inria-covidontheweb.md

Covid-on-the-Web Dataset BY University Côte d'Azur, Inria, CNRS

Context

Covid-on-the-Web Dataset is an RDF dataset that provides two main knowledge graphs produced by analyzing the scholarly articles of the COVID-19 Open Research Dataset (CORD-19) [1], a resource of articles about COVID-19 and the coronavirus family of viruses:

• the CORD-19 Named Entities Knowledge Graph describes named entities identified and disambiguated by NCBO BioPortal annotator, Entity-fishing and DBpedia Spotlight.
• the CORD-19 Argumentative Knowledge Graph describes argumentative components and PICO elements (Patient/Population/Problem, Intervention, Comparison, Outcome) extracted from the articles by the Argumentative Clinical Trial Analysis platform (ACTA).

Challenges

1. Generate a consistent KG from multiple inputs: CORD-19 corpus, named entities extraction tools, ACTA. Each input comes with its own format that requires adapting of the KG generation pipeline.
2. Huge amount a JSON data to process: 50M+ named entities (extracted from 50K+ articles) to describe in RDF. Our mapping tool generates RDF in memory before saving the result. Needs fine-tuning to keep acceptable memory/cpu/time usage. A streaming mode would be more appropriate.
3. Input JSON documents require pre-processing (cleanup of erroneous characters, filtering) that we have done using MongoDB, but having everything in the mapping description would avoid too many steps.
4. Frequent updates of CORD-19 require regular update of the KGs
5. Specific iteration needs in the JSON-to-RDF mapping

Resources

• Website: https://github.com/Wimmics/CovidOnTheWeb
• Data: SPARQL endpoint, download: see website
• Mappings: Morph-xR2RML, an implementation of the xR2RML mapping language for MongoDB databases. Mapping files: https://github.com/Wimmics/CovidOnTheWeb/tree/master/src/xR2RML
• Ontology: BIBO, Dublin Core, FABIO, FOAF, Web Annotations Ontology, PROV-O, Schema.org, Argument Interchange Format, Argument Model Ontology, SIOC Argumentation Module
• Output: we experiment with several visualization tools to display the co-occurrence network of named entities (e.g. co-occurence of a cancer and a virus of the coronavirus family). Examples with a specific lib called MGexplorer, or with Jupyter Notebooks