Similar Cases Recommendation On Online Medical Diagnose Platform

As a complementary and optimization tool of traditional medical system,the so-called internet interrogation platform,such as Chunyu Doctor,39 Healthy Net,Good Doctor Online,Clove Garden,etc.,is playing an increasingly important role in daily lives.By using the Online Medical Diagnoseplatform,patients can describe the disease or ask questions online.Doctors can diagnose the condition according to the patients' description and provide professional answers or medical advices to achieve remote diagnosis and treatment.For those common diseases or symptoms,doctors have given the professional answers.Therefore,it is of great significance to dig out the similar cases that have been given high quality answers from the historical case base,one is to make full use of the accumulated knowledge resources of the platform,provide reference for the patients' medical treatment,and the other is to realize the knowledge automation so that the computer can achieve the automatic diagnosis and treatment of the disease.From the similar cases retrieval algorithm and the answer quality estimation algorithm,this thesis studies the semantic similarity calculation,the answer feature extraction,the training of single-label sample model,etc.,based on the requirement of similar cases recommendation for practical applications.Firstly,a number of historical cases similar to the patients' question are retrieved by similar case retrieval algorithm.These cases make up the recommended candidate set.Then,the candidate answers in the candidate set are evaluated according to the answer quality assessment algorithm.Finally,those optimal historical cases are recommended based on the comprehensive evaluation of similar cases retrieval and answer quality assessment results.Therefore,similar cases retrieval can be treated as basic tasks in similar cases recommendation,and the results of the retrieval determine the recommendation quality.Meanwhile,the recommended results for the patients' reference value are determined by the quality of the doctors' answers and suggestions.The main works of this thesis are drawn as follows:1.Combining the query likelihood language model with the word2 vec for similar case retrieval.By using the word vector to map the text content to the low-dimensional continuous space,the relationship between the words can reflect the relationship between the words,and this can effectively overcome the "lexical gap" problem,and then combine with the word2 vec language model,to make semantic similarity between problems more accurate.2.Extracting the characteristics from the four aspects: the superficial characteristics of the answer,the content characteristics of the answer,the relevance characteristics between question and answer and the authority characteristics of the doctor.In face of massive unmarked samples,positive and negative training data are constructed by negative sample screening and "pseudo-negative samples",and then the machine learning model is used to assess the answer quality.3.Utilizing the large-scale platform data to test the algorithm.By comparing the performance of different similar cases retrieval models,the results show that models based on the word vector can achieve good performance.Moreover,by using platform data,the problem of unlabeled sample can be solved effectively,and the answer quality estimation model is welltrained.Finally,combining the results of problem similarity algorithm and answer quality assessment algorithm,the influence of the similarity of the problem and the quality of the answer on the final recommendation result are compared under different weighting factors.This thesis realizes the application of similar cases automatic recommendation algorithm in internet interrogation platform practically.