Research On Named Entity Recognition Based On Deep Learning

Named Entity Recognition(NER)is an important basis for Relationship Extraction,Entity Linking,Machine Translation and other tasks in Natural Language Processing(NLP).In recent years,with the huge amount of data generated by various industries,new requirements have been put forward for the accuracy and applicability of Named Entity Recognition technology.The traditional NER method needs to spend a lot of time designing the feature manually.The quality of feature engineering directly affects the final performance of the model.In recent years,deep learning model has been used to replace the artificial construction of features by human,and it has achieved good performance in some public data sets.The main work of this paper is as follows:1.In view of the English NER,the paper uses a module called SATT-Parallel RNNs.The module is based on the Parallel RNNs model and adds a Self-attention mechanism.Finally,a new NER architecture Self-Attention + Parallel RNNs(SAPR)is proposed.Through experiments,the F1 value on the public Co NLL 2003 English NER dataset reaches 93.76%,which has achieved optimal performance without the need for additional external resources.2.The difference between Chinese and English NER is analyzed.Inspired by the English NER method,the hybrid word embedding method is used for the Chinese NER task,and the multiple Attention mechanism is added,and a Chinese NER model based on CWSA+ Bi LSTM-CRF is proposed.Finally,the experiment shows that the F1 value of the model reaches 91.23% on MSRA and 73.23% on Literature NER,both of which achieve good results.This paper expands the task of NER in English and Chinese,proposes two new models and achieves the optimal performance in public data,which is of certain applicable value,and also provides new ideas for the same type of sequence labeling problems.