Research On Named Entity Recognition Method On Deep Learning

With the continuous growth of the text size on the Internet,the semantic information contained in the text has become more and more abundant.How to obtain useful structured knowledge from these massive text data has become a hot research topic.Named Entity Recognition(NER),as a popular research direction in natural language processing,has been widely used in various natural language processing tasks such as relation extraction,question answering systems,and machine translation.This article mainly focuses on the problems in named entity recognition,and conducts related research.First,most of the current Chinese named entity recognition methods mostly use single-level features for recognition,and do not consider both character and word level features at the same time,and it is difficult to obtain sufficient character shape information and word meaning information.Therefore,the effective combination of glyph information and word meaning information has important practical significance for the Chinese named entity recognition method.Second,in the field of named entities,the model of recurrent neural network is mostly used.Because the recursive neural network is mainly recursive in the calculation process,the time cost is relatively high.In comparison,the convolutional neural network is a feedforward neural network and has a small time cost,but the convolutional neural network is not good at handling time series tasks.Therefore,only one type of network is used to realize the functions of two types of networks,which has important practical significance.This article focuses on the neural network-based named entity recognition method.The main content of the thesis has the following three aspects:(1)For most Chinese named entity recognition methods,most of the characteristics of a single level are identified,and there is no simultaneous considering word and word level characteristics,it is difficult to obtain sufficient glyph information and word sense information.In order to explore the word level and the word level characteristics,it improves only a single model recognition effect based on word or word-level characteristics,and presents a Chinese name entity identification method based on multi-level feature sensing networks.Finally,the MSRA and the "People's Daily" data set are widely conducted.At the same time,it is compared to the Chinese named entity recognition method in recent years.The experimental results show that the method identified by the MSRA and RESUME data concentration reached 92.15% and94.32%,respectively,and the F1 of the Name,the name and the institution name is 94.28%,94.17% and 90.33,respectively.The overall Chinese named entity recognition method is generally superior in recent years.(2)In the field of naming entity,the model of circulating neural network is used in the field,due to the main recursive process during the calculation process,the time cost is large.In contrast,convolutional neural networks are small in the feedforward neural network,but the convolutional neural network is not good at handling time series tasks.In response to the above,this paper constructs a multi-channel void self-focus convolution network,thereby better "replacing" circulating neural networks.The experimental results show that the methods mentioned herein are generally better than the mainstream named entity recognition method in the two English data sets in Co NLL-2003 English NER and Onto Notes5.0.(3)Nowadays,structured electronic medical records have gradually become the mainstream method for various medical structures to collect patient information,but before this,hospitals still retain many unstructured medical records.Since we are still unable to directly retrieve from the massive unstructured medical records,the massive medical records cannot play a vital role.Therefore,how to fully excavate the effective information from these massive unstructured cases has become a difficult problem in the field of modern medicine.In recent years,with the rapid development of computer technology,the combination of natural language processing and electronic medical records has become a hot issue.The most basic link is to extract the entities in the electronic medical record.The text applies the Chinese named entity recognition method based on multi-level feature perception network to the actual application of entity recognition in electronic medical records.This article first introduces a named entity recognition system for Chinese electronic medical records,which is mainly used to obtain some unstructured Chinese electronic medical records,enter these texts into the trained model,and finally display the recognition in the text,and follow them according to the entity Category storage is convenient for constructing structured electronic medical records in the future.