| The 21st century belongs to the information age,the rapid development of Internet technology,artificial intelligence has been widely used in all walks of life,bringing great convenience to people’s life,such as algorithmic trading,information extraction,intelligent question and answer.Named entity recognition technology is the basis of natural language processing tasks,can quickly extract key information from a large number of complex structure of text data,promote the further development of various industries.In natural language processing,deep learning makes use of its powerful feature learning ability to fully mine the deep information of text and better understand the semantics of text,thus providing strong support for solving the problem of insufficient feature learning in chinese text named entity recognition task.Named entity recognition technology has achieved great success since 1991.However,there are still some problems in the existing methods:Traditional word embedding methods can not effectively solve the polysemy problem of chinese character expression,and most methods have large parameters,long time,and cannot fully extract text information,which will lead to poor entity recognition effect.Based on the above problems,this paper improved the traditional named entity recognition model,proposed a more accurate recognition method,and carried out several groups of experiments to verify the effectiveness of the proposed model.The specific work is as follows:(1)Solve the problem of polysemy in chinese text expression.At present,BilSTM-CRF,a classical named entity recognition model,uses the traditional word embedding technology to express word vector,but it cannot accurately express polysemy in different contexts with text features.In this paper,BERT model is used to replace the traditional word embedding model,which can effectively solve the problem because multi-layer transformer coding structure can express the word vector dynamically.(2)Construct the ALBERT-BilSTM-Attention-CRF model.Since BERT model training has high requirements on computational power,and the number of parameters is large,up to hundreds of millions of orders of magnitude,it takes a long time to train and even gets worse results when the number of parameters is larger.Therefore,this paper changed BERT to ALBERT model to greatly reduce the training parameters and improve the accuracy of results.In addition,the attention mechanism is added to give different weights to different influences of text context on target words,fully obtain the context association to extract potential semantic feature information,and effectively identify entities with ambiguity and complex nested structure.Then,the ALBERT-BiLSTM-Attention-CRF model is proposed.(3)Experimental verification analysis.The model presented in this paper is compared with several combined models.The accuracy rate,recall rate and fl-value were 93.75%,92.21%,92.97%and 92.01%,91.38%,91.7%on the general and professional data sets respectively.Aiming at the existing problems in chinese named entity recognition,this paper optimized and innovated the model by improving BERT and incorporating attention mechanism,which improved the model effect and achieve more accurate and efficient chinese named entity recognition. |
