Research on Named Entity Recognition Based on ERNIE-BiLSTM-CRF

Named entity recognition, as well as the knowledge extraction of text data, is one of the important tasks of information mining. However, due to the unique text structure and word distribution of some specific sentences, the traditional models cannot well inject the related entities into the exact representation. To better represent entities by word vectors to improve the annotation performance of named entities of the model, this paper proposes an ERNIE-BiLSTM-CRF model. Firstly, the dynamic word vectors of text input are obtained by ERNIE pre-trained model, and then the corresponding feature vectors are acquired through BiLSTM, and finally, the features are decoded and the label sequence is calculated by CRF. The experiment result shows that the Precision, Recall, and F1-score of the proposed model are 69.74%, 66.55%, and 68.10% respectively, which proves the effectiveness of the method.