Smart-MD: Neural Paragraph Retrieval of Medical Topics

Abstract

We demonstrate Smart-MD, an information retrieval system for medical professionals. The system supports topical queries in the form [disease topic], such as ["lyme disease", treatments]. In contrast to document-oriented retrieval systems, Smart-MD retrieves relevant paragraphs and reduces the reading load of a medical doctor drastically. We recognize diseases and topical aspects with a novel paragraph retrieval method based on bidirectional LSTM neural networks. We demonstrate Smart-MD on a dataset that contains 3,469 diseases from the English language part of Wikipedia and 6,876 distinct medical aspects extracted from Wikipedia headlines.