Published February 14, 2021 | Version 3.0

CORD-19 Diachronic Concept Embeddings

  • 1. University of Pittsburgh
  • 2. Carnegie Mellon University
  • 3. The Ohio State University

Description

Medical concept embeddings trained on monthly releases of the CORD-19 dataset, for analysis of diachronic change.

This resource includes:

  1. Concept embeddings, trained using the JET toolkit on the set of articles added to CORD-19 in each month of March 2020 - October 2020.  Concepts are SNOMED CT codes.
  2. Nearest neighborhood-based analysis results of the concept embeddings, stored as a SQLite database.

Database content excludes terms and definitions for SNOMED CT codes, which are licensed content owned by SNOMED International. Information on accessing SNOMED licensed content may be found here.

Files

CORD-19_monthly_embeddings.zip

Files (1.3 GB)

Name Size
md5:5bc7c50f6bfd89c1c33d34d394556c92
340.3 MB Download
md5:be0737c6c04a9e51fc04e94f384dc1f9
998.5 MB Preview Download

Additional details

Related works

References
Conference paper: 10.18653/v1/W18-3026 (DOI)