Research On Methods Of Symptom Recognition And Disease-assisted Diagnosis For Orthopedic Consultation

With the increasing awareness of healthcare and the development of Internet technologies,more and more people are willing to spend more time in exploring various medical and health information on the Internet.At present,there are mainly two ways for obtaining online medical and health information: one relies on the search engine,which is simple and fast,however,it is much difficult for non-medical professionals in searching,understanding and so on;the other depends on the health portal,which provides an consulting platform for patients,however,due to doctors’ limited time and energy,they often cannot reply to patients in time.To satisfy those strong needs,medical automatic question answering system has come into being.Furthermore,the involved disease inference cannot be solved directly only through the knowledge base,thus it becomes the biggest stumbling block in medical automatic question answering.An effective auxiliary diagnosis method for diseases is the key to overcome that obstacle,and this paper does take it as the research emphasis.Considering the characteristics of medical question-answering data,the paper first strengthens the data features through the symptom recognition,and then on that basis carries on the auxiliary diagnosis of diseases.In addition,in order to be more targeted,the orthopedics consultation texts are taken as the whole paper’s study object.Aiming at difficulties of the symptom recognition in orthopedics consulting,the paper adopts the ideas of “from small to big” and “dividing first and combing later”,and takes advantage of the constitution patterns of symptom entities.This paper firstly concatenate the part-of-speech feature,suffix feature and deep feature to recognize the basic symptom for determining the location of every symptom entity,then does incremental iteration using the parts-of-speech determined by the constitution patterns to obtain the boundary of every symptom entity for identifying symptom entity,and finally merges all symptom entities identified in each question text to achieve the identification of the complete symptom.The experimental results show that,for the identification of orthopedics symptoms in medical question answering,the accuracy of the part-of-speech incremental iteration based method proposed by this paper is 5.4% higher at least than those of the common bio-medical named entity recognition methods.As for the auxiliary diagnosis of diseases,the paper regards it as a multi-label classification problem.Aiming at the difficulty,firstly,on the basis of orthopedics symptom recognition,this paper combines it with the patient label feature and the deep feature of the original question text to represent the question text effectively,and utilizes Softmax classifier to obtain all the disease labels whose corresponding probabilities are highest of Top K as the initial candidate set of labels.Secondly,the “second-order” correlations among candidate labels are obtained by calculating the conditional probability.Finally,the initial candidate labels are rejected by both the threshold and the label correlations to return the labels satisfying the condition as the output result of the auxiliary diagnosis of orthopedics diseases.The experimental results show that in the proposed method based on symptom feature the values of hamming-loss and coverage are 0.0818 and 1.3558 lower,respectively,and the value of average-precision is 0.047 higher than those of the traditional methods when the auxiliary diagnosis of diseases is made.Finally,according to the above mentioned methods,in the patient-based medical automatic question answering system,the subsystem of symptom recognition and auxiliary diagnosis of diseases is designed,and the corresponding UML modeling and core codes presentation are carried out.