A role distinguishing Bert model for medical dialogue system in sustainable smart city

Abstract

Smart medicine is a vital component for building sustainable smart city. In smart medicine, intelligent dialogue system is playing an important role in providing personalized and efficient healthcare services to improve the quality of life for human beings. The Bert model has become a popular way to construct intelligent medical dialogue system. However, the current Bert model is difficult to achieve desirable results for this task since its input does not reflect the difference in roles. To overcome this drawback, we present a role distinguishing Bert model for intelligent medical dialogue system to help construct the sustainable smart city. Particularly, we segment and label the utterances depending on different dialogue roles, and then construct the corresponding segment embedding as the input of our model. Furthermore, we substitute the NSP task with the SOP task to better learn the coherence between sentences. Finally, we verify the proposed model by comparing it with Ernie on some online E-commerce datasets for intent recognition, semantic matching, and session dialogue classification. The results demonstrate that our proposed model improves the average 1% accuracy for different tasks in dialogue system, proving the potential of the proposed model for establishing intelligent medical dialogue system in smart city.