Enhancing Dialogue Symptom Diagnosis with Global Attention and Symptom Graph

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 5033–5042, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 5033 Enhancing Dialogue Symptom Diagnosis with Global Attention and Symptom Graph Qin Chen1§, Xinzhu Lin1§, Xiahui He1, Huaixiao Tou2, Ting Chen3, Zhongyu Wei1* 1School of Data Science, Fudan University 2Alibaba Group, China 3Ping An Technology, China {qin chen,18210980055,18210980050,17210980029,zywei}@fudan.edu.cn huaixiao.thx@alibaba-inc.com Abstract Symptom diagnosis is a challenging yet pro- found problem in natural language processing. Most previous research focus on investigat- ing the standard electronic medical records for symptom diagnosis, while the dialogues be- tween doctors and patients that contain more rich information are not well studied. In this paper, we first construct a dialogue symptom diagnosis dataset based on an online medical forum with a large amount of dialogues be- tween patients and doctors. Then, we pro- vide some benchmark models on this dataset to boost the research of dialogue symptom di- agnosis. In order to further enhance the perfor- mance of symptom diagnosis over dialogues, we propose a global attention mechanism to capture more symptom related information, and build a symptom graph to model the asso- ciations between symptoms rather than treat- ing each symptom independently. Experimen- tal results show that both the global attention and symptom graph are effective to boost di- alogue symptom diagnosis. In particular, our proposed model achieves the state-of-the-art performance on the constructed dataset. 1 Introduction With the widespread use of electronic health records (EHRs) in medical treatment, symptom diagnosis based on EHRs have received a lot of attention in the natural language processing (NLP) research community (Linder et al., 2007; Shivade et al., 2013). Previous work on EHRs achieved great success in determining the diagno- sis of clinical depression (Trinh et al., 2011), iden- tifying community-acquired pneumonia (DeLisle et al., 2013), improving medication reconciliation (Persell et al., 2018) and infection detection (Tou et al., 2018). § Contributed equally * Corresponding author Conversation . . . . . . Patient: 孩子50天，嗓子有痰，而且咳嗽，是感冒吗？ Patient: My kid has been born for 50 days. He has a cough, Patient: with phlegm. Does he have a cold? Doctor: 咳嗽，有痰，流黄鼻涕，说明有炎症了。 Doctor: Coughing, sputum and a runny nose, indicate Doctor: inflammation. Doctor: 咳嗽频繁吗？发烧吗？ Doctor: Does he cough frequently? Had a fever? Patient: 一阵一阵的，咳起来厉害，不发烧。 Patient: Occasionally, but his cough is very serious. No fever. . . . . . . Symptom Diagnosis True: . . . . . .有痰 (Phlegm) 咳嗽 (Cough) 流黄鼻涕 (Runny nose) 炎症 (Inflammation) . . . . . . False: . . . . . .发烧 (Fever). . . . . . Uncertain: . . . . . .感冒 (Cold). . . . . . Table 1: An example of a doctor-patient dialogue and symptom diagnosis. Underlined phrases are symptom descriptions. True, False, and Uncertain are the infer- ence results for whether the symptom exists in the pa- tient. However, EHRs usually contained historical in- formation, such as the medical records or health records, …

Content truncated to 500 words