Named entity recognition systems have the untapped potential to extract information from legal documents, which can improve information retrieval and decision-making processes. In this paper, a dataset for named entity recognition in Brazilian legal documents is presented. Unlike other Portuguese language datasets, this dataset is composed entirely of legal documents. In addition to tags for persons, locations, time entities and organizations, the dataset contains specific tags for law and legal cases entities. To establish a set of baseline results, we first performed experiments on another Portuguese dataset: Paramopama. This evaluation demonstrate that LSTM-CRF gives results that are significantly better than those previously reported. We then retrained LSTM-CRF, on our dataset and obtained F 1 scores of 97.04% and 88.82% for Legislation and Legal case entities, respectively. These results show the viability of the proposed dataset for legal applications.

PDF Abstract

Datasets


Introduced in the Paper:

LeNER-Br

Used in the Paper:

CoNLL 2003
Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Named Entity Recognition (NER) LeNER-Br LSTM-CRF Micro F1 (Tokens) 0.9253 # 1
Micro F1 (Exact Span) 0.8661 # 1

Methods


No methods listed for this paper. Add relevant methods here