What public health campaigns can learn from people’s Twitter reactions on mask-wearing and COVID-19 Vaccines: a topic modeling approach

Figures & data

Table 1. Keywords to subset tweets about mask-wearing and vaccination

Corpus	Keywords/hashtags
Mask	mask, facemask, fac cover, faceshield wearamask billionshieldschallenge, billionshields, nomaskonme, nomask, nomask, maskless, maskup, nonewnormal, nomuzzles, nomaskselfie, showyoursmile, facecoverings
Vaccine	vaccine immunity, immune, immunization, pharm

Table 2a. Topic on mask by LDA model

Topic name	Topic size	Topic sentiment	High-frequency tokens
Mask as new norm	10.8%	0.11	Wear, social distance, pandemic, protect, keep, stay
Face shield DIY	10.3%	0.10	Bottle, empty, soda, recycle, billionshield, challenge
Mask for love	18.8%	0.02	Love, besafe, staysafe, trust, savelife, superhero
Ads/sales	46.8%	0.14	Sale, washable, design, fashionable, price
Government criticism	13.3%	0.11	Trump, operation, republican, idiot, leader, law, court

Table 2b. Topic on vaccine by LDA model

Topic name	Topic size	Topic sentiment	High-frequency tokens
Science	25.8%	−0.006	Pandemic, scientist, herd immunity, compromise, immune_system
Coping without vaccine	25.0%	0.005	Deal, safe, distance, stay, work, home
Immunity boost	14.9%	0.109	Immune_system, boost, energy, vitamin, sleep
Vaccine race	14.3%	0.064	Vaccine, race, test, headline, process
Politics	20.0%	0.127	Mail, voter, rule, ballot

Figure 1. Topic sizes and sentiment scores on mask-wearing over time. (a) Topic size increment normalized to corpus topic size. (b) Average sentiment scores

Figure 2. Topic sizes and sentiment scores on vaccination over time. (a) Topic size increment normalized to corpus topic size. (b) Average sentiment scores