Welcome to the CovidNews-NER dataset. It is an English COVID-19 Named Entity Recognition dataset in the pandemic news domain, addressing current NER models’ lack of ability to tackle new and out-of-domain topics. For more information please reference the original paper "How to tackle an emerging topic? Combining strong and weak labels for Covid new NER".

Structure

There are 5 folders included in dataset/, dataset is in CONLL format. Files ending with _text.txt are the text for each entry, files ending with _true.txt are the strong labels for each entry and files ending with _dist.txt are the weak labels for each entry.

• train/: 2100 entries of strong and weak training data
• valid/: 600 entries of strong validation data
• test/: 300 entries of strong test data
• rest/: 10000 entries of additional weak data
• other/: contains all strong and weak data seperated by original language translated from and combined

Entity Categories

The dataset covers 10 entity categories listed below. For detailed information about the dataset please see the corresponding paper.

• person
• location
• organization
• time
• disease
• virus
• product
• animal
• symptom
• bacterium

Code

The code uses the RoSTER repository with some modifications to it, to view the original please click here.