Chinese Named Entity Recognition Based On Deep Learning

The rapid development of Internet make online all the time to produce a large number of text data,the text data contain a higher value,in order to from the extracted text data to valuable information,named entity recognition as one of the basic technology,its effect on subsequent recognition of semantic role labeling,information retrieval and text classification tasks will have a huge impact,Therefore,more and more scholars have paid attention to it.At present,there are some problems in Chinese named entity recognition,such as the single word vector representation in the pre-processing stage and only paying attention to the global feature extraction while ignoring the local feature extraction.This paper first proposes a Chinese named entity recognition method based on ATT-BERT to solve the single problem of word vector representation,and then proposes a Chinese named entity recognition method based on TCNN-BERT,which is the final method of this paper,and can effectively solve the above two problems.The research work of this paper mainly focuses on the following two parts:(1)Chinese named entity recognition based on ATT-BERT.Firstly,the BERT pre-trained language model is used to generate context-dependent dynamic word vectors according to the context of the text to fully represent the word vectors.Then,the obtained sequence of word vectors is input into BiLSTM,and the contextual features of the text,namely the global features,are obtained through two LSTM in different directions.Then,the attention mechanism is integrated,and different weights are given selectively according to the role of named entity recognition.The features that play an important role are given a larger weight,while irrelevant features are weakened or even ignored.Finally,input to CRF to learn the dependency between tags to optimize the whole tag sequence,obtain the global optimization of the prediction tag sequence,and realize the Chinese named entity recognition.Experimental results show that this method can achieve good recognition effect.(2)Chinese named Entity Recognition Based on TCNN-BERT.Methods(1)can solve the problem of said a single word vector,and joined the attention mechanism,enhances the ability of named entity recognition,but didn't solve the problem of ignoring local characteristics,it is because the BiLSTM can capture the global feature of the text,but can't give attention to both the local characteristics of the text,so consider joining the local characteristics of IDCNN modeling study text information.Therefore,on the basis of Method(1),this paper also proposes a Chinese named entity recognition method based on TCNN-BERT.Firstly,the dynamic word vector containing contextual semantic information is still generated by BERT model.Then,the text is input to the two-channel Neural Network(TCNN),which is composed of BiLSTM and IDCNN,to obtain the global and local features of the text.Secondly,the attention mechanism is added to divide the features into dynamic weights to mine the deep semantic information of the text.Finally,the input to CRF is used to restrict the tag sequence to make it conform to human language logic,and the prediction tag sequence with the highest probability is output to realize Chinese named entity recognition.Experimental results show that this method can achieve ideal recognition effect.Figure[18]Table[10]Reference[61]...