Large-Scale Multi-Label Text Classification on EU Legislation

We consider Large-Scale Multi-Label Text Classification (LMTC) in the legal domain. We release a new dataset of 57k legislative documents from EURLEX, annotated with ~4.3k EUROVOC labels, which is suitable for LMTC, few- and zero-shot learning. Experimenting with several neural classifiers, we show that BIGRUs with label-wise attention perform better than other current state of the art methods. Domain-specific WORD2VEC and context-sensitive ELMO embeddings further improve performance. We also find that considering only particular zones of the documents is sufficient. This allows us to bypass BERT's maximum text length limit and fine-tune BERT, obtaining the best results in all but zero-shot learning cases.

PDF Abstract ACL 2019 PDF ACL 2019 Abstract

Datasets


Introduced in the Paper:

EURLEX57K

Used in the Paper:

RCV1
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Multi-Label Text Classification EUR-Lex bert-base nDCG@5 82.3 # 1
P@5 68.7 # 1
Micro F1 73.2 # 1
RP@5 79.6 # 1

Methods