Online Diagnosis-Treatment Department Recommendation based on Machine Learning in China

Abstract

Abstract [Background] As a supplement to the traditional medical service mode, online medical mode provides services of online appointment, consultation, online remote treatment, etc. Due to the complexity of online medical service application scenarios and the necessity of professional knowledge, the accuracy of the department during patient online consultation is limited, and traditional recommendation methods in the medical field face problems such as low computational efficiency and poor effectivenessMachine learning has been widely and successfully applied in medical fields. Hence, this study applies machine learning technology to intelligent department recommendation for online diagnosis-treatment services.[Objective] This paper compares the accuracy rate of different machine learning algorithms utilized for intelligent medical department recommendation at online diagnosis-treatment platforms, aiming to extract new features from the research data to improve recommendation effect. [Method] Based on online diagnosis-treatment data from 20 second-level departments at WeDoctor platform, the accuracy rates of two text vectorization methods in implementing intelligent department recommendation pairing with four classifiers, namely support-vector machine, random forest, multinomial Bayes and logistic regression, were compared from the perspective of hierarchy classification of department and secondary department. Furthermore, the paper also introduces variable of gender and age to improve accuracy rate of department recommendation. [Results] The recommendation accuracy rate is the best when text vectorization method is word2vec and classification algorithm is support-vector machine. The accuracy rate is 79.40% after adding age and gender into the model. The accuracy rate of intelligent recommendation was only about 52.7% for general surgery department and the reason behind is probably that online consultations from patients are usually so complicated that department functions are prone to be confused. [Conclusion] The effect of machine learning for online diagnosis-treatment platforms’ intelligent department recommendation is particularly significant. And the recommendation accuracy rate can be further improved by integrating age and gender into the algorithm. Moreover, considering the fact that some disease symptoms are confusing and can affect the recommendation accuracy rate to some extent, multiple departments shall be recommended to improve patients' online diagnosis-treatment experience and satisfaction.Fund program: National Natural Science Foundation of China(71571162)