Automatic Identification Of Online Physician Communication Style By Using Three-dimensional Feature Fusion Model

Tele-consultation is one of the important ways of patient-physician consultation.Especially in the recent global outbreak of Corona Virus Disease 2019,in order to reduce the risk of nosocomial infection in ordinary patients,tele-consultation has been increasingly advocated.The tele-consultation is very dependent on the physician communication style,since the information transmission channel of the teleconsultation is more limited than the face-to-face consultation.And patient-centered communication(PCC)is widely recommended currently.In order to optimize the performance and management of online health services,it is necessary to carry out research on automatic identification of Patient-centered communication style in teleconsultation.Due to the high complexity of communication style identification,the commonly used methods of identification about physician communication style mainly include "direct observation using standardized patients ","self-report by patient and physician","rating scales ","interactive analysis system ".The above methods are either not based on objective dialogue as the analysis object,which is not suitable for research in teleconsultation situations;or although the objective dialogue is used as the research object,but the identification method stays in the manual coding stage,which leads to low analysis efficiency.In view of the above situation,this study proposes an automatic identification model of physician communication style based on fusion of three dimensional features for tele-consultation.The construction of this model consists of three steps: acquisition and annotation of text set;structured representation of text;learning and optimization of classifier.(1)Regarding the acquisition and annotation of text sets: this study crawled the doctor-patient communication data on the online health website as data processing objects.Next,the labeling example is proposed based on the definition of PCC connotation elements,and two experts were asked to label the doctor’s dialogue text separately.(2)About the text representation model: this study constructs sub-features that can characterize the doctor’s dialogue text from three aspects: vocabulary feature dimension,sentence feature dimension,and conversation feature dimension,based on the analysis of semantic features of PCC texts.Then fuse the above features according to the basic idea of the vector space model,so as to obtain the "text representation model of three-dimensional feature fusion".(3)Regarding the learning and optimization of classifiers: a balanced text sets was obtained by the clustering-based undersampling approach using the nearest neighbors of the cluster centers,which are used to train and optimize the SVM algorithm in order to obtain the classifier with the best performance.Comparative experiments based on real online doctor-patient consultation proved the effectiveness of this model.The experimental results showed that the F1 of the three-dimensional feature text representation model is better than that of the ordinary text representation model.And the F1 of SVM algorithm is significantly better than Naive Bayes,KNN and C4.5.The model will achieve the best F1 of 0.941,when the confidence threshold of vocabulary expansion features is 0.7 and the classification algorithm is SVM.This research can promote the development of online health community behavior and doctor-patient communication style theory,and can also provide direction guidance and scientific basis for online health service optimization and service management.