Process-Aware Dialogue System With Clinical Guideline Knowledge

Abstract

Task-oriented dialogue systems aim to engage in interactive dialogue with people to ultimately complete specific tasks. Typical application domains include ticket booking, online shopping, and healthcare providing. Medical dialogue systems can interact with patients, provide initial clinical advice, and improve the efficiency and quality of healthcare services. However, current medical dialogue systems lack the ability to utilize domain knowledge. This paper extracts regular domain knowledge as well as medical process knowledge from clinical guidelines to improve the performance of dialogue systems. Regular knowledge is used to generate accurate responses for a given input, and process knowledge is used to steer the conversation. The authors divide the task of multi-turn conversation generation into four sub-tasks and propose a 4-layer knowledge-based process-aware dialogue model that incorporates the domain knowledge to generate responses. Results indicate that the approach can lead medical conversations actively while providing accurate responses.