EuroVoc: 5.8 Million EU Documents with Multidisciplinary Thesaurus Labels

5,820,634 documents are labeled with the EuroVoc thesaurus, a hierarchical system of 7,049 classes covering EU activities. The dataset was created by the European Parliament to address challenges in extreme classification, such as scalability and data sparsity. It was last updated on December 1, 2025.

Use Cases

• Training models for extreme multi-label text classification based on a large label space of 7,049 classes.
• Benchmarking hierarchical classification algorithms on a large-scale, real-world document corpus.
• Developing domain-specific NLP applications for EU policy, legal, or administrative text analysis.
• Studying data sparsity and scalability challenges in large-scale machine learning tasks.

Strengths

• Large scale with 5,820,634 documents.
• Structured around a well-defined, multidisciplinary thesaurus of 7,049 hierarchical classes.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but other details like file formats and sample data are unavailable.
• Freshness should be verified as the last update timestamp is from the future (2025-12-01).

Provenance

Source

EuropeanParliament

Freshness

Last updated 2025-12-01 14:31:12

Geography

European Union institutions