Diaformer: Automatic Diagnosis via Symptoms Sequence Generation

• DOI: 10.1609/aaai.v36i4.20365
• Publication Date: 2022-07-04T11:02:19Z
• Authors: Junying Chen; Dongfang Li; Qingcai Chen; Wenxiu Zhou; Xin Liu
• Keywords: Sequence (biology)

Automatic diagnosis has attracted increasing attention but remains challenging due to multi-step reasoning. Recent works usually address it by reinforcement learning methods. However, these methods show low efficiency and require task-specific reward functions. Considering the conversation between doctor patient allows doctors probe for symptoms make diagnoses, process can be naturally seen as generation of a sequence including diagnoses. Inspired this, we reformulate automatic Sequence Generation (SG) task propose simple effective Diagnosis model based on Transformer (Diaformer). We firstly design symptom framework learn inquiry disease diagnosis. To alleviate discrepancy sequential disorder implicit symptoms, further three orderless training mechanisms. Experiments public datasets that our outperforms baselines 1%, 6% 11.5% with highest efficiency. Detailed analysis prediction demonstrates potential applying