Chinese Named Entity Recognition for Hazard And Operability Analysis Text

摘要

To solve the problem that it is difficult to identify the key Chinese entity information in the hazard and operability analysis text, a deep neural network model based on bidirectional long short-term memory and conditional random field (BiLSTM-CRF) is proposed to identify key named entities in the text. In the word vector pre-training process, bidirectional encoder representation from transformers (BERT) model is used to pre-train word vectors instead of the static word vectors in the traditional word2vec model, and then obtain context-related dynamic word vectors, to improve the representational ability of word vectors, and solve the problem of word boundary division when word vectors are used in Chinese corpus training. This model has the ability of complete the task of Chinese named entity recognition. The F1 value on the test corpus reaches 93.31%, which is 21.82% higher than conditional random field (CRF) Baseline, and 4.49% higher than the traditional BiLSTM-CRF model. The experimental results show that the BERT-BiLSTM-CRF model is effective for the named entity recognition (NER) task of the hazard and operability analysis text, and it is helpful to automatically extract the relationship between the entities in the hazard and operability analysis text and build safety analysis knowledge graph.