digitalepidemiologylab/covid-twitter-bert-v2

This model is identical to covid-twitter-bert - but trained on more data, resulting in higher downstream performance . It was trained on 97M unique tweets (1.2B training examples) collected between January 12 and July 5, 2020 containing at least one of the keywords "wuhan", "ncov", "coronavirus", "covid", or "sars-cov-2" The model can e.g. be used in the fill-mask task (see below) You can also use the model without the MLM/NSP heads and train a classifier with it . The model was evaluated based on downstream Twitter text classification tasks from,