Research Of Chinese Named Entity Recognition Based On Deep Neural Network

Named entity recognition is one of the key technologies in the field of natural language processing,and it is also the basic work of information extraction,intelligent question and answer,machine translation and other applications.With the rapid growth of the Internet industry,the commercial value of text data is getting higher and higher.In order to accurately mine information in massive texts,named entity recognition technology has become a focus of researchers.The main task of naming entity recognition is to identify the name,place and organization of the text.It is found that traditional statistical learning-based named entity recognition methods require manual features,and feature engineering has a great influence on model.In order to weaken the dependence of the model on the artificial features,our article introduces the current popular deep learning sequence annotation model—BiLSTM-CRF model as a baseline model.And we improved this model to make it better applied to Chinese named entity recognition task.The main work of this article is as follows:(1)In order to facilitate the input of text sequences,the word2 vec model is introduced to convert Chinese characters into multidimensional vector and input models.We introduce the convolutional neural network into the BiLSTM-CRF model,and propose a Chinese named entity recognition model based on BiLSTM-CNN-CRF.The model can extract the spatial feature of text sequences effectively.Finally,experiments show that the BiLSTM-CNN-CRF model can achieve good results in the RMRB corpus,especially the recall and F-value are 2.07% and 0.86% higher than the BiLSTM-CRF model.(2)We introduce the attention mechanism into the BiLSTM-CRF model,and propose a Chinese named entity recognition model based on Att-BiLSTM-CRF.The attention mechanism combine with the training of neural network can highlight the keywords in the text to improve the annotation effect.Finally,experiments show that the accuracy,recall,and F-value of the Att-BiLSTM-CRF model are respectively 0.13%,2.03%,and 0.71% higher than the BiLSTM-CRF model.(3)Combining convolutional neural network and attention mechanism,we proposed a Chinese named entity recognition model based on Att-BiLSTM-CNN-CRF.Experiments show that the accuracy,recall and F value of the model are improved by 0.9%,1.77% and 1.31% respectively compared with the BiLSTM-CRF model,and are 1.12%,-0.2% and 0.45% higher than theBiLSTM-CNN-CRF model.Compared with Att-BiLSTM-CRF model,they increased 0.77%,0.41%,and 0.6%,respectively.