Medical Named Entity Recognition Method Based On Dynamic Networks

In recent years,with the rapid development of artificial intelligence technology,the intersection of medical health and artificial intelligence has become a hot research direction,which promotes the production of a large number of new intelligent medical equipment and electronic medical systems.The electronic medical system contains a wealth of medical and health data,which has laid a strong data foundation for medical research and medical practice.However,it is still challenging in making full use of these data for exploration and analysis to support clinical decision-making and public health better.Among them,named entity recognition(NER)in medical texts,especially named entity recognition in Chinese medical texts,is one of the most severe challenges.Due to the complexity,scarce resources,and entity nesting of Chinese medical texts,Chinese NER in the medical field is more challenging,although NER has attracted great attention in various fields.To cope with the problems of NER in the current medical field,this paper has conducted a lot of researches on electronic medical record data and existing technologies,and further proposed a series of methods of medical named entity recognition based on dynamic networks,and carried out abundant experimental verification.The contributions of this thesis are as follows.(1)The paper proposes a Chinese medical NER method based on multi-semantic dictionaries and multi-modal trees(MSD＿DT＿NER)to handle the problem of lacking semantics in Chinese medical texts.This method can use four different path modes in a tree structure to obtain word information of different lengths,and splicing word information and character information,which can realize the fusion of multi-granularity features and enrich the semantic embedding expression.(2)The paper presents a Chinese medical NER method based on graph neural network and cross-language(GC＿NER)to solve the problem of insufficient resources of Chinese medical texts.The method of cross-language knowledge transfer is adopted to transfer high-resource language knowledge to Chinese medical texts,supplement knowledge,and supervise the NER tasks of Chinese medical texts with the help of external language knowledge.(3)The paper proposes a Chinese medical NER method based on dependency syntax analysis and dynamic stacking network(SD＿NER)aiming at the problem of named entity nesting in medical texts.This method can perform adaptive network stacking according to the number of layers of entity nesting,and use the features of embedded entities to help identify external entities,and realize dynamic stacking recognition of nested entities.Finally,we conduct extensive experiments on the electronic medical record evaluation data set of the China Conference on Knowledge Graph and Semantic Computing(CCKS),and use multiple indicators to evaluate the performance of all of our models.The results prove that the series of Chinses medical NER methods proposed in this paper achieve the SOTA results.