Named Entity Recognition For Chinese Electronic Record With Neural Network

With the gradual improvement of informatization in the domestic medical field,more and more electronic medical record data are accumulated.These huge amounts of data not only have great commercial value but also have great research value.However,these data are not structured texts.Useful information is distributed in cluttered data and can't be utilized quickly and effectively.Therefore,it is very important to analyze and use Chinese electronic medical records intelligently,which can not only create great value,but also accelerate the development of the medical field faster.Based on this,this work studies the task of named entities recognition in Chinese electronic medical record texts.Named Entity Recognition as the basic work of information extraction aims to identify the entity unit with the most basic semantic meaning in the medical record texts and provide support for other follow-up information extraction tasks.Based on the common domain named entity recognition,this paper proposed a corresponding improvement plan for the Chinese medical electronic texts,then designed and implemented a complete Chinese electronic medical records named entity recognition system.The main work of this paper is as follows:1).Collected and annotated a batch of data of Chinese electronic medical record named entities.Due to the late start of related research in China,there is no open and influential data set for research.One of the few personal studies did not make their own datasets public.At the beginning of the study,this work carried out data annotation work by collecting medical records and related entity dictionaries.2).Designed and implemented a Chinese electronic medical records named entity recognition system.Firstly,a complete named entity recognition system was designed and implemented to provide a basis for further analysis and utilization of electronic medical records.Then,an algorithm was designed based on the existing literature,and achieved a recurrent neural network and conditional random field joint model with abundant word features.This model designs the fine-grained word embedding as the input by re-splitting the Chinese word segmentation results and it reduce the influence of the ambiguous participle on the named entity recognition.According to the characteristics of Chinese electronic medical record entity,this paper designed the features of part of speech,dictionary and other features as a supplement to the automatic learning feature representation of neural network,which improves the recognition effect of complex long entity.3).The design and implementation of a distance-sensitive Seq2Seq model.This work migrates Seq2Seq model to named entity recognition task of sequence labeling by constraining it to be of equal length structure.Combining with the linguistic features of Chinese medical record texts,this paper proposes an improved attention mechanism based on distance modification and achieves results.In addition,the improved model proposed in this paper has achieved good results in the evaluation task of electronic medical records named entity recognition issued by China Conference on Knowledge Graph and Semantic Computing this year.