Enhanced Medical Dialogue Diagnosis with Intra-inter Window Attention Encoder

Information extraction from clinical dialogue to build a knowledge graph that can generate electronic medical charts has been an important application of NLP in the medical industry. Although substantial research has been carried out on such information extractor, existing models underperform due to lack of contextual background between dialogue windows. This study proposes a model that can extract relevant medical items from the doctor-patient dialogues by leveraging a multi-layer encoder-decoder framework. The experimental results on a well-studied dataset shows that our model outperform the current baseline models, which prove the model effectiveness.