Research And Application On Intelligent Disease Guidance And Medical Question Answering Method

The internet is the key bridge to connect patients with medical services. When people do not feel well, nearly 90% of them first go on the internet to search for related medical information. The internet has already changed the eco-system of medical services in all major steps. These include medical consultation, clinic visits, treatment and recovery as well as buying medication online. In a recent report, the entire internet healthcare will have a huge share of the market value of one trillion.Online medical guidance has been a very import step. In this paper, we will focus on data mining and machine learning technologies to research on medical guidance and medical question and answer system(QAs), aim at providing human-like, comprehensive and informative automated medical consultation.Most people when they get sick, their lack of medical knowledge and experience will cause them to describe their symptoms inaccurately in medical terms. The medical guidance model mentioned in this paper uses deep learning technologies such as CNN and NLP, to do feature construction and transformation on the raw noisy data. The current model covers 500 different types of diseases and has rank-1 accuracy of nearly 70%. For disease knowledge QAs, when people search a question on the internet they may get many different answers, for some questions it is difficult to have authoritative answers. If they want to know how most doctors say about that they may reference many doctors answers, which takes a lot of effort. The disease knowledge automatic question answering method mentioned in this paper uses multi-label classification technology automatically finds similar question from the existing question and answers, and using information retrieval ranking methods to rank all sentences from answers by a number of doctors. The important sentences extracted will be returned as the answer to the user.While modeling the two model and method, we solve many problems. For the problem that word segmentation tool does not work well on the medical field words, we used mutual information based new word discovery method; for the problem to extract symptoms from the user query we use named entity recognition algorithm; for the problem to search similar questions with the user query we use multi-label classification technology; for the problem to summarize the answers we used TextRank algorithm to find the most important sentences. The data we used in this paper are crawled by the distributed web crawler, in addition, according to the above method and system we combined with WeChat public platform to build a system for users.In this paper we did experiments and analysis to test the model and method, the result show that the model and method are effect.