neulab / cord19

cord19 related stuff

CORD-19 Information Aggregator

by Graham Neubig, Emma Strubell, Zhengbao Jiang, Zi-Yi Dou and others at the Carnegie Mellon University Language Technologies Institute

View the extracted information here

This is a tool to browse answers the scientific literature may provide regarding various questions about the novel coronavirus and COVID-19. Click the questions below to see a list of answers with links to the sources that provided them.

We are looking for help improving this tool! If you are familiar with reading the medical literature and could give fine-grained feedback please contact us at [email protected]. Or, if you are a programmer and could help please feel free to contribute to this repository!

Information extraction script

See the extraction directory.

Docs

• Datasets doc
• OIE Templates Doc

Data extraction

Our data is based on the CORD-19 dataset. Ask the authors of this repository for access if you're interested.

• text-only/*.oie

  Each line has <sub, rel, obj> triplets extracted from the corresponding sentence in text-only/*.sent. Triplets are separated by \t and each triplet is of the format subject|||relation|||object, e.g. he#0,2|||'s on#3,8|||outside#13,20. Subjects, relations, and objects are continuous spans of tokens in the format of text#start_char,end_char, e.g. he#0,2. Note that text of the span might only be a substring of the string spanning from start_char to end_char.

cord19's Issues

Better display of extracted information

A few improvements to the display of extracted information may be warranted:

1. Make it clear what keywords are being searched for each of the questions.
2. Better explanation of the OpenIE extractions. Maybe highlight the words in the extracted sentences.

Pretty output of extraction results in HTML

Currently extraction results are output in text on the command line, it'd be nice if we could output them in a prettier format. Also, maybe machine readable.

Sorting of extractions by the reliability of the source

Currently the extraction algorithm is not considering how reliable the source is (peer reviewed journal of what quality, or biorxiv). It'd be nice if we could do this.

Make it possible to do regexes over noun phrases in text

Currently regexes in text can't be on the borders of the template. For example the following is not allowed:

[X] is a risk factor of [Y]

It would be nice if we could do regexes that matched NPs, so we would be able to solve this issue.