| Entering a new era,with the advancement of medical information,the requirements for medical information processing capabilities are also increasing.Artificial intelligence technology is applied to the medical field to help medical workers improve work efficiency,discover potential medical value information,and more serve the people well and meet their health needs.Named entity recognition,as the basic step of text data processing,provides basic data services for subsequent data mining,information extraction,and knowledge graph tasks.Therefore,the research of named entity recognition technology in the medical field is of great practical significance.In the trend of continuous promotion of natural language processing technology,named entity recognition technology has attracted more and more attention,and certain achievements have been made in certain fields such as technology and finance.Nevertheless,in the medical field,due to the particularity of the medical profession and the complex and changeable characteristics of the semantic expression of the Chinese language itself,the research of named entity recognition in this field still has the following problems:(1)There is a lack of currently published medical text data,and high-quality medical texts based on annotations,such as electronic medical records,are even scarce;(2)In Chinese medical text named entity recognition,traditional word vectors have insufficient contextual information.Based on the above-mentioned practical problems,this article conducts related research on the basis of in-depth investigation of the status of named entity recognition technology applied in the medical field.The data set used in the research comes from biendata’s CHIP-2020(Entity)Chinese medical text naming recognition competition.The research work includes:(1)Analyze the defacts of machine learning related algorithms,fully explore the application of the relatively mature Bi LSM-CRF model in production practice,and analyze its shortcomings and shortcomings.Although Bi LSTM can consider the semantic information of the context,it can identify The corresponding features are found,but it is still difficult to identify the importance of the features of the sequence.To solve this problem,an attention mechanism can be introduced.(2)Aiming at the named entity recognition method in the neural network model,in the model training process,the traditional word vector expression is too singular.The BERT model is introduced for pre-training,and the Bi LSTM model is improved to a two-layer Bi LSTM model.The dual-layer Bi LSTM BERT-Bi LSTM-CRF recognition model adds features from a large number of unlabeled data to the network.Comparative experiments prove that the model can improve the recognition effect.(3)System visualization design and implementation.Visualize the research results of this topic,integrate the management and maintenance of the algorithm model into the Web system to realize a set of medical NER visualization application system that can be used in production practice.This system is mainly for medical workers and can retrieve medical texts.And visually analyze the recognition results.For new words that are incorrectly identified or not included,they can be manually labeled and supplemented to the data set according to certain rules to enrich the data set.On the one hand,the model is retrained to improve the recognition rate;on the other hand,the source of the data set can be enriched.To a certain extent,it can solve the problem of the lack of labeled data sets. |
