Using automation to produce a ‘living map’ of the COVID-19 research literature

Abstract

The COVID-19 pandemic has disrupted life worldwide and presented unique challenges in the health evidencesynthesis space. The urgent nature of the pandemic required extreme rapidity for keeping track of research, andthis presented a unique opportunity for long-proposed automation systems to be deployed and evaluated. Wecompared the use of novel automation technologies with conventional manual screening; and Microsoft AcademicGraph (MAG) with the MEDLINE and Embase databases locating the emerging research evidence. We foundthat a new workflow involving machine learning to identify relevant research in MAG achieved a much higherrecall with lower manual effort than using conventional approaches.