Medical Q&A Detection And Recommendation System Based On Machine Learning And Link Prediction

With the continuous promotion of the "Internet + medical" policy,there are more and more doctor-patient Q&A platforms in the market,and people can now complete their diagnosis of their diseases without leaving their homes.However,there are still many problems.For example,the data of each doctor-patient Q&A platform is not interoperable,the quality of the platform doctors is uneven,the questions cannot be answered within a limited time,and the condition could be easily misdiagnosed according to the one-sided descriptionThrough investigation and analysis,this thesis develops a medical question and answer combination and recommendation system using the B/S framework.Named entity recognition algorithm and link prediction algorithm are used to solve problems such as question and answer analysis,user’s disease self-checking,history records,disease complication viewing,disease prediction,etc.Besides those mentionaed above,the system can also be updated with newest information conveniently,for example information such as new diseases,symptoms,and Q&A records can be added to the system regularly.Over all,this thesis mainly includes the following aspects:1)Corresponding crawler rules are developed using java’s htmlunit tool class for several most popular doctor-patient quiz platforms.Obtained the currently recorded diseases,illnesses,and doctor-patient quiz records,and structural information.2)Named Entity Recognition is performed using Bi-LSTM and CRF algorithm for each question and answer record.Relevant information of diseases and symptoms of each question and answer is extracted and corresponding "disease-symptoms" networks are constructed.3)This thesis proposes an n-point link accuracy index that can better show the accuracy of single node prediction.This thesis develped a new link prediction algorithm(NIS)bases on an older algorithm(IS)from literature.The NIS algotithm is used to predict the connection between diseases and the connection between symptoms.At the same time,this thesis analyzes the distribution of disease frequency and the distribution of symptoms in each disease on the doctor-patient platform,and proposes two calculation methods of weight matrix to predict the disease.4)The system was built using the SSM framework.It integrates the named entity recognition algorithm,link prediction algorithm and disease disease prediction algorithm into the system.For the front-end development of the system,we used AJAX to transfer all information asynchronously,increasing the fluency of the user’s use of the system.In order to improve the maintainability of the system,we used front-end separation,and the front-end interaction uses JSON format to transfer data.At the same time,we provide an API interface to facilitate other developers to use the system’s methods and results.