Arabic, Japanese,...). Note that since the Irish language (gle) was granted full official and working status in the EU in 2022, there are very few documents in that language. Additionally, Croatian (hrv) is also less represented in the dataset as Croatia is the latest country to have joined the EU in 2013.

The lengths of the documents also varies depending on the language it is written in. Note that this boxplot does not present the outliers, since the length of certain documents can contain up to almost 100 million characters. The red lines in the boxplot indicates the median length of the documents for each language.

How is the data structured ?

An example of a sample of this dataset is the following :

{"title": "Directive 96/9/EC of ...",
"date": "1996-03-11",
"eurovoc_concepts": ["copyright", "data protection", "database", "free competition", "national implementing measure"],
"eurovoc_concepts_ids": ["529", "5181", "4821", "1635", "5739"],
"url": "http://publications.europa.eu/resource/cellar/b48976b5-b31a-4dba-a052-87133e17d65e",
"lang": "eng",
"formats": ["fmx4", "html", "pdfa1b", "print", "xhtml"], 
"text": "DIRECTIVE 96/9/EC OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL of 11 March 1996 on the legal protection of databases ..."}

• title : title of the document
• date : publication date of the document
• eurovoc_concepts : list of the EuroVoc concept labels related to this document
• eurovoc_concepts_ids : list of the EuroVoc concept IDs related to this document
• url : URL to access the document
• formats : list of formats in which the original document is available
• text : text content of the document

Bibliography

• Ilias Chalkidis, Emmanouil Fergadiotis, Prodromos Malakasiotis, Nikolaos Aletras, and Ion Androutsopoulos. 2019. Extreme Multi-Label Legal Text Classification: A Case Study in EU Legislation. In Proceedings of the Natural Legal Language Processing Workshop 2019, pages 78–87, Minneapolis, Minnesota. Association for Computational Linguistics.
• I. Chalkidis, M. Fergadiotis, P. Malakasiotis and I. Androutsopoulos, Large-Scale Multi-Label Text Classification on EU Legislation. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL 2019), Florence, Italy, (short papers), 2019.
• Andrei-Marius Avram, Vasile Pais, and Dan Ioan Tufis. 2021. PyEuroVoc: A Tool for Multilingual Legal Document Classification with EuroVoc Descriptors. In Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021), pages 92–101, Held Online. INCOMA Ltd..
• SHAHEEN, Zein, WOHLGENANNT, Gerhard, et FILTZ, Erwin. Large scale legal text classification using transformer models. arXiv preprint arXiv:2010.12871, 2020.

Author(s)

Sébastien Campion sebastien.campion@europarl.europa.eu

Andreas Papagiannis andreas.papagiannis@europarl.europa.eu