Question answering (QA) is a prominent challenge in natural language processing research that requires machines to predict the correct answer to a posed question by extracting it from a given context. In some cases, QA tasks also involve determining "answerability": whether the answer is present at all in the passage. Recent research has begun to explore domain-specific QA systems, such as for usage in medical contexts. The growing adoption of electronic health records (EHR) in the healthcare system poses a specific QA challenge: retrieving answers from clinical notes to inform medical decisions. This paper introduces the EHReader model based on the Retrospective Reader architecture. The EHReader model incorporates quick reading and deep reading modules, enabling it to evaluate answerability and then verify the answer more comprehensively quickly. We compare EHReader to baseline DistilBERT and BioBERT models for medical QA tasks. The proposed model incorporating only the QuickReader module achieves state-of-the-art results on the benchmark EmrQA medical dataset and outperforms the baseline DistilBERT and BioBERT models.
lcwong0928/ehreader
A question-answering system for electronic health records to ease physician workload.
Jupyter NotebookMIT