TABLE 3. Comparison of Research Approaches.

No	Authors	Search query	Method of data collection	Software used	Aim of data analysis	A qualitative approach to develop themes further
1	[47]	Using a list of 25 hashtags as search terms to fetch tweets (e.g. #coronavirus, #2019nCoV, #COVID19, #coronaoutbreak, #quarantine, #StayHome, #SARsCov2, etc.)	Twitter’s open application programming interface (API)	Python and the NRC Emotion Lexicon	13 topics were identified by topic modeling, and sentiment analysis (SA) was performed.	(1) Becoming familiar with the keyword data, (2) generating initial codes, (3) searching for themes, (4) reviewing potential themes, (5) defining themes, and (6) reporting. The thematic approach relied on human interpretation. Coding in NVivo.
2	[50]	Explicit Covid-19 keywords such as “coronavirus”, and keywords such as “school” and “cancelled” in order to include tweets about a wider array of topics impacted by the pandemic.	Twitter streaming API	ANOVA, VADER software	Clustering & subclustering and sentiment analysis Clustering techniques organized tweet data into 15 high-level themes and 15 specific topics within each theme.	(1) Tweets were embedded with Universal Sentence Encoder, (2) a single label was computed using the eight words with the highest overall frequencies, (3) then manual annotations were provided of the prominent themes that arose, by inspecting small samples of tweets within each cluster. To augment human interpretations of each cluster and subcluster, the authors generated summaries using DistilBART, which aims to generate concise summaries without relying on extractive summarization strategies. The BART-based decoder.
3	[52]	A combination of relevant keywords and hashtags: (#covid OR covid OR #covid19 OR covid19) AND (#vaccine OR vaccine OR #vacine OR vaccine OR vaccinate OR immunization OR immune OR vax).	Not mentioned	Python	Behavioral intentions regarding Covid-19 vaccines were mapped to constructs (capability, opportunity, motivation) in the adapted COM-B model.	(1) The coding schema was developed iteratively based on the definitions of constructs in the adapted COM-B model, (2) two reviewers independently coded 1000 tweets in each round, (3) after completing one round of coding, the two reviewers met with a third reviewer to discuss disagreements and update the coding schema until a consensus was reached. The thematic approach relied on human interpretation. The coding tool was not mentioned.
4	[49]	13 keywords: COVID19, CoronavirusPandemic, COVID-19, 2019nCoV, CoronaOutbreak, coronavirus, WuhanVirus, covid19, coronaviruspandemic, covid-19, 2019ncov, coronaoutbreak, and wuhanvirus.	Tweets were collected by the Panacea Lab	R, RStudio Version 1.4.1103 and the National Research Council of Canada Emotion Lexicon	16 topics were identified by topic modelling and SA was performed.	(1) The textmineR package topic label function was used to generate initial labeling for the topics, (2) the authors labeled topics by reading representative tweets for each topic, (3) through discussions, they further grouped the topics into 5 overarching themes. The thematic approach relied on human interpretation. The coding tool was not mentioned.
5	Present work	Keywords: “covid-19” OR “vaccination” OR “vaccine” OR “covid” OR “coronavirus” OR “SARS-CoV-2” OR “Johnson & Johnson” OR “Moderna” OR “Oxford / AstraZeneca” OR “Pfizer / BioNTech”.	QDA Miner software	ProSuite software	33 topics were identified by topic modeling and then the topics were mapped to determinants.	(1) Topics were labeled to create the first version of labels based on the keywords with the greatest weight, (2) the names of labels were polished through in-depth reading of the most representative topic tweets, and (3) a final set of topic labels was created. The thematic approach relied on human interpretation. Coding in Excel.

Note: For more information on the details of each study (e.g. the goal, findings, etc.), see Table 1.