Research On The Recommendation Method Of Similar Cases For Traditional Chinese Orthopedics Inquiries

With the rapid development of Wise Information Technology of 120,Artificial Intelligence technology has gradually integrated into the medical industry,and the healthy medical information platforms which interacts between patients,medical staff and medical institutions have emerged.At present,the research of clinical assistant diagnosis and treatment is mainly through analysising and processing the massive medical data and diagnostic data which in the form of multimedia.Therefore,feature extraction from massive data is indispensable,and the significance of the research of how to carry out personalized diagnosis and treatment is also far-reaching.At present,Chinese TCM(Traditional Chinese Medicine)Electronic Medical Record is a new thing which developed along with the informationization process of Chinese medicine hospitals.It has the characteristics of historical,humanistic and qualitative description.However,because of the difference of TCM Electronic Medical Records between natural language texts,the different medical staff's consultation habits,and the lower platform efficiency,so clinical consultation often fails to achieve the expected diagnosis and treatment effect.This paper focuses to study on the comprehensive feature representation and similar case recommendation by setting up a TCM Orthopedic consultation platform:Firstly,because of the coexistence of structural features and continuous features in the TO(Traditional Chinese Medicine Orthopedics)Electronic Medical Records,traditional text feature representation methods are not fit and often inferior.At the same time,most feature engineering relies on existing medical data too much and cannot learn hidden features well.Aiming at two problems above,this paper proposes a comprehensive feature representation method based on GCNN-based VAE(Gated Convolutional Networks-based Variational Autoencoder).This method combines feature partition and represents the data of different structure types separately as feature vectoration.After feature fusion,training comprehensive feature representation by supervised learning method.Experiments show that this method has an increase of 2.8 percentage points over the comprehensive index value of the existing method.Secondly,usually there are multiple target users in the consultation platform,and medical staff's consultation habits are different.At the same time,the clinical diagnosis requires the platform need to be fast and accurate.The traditional methods of rule-based diagnosis and retrieval-based case recommendation cannot achieve the expected results.Aiming at two problems above,this paper proposes a similar case recommendation method based on the improved Wide&Deep Model,which incorporates doctor user's preference characteristics and recommends a list of similar cases to the doctor with his currently diagnosing.The experimental results show that the method improves the recommendation efficiency of similar cases,and increase of 2.6 percentage points over the comprehensive index value of the existing method.Finally,based on the theoretical knowledge above,the analysis and design of the TO consultation platform is used to help obtain comprehensive feature representation and conduct similar case recommendation more accurately.In this way,to verify the application value of the above research in real projects.