Table 1.

Precision, recall, and F₁-scores of deep neural network classifiers for the class of tweets that self-report a COVID-19 diagnosis, evaluated on a held-out test set of 2000 manually annotated tweets.

Classifier	Precision	Recall	F1-score
BERT-Base-Uncased	0.82	0.85	0.84
DistilBERT-Base-Uncased	0.83	0.77	0.80
RoBERTa-Large	0.87	0.92	0.90
BERTweet-Large	0.90	0.91	0.91
COVID-Twitter-BERT	0.96	0.91	0.94