Research And Application Of Personalized Health Service Recommendation System Based On Neural Network

With the application of IT in the field of health,the EHRS has been rapidly promoted,and many EHRs have been accumulated.These data contain many practical information.Data analysis and mining of EHRs has become one of the hot topics in the field of health.Therefore,it has important research significance and application value to establish a personalized drug recommendation model based on the EHRs,recommend the drugs needed by the patients through their medical information,and assist doctors in diagnosis.In order to improve the effect of the recommendation model,the main contents are as follows:Based on the complexity and timing of data,a model combining GAT and TNN,which is mainly optimized as follows: 1)Using the correlation between clinical events during the visit,the EHRs is regarded as a graph.The drug recommendation is transformed into a task of modeling based on graph data and time correlation.2)The GAT is used to extract the structural correlation,and the time updating module is used to learn the time dependence between multiple visits of patients.3)Use the attention to replace the Update Gate in GRU,so as to avoid the vectors with zero correlation affecting the learning process of GRU.On the MIMIC-III,GA-2DGRU in Jaccard,F1 and AUC indexes is 0.4719～0.4972,0.6359～0.6581 and 0.7198～0.7468.The best performance of GA-2DGRU in Jaccard is0.0764,0.0955,0.0351 and 0.0237 higher than other models LR,LEAP,RETAIN and PREMIER;The best performance of F1 is 0.079,0.1062,0.0330 and 0.0332 higher than other models LR,LEAP,RETAIN and PREMIER;The best performance of AUC is 0.0511,0.1756,0.0488 and 0.0314 higher than other models LR,LEAP,RETAIN and PREMIER.It is verified that the model can capture the inherent structure and time characteristics of EHR well in prediction.Finally,the demand analysis,overall design and system testing of the medicine assisted recommendation system are carried out.The practical application of the improved model was explored,and it was integrated into the medicine recommendation module of the treatment assistant system,which verified the availability of the model.Figure [30];Table [15];Reference [55]...