language understanding in clinical conversation

NLP

本文分享3篇医疗对话中与语言理解(可能也包括信息抽取/信息挖掘)任务相关的论文
- Extracting Symptoms and their Status from Clinical Conversations (Google, ACL2019)PDF 不开源
- Learning to Infer Entities, Properties and their Relations from Clinical Conversations (Google, EMNLP2019) PDF 不开源
- Enhancing Dialogue Symptom Diagnosis with Global Attention and
Symptom Graph (Fudan, EMNLP2019) PDF 开源

Learning to Infer Entities, Properties and their Relations from Clinical Conversations (Google, Emnlp2019)

学习从临床对话中推理症状/属性和它们的关系

这篇论文提出了两个Task, the Symptom Task(Sx)和the Medication Task(Mx). Medication指的是用于治疗的药物.

- Sx任务抽取对话中的如下三元组(symType, propType, propContent) 即(症状类别, 属性类别, 属性内容) 比如(pain, frequency, everymorning)
- Mx任务抽取对话中的以下三元组(medContent, propType, PropContent) 即(药物内容, 属性类别, 属性内容) 比如(Ibuprofen, freqency, twice a day). medContent种类非常多并且在不断更新, 因此作者有时抽取的是indirect references例如"pain medications". 作者定义了三种medication的属性类别, 包括dosage剂量/duration持续时间/frequency频率).