[Paper Review] CORD-19: The COVID-19 Open Research Dataset

Lucy Lu Wang, Kyle Lo+26 more|PubMed|Apr 22, 2020

COVID-19 diagnosis using AIMedicine86 references586 citations

TL;DR

CORD-19 describes the construction, processing, and use of a growing open dataset of COVID-19 and related coronavirus literature, including metadata harmonization, full-text parsing, and community task support.

ABSTRACT

The Covid-19 Open Research Dataset (CORD-19) is a growing resource of scientific papers on Covid-19 and related historical coronavirus research. CORD-19 is designed to facilitate the development of text mining and information retrieval systems over its rich collection of metadata and structured full text papers. Since its release, CORD-19 has been downloaded over 200K times and has served as the basis of many Covid-19 text mining and discovery systems. In this article, we describe the mechanics of dataset construction, highlighting challenges and key design decisions, provide an overview of how CORD-19 has been used, and describe several shared tasks built around the dataset. We hope this resource will continue to bring together the computing community, biomedical experts, and policy makers in the search for effective treatments and management policies for Covid-19.

Motivation & Objective

• Provide a large, up-to-date collection of COVID-19 and historical coronavirus papers from multiple sources.
• Harmonize metadata and deduplicate papers to create a persistent, canonical dataset identifier.
• Extract and structure full-text content (via S2ORC JSON) and tabular data to enable NLP and IR tasks.
• Support a range of applications including clinical reviews, information retrieval tools, and shared tasks for the research community.
• Facilitate daily updates and ongoing expansion with licensing-aware distribution and data processing pipelines.

Proposed method

• Ingest paper metadata and documents from PubMed Central, PubMed, WHO COVID-19 Database, bioRxiv, medRxiv, arXiv, and publishers via Semantic Scholar.
• Cluster papers using identifiers (doi, pmc_id, pubmed_id, arxiv_id, who_covidence_id, mag_id) to form cord_uid clusters and deduplicate conservatively.
• Parse PDFs to TEI XML with GROBID, convert to S2ORC JSON, and also parse JATS XML for PMC papers to the same target format.
• Release full-text parses (PDF-derived and XML-derived) in S2ORC JSON with metadata and bibliographies, including SHA-1 for PDFs.
• Release HTML table parses and apply a separate PDF table processing pipeline (IBM SDU and GTE) to extract and match 188K tables.
• Provide licensing-aware metadata and note copyright constraints; commit to daily updates and openness for research use.

Experimental results

Research questions

• RQ1 How can a rapidly growing corpus of COVID-19 literature be assembled from heterogeneous sources with consistent metadata?
• RQ2 What processing pipeline yields machine-readable, structured full-text representations suitable for NLP and information retrieval?
• RQ3 How can updates be regularized and identifiers be persistent across releases to support reproducible research?
• RQ4 What is the role and impact of CORD-19 in enabling clinical reviews, tool development, and shared tasks around COVID-19 literature?

Key findings

• The dataset grew to over 140K papers with over 72K full texts as of version updates.
• More than 47K papers and 7K preprints on Covid-19 and coronaviruses were released since the start of 2020, comprising nearly 40% of the corpus.
• Approximately 48% of papers have PDF full-text parses and about 37% have XML parses (PMC subset).
• The dataset has been downloaded over 200K times in the first three months after release, driving numerous text mining and discovery systems.
• CORD-19 enables a wide range of applications, including systematic reviews by clinicians and numerous AI-driven information retrieval and extraction tools; it also spurred shared tasks like Kaggle challenges and TREC-COVID.