Abstract

Beyond current conversational chatbots or task-oriented dialogue systems that have attracted increasing attention, we move forward to develop a dialogue system for automatic medical diagnosis that converses with patients to collect additional symptoms beyond their self-reports and automatically makes a diagnosis. Besides the challenges for conversational dialogue systems (e.g. topic transition coherency and question understanding), automatic medical diagnosis further poses more critical requirements for the dialogue rationality in the context of medical knowledge and symptom-disease relations. Existing dialogue systems (Madotto, Wu, and Fung 2018; Wei et al. 2018; Li et al. 2017) mostly rely on datadriven learning and cannot be able to encode extra expert knowledge graph. In this work, we propose an End-to-End Knowledge-routed Relational Dialogue System (KR-DS) that seamlessly incorporates rich medical knowledge graph into the topic transition in dialogue management, and makes it cooperative with natural language understanding and natural language generation. A novel Knowledge-routed Deep Q-network (KR-DQN) is introduced to manage topic transitions, which integrates a relational refinement branch for encoding relations among different symptoms and symptomdisease pairs, and a knowledge-routed graph branch for topic decision-making. Extensive experiments on a public medical dialogue dataset show our KR-DS significantly beats stateof-the-art methods (by more than 8% in diagnosis accuracy). We further show the superiority of our KR-DS on a newly collected medical dialogue system dataset, which is more challenging retaining original self-reports and conversational data between patients and doctors.

Contributions

Our contributions are summarized in the following aspects.

1) we propose an End-to-End Knowledge-routed Relational Dialogue System (KR-DS) that seamlessly incorporates medical knowledge graph into the topic transition in dialogue management, and makes it cooperative with natural language understanding and natural language generation.

2) A novel Knowledge-routed Deep Q-network (KRDQN) is introduced to manage topic transitions, which integrates a relational refinement branch for encoding relations among different symptoms and symptom-disease pairs, and a knowledge-routed graph branch for topic decision-making.

3) We construct a new challenging end-to-end medical dialogue system dataset, which retains the original self-reports and the conversational data between patients and doctors.

4) Extensive experiments on two medical dialogue system datasets show the superiority of our KR-DS, which significantly beats state-of-the-art methods by more than 8% in diagnostic accuracy.

Architecture

Conclusion

In this work, we move forward to develop an End-to-End Knowledge-routed Relational Dialogue System (KR-DS) that enables dialogue management, natural language understanding, and natural language generation to cooperatively optimize via reinforcement learning. We propose a novel Knowledge-routed Deep Q-network (KR-DQN) upon a basic DQN to manage topic transitions, which further integrates a relational refinement branch for encoding relations among different symptoms and symptom-disease pairs, and a knowledge-routed graph branch for policy decision guided by medical knowledge. Additionally, we construct a new benchmark focusing on end-to-end medical dialogue systems, which retains the original self-reports and the conversational data between patients and doctors. Extensive experiments on two datasets show the superiority of our KR-DS, which generates the most precise and reasonable results.