Research And Application On Data Driven Diagnosis,Treatment And Doctor Recommdation Methods

With the implementation of Internet medical policies,as a part of the massive network information,medical resources have shown an explosive growth trend.Various medical resources analysis services support patients greatly for better medical treatment.However,with the great increase of medical information,patients have difficult to seek the meanful information.This is also the disadvantages caused by "information explosion" and"information overload" in medical sector.In this paper,we focus on the research of multi-source heterogeneous medical data mining problem.We would give a solution to analysis the process of disease,forecast the drug-drug interaction and recommend the right doctors with big data,intelligent information processing and knowledge management theory and technology.The main context of this paper includes:(1)Research chronic diseases course forecast method based on dynamic time match of RNN model.Firstly,the similarity of the sequence is encoded by a RNN model,and the temporal patterns in the data are dynamically matched.Then,the relationship between the events from the two sequences is regarded as a two-dimensional domain.The gradient is overcome by a gated loop unit(GRU).Finally,the similarity between sequences is calculated and the temporal pattern of the patient data is matched dynamically.That is the case classification for diseases.(2)Study on prediction methods of adverse drug reaction based on more task return model.First,each Drug-Drug Interaction(DDI)is considered a forecast task.Then the loss function is minimized based on proper regularization of all tasks.In order to solve the optimization problem,an effective approximate gradient method is proposed,and the types of potential drug interactions for muti-durgs are also detected.(3)Multiple labels based on text mining and prediction model on recommended method and its application to medical resources.Firstly,the doctors were labelled.We employed new word discovery and named entity recognition technology.Then,methods of cluster sampling and frequency sampling are proposed to predict the doctor's label.The labels are transmitted within the scope of all doctors.Finally,we give out a KOL ranking recommendation method for a specific disease.The purpose of this paper is through the study of medical data analysis on three important issues:to improve the level of medical information service in China,to meet the needs of patients for medical information better,to provide medical theory and practice support for a more rational allocation of medical resources.