Named Entity Recognition for Russian Judicial Rulings Text

Abstract

The article presents the solution of named entity recognition problem for legal Russian-language texts. We studied CRF, LSTM, BERT and BiLSTM and their combinations. The models were tested with various parameters of text preprocessing and words vector representations. The best result was shown by fastext vectorization with BiLSTM and CRF model, the value $F-score$ is 0.86.