Research On Chinese Disease Names Normalization Based On Deep Learning

Medical and health information extraction and knowledge mining are important ways to improve the level of medical and health services and medical management decision-making in my country,especially in the era of big data.On the one hand,due to the uneven distribution of medical resources,the contradiction between the current medical knowledge management and the growing demand for medical and health services has led to the public’s demand for other forms of medical services;on the other hand,with the continuous development of the Internet model,different The form of social media has gradually changed the way people obtain and share medical information in the medical and health field of our country.In this context,in recent years,the medical and health service communities dominated by online health communities have gradually attracted a large number of user groups to participate,and accumulated a large amount of health data.These data contain valuable medical and health knowledge,which is an important basis for in-depth understanding of user needs and improvement of medical and health service level.It has become an important data source for medical academic research,such as knowledge mining and knowledge discovery,knowledge graph construction,knowledge reasoning and knowledge recommendation methods.Disease name normalization refers to the process of mapping the user’s reference to a disease to a standard disease name.As a key link in information extraction and knowledge mining in the medical and health field,this research is the basis for knowledge map construction and deep information mining.However,due to the lack of strict terminology standards for online text representations,online text representations are often more casual and colloquial,and the normalization of disease names for online health communities faces greater challenges.Deep learning model is an important research in the field of artificial intelligence.It has strong learning and adaptability,wide coverage,and high transferability.It can make full use of large-scale semantic features,discover and learn the intrinsic feature relationship between texts,and is very suitable for complex and professional knowledge discovery research in the field of health care.Therefore,in order to effectively solve the challenges and problems in the medical and health field,this paper designs a disease name normalization model based on deep learning,so as to further improve the effect of medical and health information extraction and knowledge mining.This study firstly constructed a normalized dataset of Chinese disease names based on the online health community,conducted a Chinese-English controlled experiment using the basic neural network model,and introduced a variety of external semantic features to discuss its impact on the experiment.Secondly,from the perspective of fusing local and global semantic feature vectors,a Chinese disease name normalization model based on multi-feature fusion is proposed.Then,based on the perspective of multi-task learning,this study applies BERT to obtain dynamic semantic features of input text on the one hand,and constructs a hybrid neural network to extract semantic relations of static semantic features on the other hand,and introduces part of speech and medical dictionary to generate attention weight matrix as Auxiliary task conditioning vector.Finally,the continuous learning method based on the self-attention mechanism is integrated into the multi-task learning model to achieve the optimal normalization effect.The best model(GCBM-BSCL)achieves76.12% on Accuracy@1,87.20% on Accuracy@5,and 90.02% on Accuracy@10.This study combines CNN,BERT,self-attention mechanism,continuous learning,deep semantic fusion and multi-task learning in the normalization task of Chinese disease names for the first time,and optimizes it from the aspects of vector,model,framework and algorithm.It has good application value in the construction of Chinese medical knowledge graph,information extraction and natural language understanding.This study for the first time in Chinese disease name normalization mission CNN,BERT,since the attention mechanism,continuous learning,combining semantic fusion depth and multitasking learning,from the aspect of vector,model,framework and algorithm has been optimized,in the Chinese medical knowledge map construction,information extraction,and has good application value in natural language understanding,This study not only helps to further promote the research progress of knowledge mining in the medical and health field in China,but also has important significance for promoting the application and deepening of artificial intelligence in the medical and health field.