Large-Scale Multi-Label Text Classification on EU Legislation
This addresses the problem of automated legal document categorization for EU policymakers and researchers, though it is incremental with improvements over existing methods.
The paper tackled large-scale multi-label text classification on EU legislation by releasing a new dataset of 57k documents with 4.3k labels and showing that fine-tuned BERT, with strategies to handle text length limits, achieved the best results in most cases.
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.