Table 1.

Bidirectional encoder representations from transformers classification of the performance of tweets about the COVID-19 pandemic that were published by Texas public health agencies between January 1 and June 30, 2020.

Variables	Precisiona	Recallb	F1 scorec
About COVID-19 or not	.93	.93	.93
Information function	.85	.92	.88
Action function	.68	.83	.75
Community function	.58	.58	.58
Handwashing	.75	1.00	.53
Social distancing	.80	.80	.80
Mask wearing or face covering	.85	.96	.90
Staying at home or sheltering in place	.74	.78	.76
Getting tested	.69	.90	.78
Learning more information	.77	.92	.84
Other behaviors	.27	.54	.36
Severity	.69	.92	.79
Susceptibility	.84	.86	.85
Benefit	.42	.70	.52

^aThe microaverage, macroaverage, weighted average, and sample average precision scores were .78, .70, .80, and .46, respectively.

^bThe microaverage, macroaverage, weighted average, and sample average recall scores were .88, .83, .88, and .50, respectively.

^cThe microaverage, macroaverage, weighted average, and sample average F1 scores were .83, .76, .84, and .47, respectively.