The Research Of Chinese Named Entity Recognition Based On Deep Learning

Named entity recognition is a key and fundamental task in natural language processing.Its task is to identify entities in the text to be processed that represent specific practical meanings.Named entity recognition is the key to high-level tasks such as relation extraction and event extraction,and it is also the cornerstone of applications such as text classification and question answering systems.The accuracy of named entity recognition will directly affect the effectiveness of follow-up work.Traditional named entity recognition methods require a lot of manual annotation and feature extraction.For this defect,we explore a new method for automating the identification of named entities.Through unsupervised training,a distributed feature model is obtained,and a feature model is used to obtain a word vector.The distributed vectors with additional features are input into the deep features of the B-LSTMs neural network to find the words,and finally the classifier outputs the named entity classes.The B-LSTMs network model trained by large corpus performs the best F value of 92.47% for the named entity recognition task.Experimental results show that this method integrates contextual information and other factors,and has a good effect.For the problem that the B-LSTM network method needs to sequentially read the entire text and can only be trained in a single thread,a named entity recognition model that takes both CNN parallelization and RNN context understanding into account is proposed.The model is based on Iterative Hole Convolutio nal Network(ID-CNN),which can accelerate the parallelization in the GPU environment.In the case of loss of the acceptable range of F values,the training speed of the model is about three times higher than that of the B-LSTMs model,and the best F value of the named entity recognition task is 90.82%.Experimental results show that the model has faster response speed,higher accuracy and recall rate.In this paper,B-LSTM and ID-CNN are used to extend the neural network language model,and a new type of deep model with slightly better performance than other methods is proposed,which has a certain use value.This study provides new ideas for solving other types of labeling problems.