Research And Application On Named Entity Recognition Based On LSTM

With the rapid development of intelligent technologies such as data mining and cloud services,a large amount of data information is growing in the Internet,which contains a large amount of text data.These huge amounts of text data contain great social value.However,since most of the text information in daily life exists in an unstructured form,the computer cannot efficiently process the data,and thus the extracted valuable information is very small.Named entity recognition is an intelligent information processing technology that can extract structured information from unstructured text,and plays a key role in the field of natural language processing.The research content of this paper is mainly to explore the performance of named entity recognition model,focusing on chinese corpus-oriented named entity recognition technology.The research work in this paper mainly includes:(1)We design and implement a chinese named entity recognition model based on Bi LSTM-QRNN-CRF.By researching and analyzing the long short-term memory network(LSTM)model,bidirectional long short-term memory network model(Bi LSTM)and bidirectional long short-term memory network-conditional random field model(Bi LSTM-CRF),we design a chinese named entity recognition model based on Bi LSTM-QRNN-CRF.This model enriches the feature information of the input data and further improves the recognition effect.This method uses the most widely used and mature Bi LSTM-CRF model as the basic model,and we use quasi-recurrent neural network(QRNN)to enhance the feature information of the model input,and because the QRNN model has the characteristics of parallel computing,the training time has not been increased.Experimental results show that the recognition effect of the model has been improved to a certain extent.(2)We design and implement a chinese named entity recognition model based on tri-training for multiple neural networks.With the increasing application of named entity recognition technology,the types of objects to be identified are more and more diverse,and the need for named entity recognition for specific fields is also increasing.The chinese named entity recognition model based on tri-training for multiple neural networks overcomes the problem that only a small amount or even no labeled corpus is available in a specific field,and we aims to improve the system practicality of named entity recognition.This method combines the advantages of neural networks and tri-training.Firstly,three different neural networks are trained as basic classifiers with a small amount of labeled data,and then the three basic classifiers are trained on a large amount of unlabeled data to optimize the model.Experimental results show that the model has a good recognition effect.(3)We design and implement an interface-based named entity recognition system,and the system was developed by the already-implemented chinese named entity recognition model.By designing and implementing the functions of registration,login and file upload,the practicality of chinese named entity recognition is improved.