Seventy articles conducted sentiment/emotion analysis through social media applications to identify public sentiments and opinions toward COVID-19 vaccines. The approaches of sentiment analysis included lexicon-based approaches (n=40), machine learning approaches (n=19), hybrid methods (n=7), and manual coding classification (n=4). Forty studies applied the lexicon-based sentiment analysis method, using predefined lexicons annotated with sentiment polarities (eg, positive, negative, or neutral) to determine sentiments expressed in the parsed text [19-58]. TextBlob and VADER (Valence Aware Dictionary and Sentiment Reasoner) were 2 well-known rule-based lexical sentiment analyzers [19,25,26,33,37-41,43,44,46,49,51,53,56-58]. Moreover, 11 studies further predicted the emotion types expressed in the tweets [20,21,23,26,32,37,45,50,53,54,58], and 8 classified emotions as “trust, surprise, sadness, joy, anticipation, disgust, fear, and anger” using the National Research Council Sentiment Lexicon [20,23,32,37,45,50,54,58].

Nineteen sentiment analysis studies were based on machine learning, and they trained machine learning classifiers by annotating tweets in the data set [59-77]. Machine learning approaches used supervised classification algorithms to extract information regarding sentiment polarity. Classical machine learning models mainly included naïve Bayes, support vector machine, random forest, decision tree, and logistic regression [60,61,67,70,76]. Additional machine learning techniques for polarity classification were Microsoft Azure cognitive services, Amazon Web Services (AWS), and Baidu’s AipNLP [59,64,75]. Deep learning techniques mainly used convolutional neural networks, recurrent neural networks, bidirectional long short-term memory (LSTM), and Bidirectional Encoder Representations from Transformers [63,65,66,68,71,77].

Seven studies employed hybrid methods combining lexicons and machine learning for polarity classification [7,78-83]. Accuracy, precision, recall, and F1 score were typically used to evaluate the performance of classification models [61,68-70]. In terms of model performance, 3 studies showed that LSTM outperformed other classifiers [68,70,78].

These sentiment analysis studies showed that public sentiments were associated with real-time news, internet information, public health events, the number of COVID-19 cases, vaccine development, the pandemic, and announcements of political leaders or authorities [7,21,27,31,51,52,55,56,58,65]. Although public sentiments on COVID-19 vaccines varied significantly over time and geography [7,33,39], positive sentiments were more prevalent than negative ones regarding COVID-19 vaccines [7,19,20,22,25,29,33,41-44,46,47,51,57,60,62,65,70, 72,76,78,84,85], with trust and anticipation being the predominant emotions [20,23,32,37,50,54,58]. However, some other studies found that negative sentiments overwhelmed positive ones, with fear being the dominant emotion [53,59,64,66,71,73,79-82,86]. Positive sentiments were found to be mainly related to increased vaccine coverage, vaccine development, vaccination research, and health services [31,57,67,69], whereas negative sentiments were positively associated with increased COVID-19 cases, misinformation, conspiracy theories, and fear regarding vaccine safety [21,55]. Socioeconomically disadvantaged groups were more likely to hold polarized opinions on COVID-19 vaccines [22]. People with bad experiences during the pandemic were more likely to hold antivaccine opinions [22], whereas comments on posts from health media and hospitals had more positive attitudes [74]. Social media posts on Pfizer and Moderna vaccines appeared to be more positive than posts on COVID-19 vaccines from other manufacturers [24,77,83].

Fifty-three studies applied social media for topic analysis on COVID-19 vaccination. Topic analysis methods included latent Dirichlet allocation (LDA) topic modeling (n=24) [20,23,25,31,34,35,40,47,50,51,54-56,58,60,61,64,87-93], manual coding (n=17) [57,80,94-108], and other algorithms (n=12) [26,43,62,109-117]. Table 2 summarizes the provaccine and antivaccine topics on COVID-19 vaccines present on social media.

Topic analysis identified various attitudes and opinions toward COVID-19 and its vaccine, with main topics focusing on vaccination policy, vaccine development, vaccine administration and access, vaccination propaganda, vaccine efficacy and side effects, vaccine hesitancy, and conspiracy theories [26,43,47,50,55,61,64,90,100,101]. Vaccine objection and hesitancy were generally more prevalent than vaccine support [57], although opinion patterns differed by studies. Vaccine hesitancy mainly stemmed from safety concerns, mistrust in the government and pharmaceutical companies, lack of knowledge, conspiracy theories, skepticism about vaccine development and approval, vaccine ineffectiveness, and loss of freedom [20,25,43,55,56,87-89,93,96-98,100,106,111,112]. The dominant concerns regarding COVID-19 vaccines were safety issues and side effects, such as fear of death and allergic reactions to COVID-19 vaccines [56]. Pain, fever, and fatigue were the 3 most common adverse reactions reported by the public [109,117]. Antivaccine topics varied across social media platforms. For example, the activities of antivaxxers on Facebook and Twitter focused on distrust in the government and allegations regarding vaccination safety and effectiveness, while discussions on TikTok focused on individual freedom [105]. Topics on social media also changed over time. Liu et al found that the prevalence of tweets with positive behavioral intentions increased over time [93].

Additionally, public discussions were mainly driven by COVID-19 vaccine–related news, major social events, pandemic severity, and statements issued by authorities [23,26,34,51,58,87,89]. Regarding information sources, authoritative and reliable information disseminators, such as government agencies, major media outlets, and key opinion leaders, played massively influential roles in polarizing opinions, which can amplify or contain the spread of misinformation among target audiences. Positive discourses were more likely to interact with verified sources, such as news organizations, health professionals, and media/journalists, while negative discourses tended to interact with politicians and personal accounts [40,114].

Three studies used social media to analyze behavioral intention and search behavior for COVID-19 vaccination [118-120]. Positive behavioral intention was influenced by reduced risk of infection, socioeconomic recovery, and normal life recovery. In contrast, negative behavioral intention was associated with misconceptions about vaccines and diseases, trust in natural immunity, distrust in the government and vaccines, and lack of knowledge [120]. Search interests regarding misinformation, generic information about vaccines, and availability of vaccines changed throughout the pandemic [118].

Nine studies conducted dissemination and engagement analysis on social media regarding COVID-19 vaccination [19,51,106,114,121-125]. Social media engagement metrics included popularity based on the number of likes, commitment based on the number of comments, and virality based on the number of post shares. The betweenness centrality score is a traditional social network analysis technique used to discover the most influential users on social media [125]. Observing and analyzing the most attention-capturing tweets may help craft a better vaccine information policy. Compared to other social media platforms, Facebook was the most popular analyzed social media platform, followed by Twitter [106]. COVID-19 posts were more widely disseminated and showed more significant influence than non–COVID-19 posts [124]. Specifically, “antivaccine” groups were highly engaged in the COVID-19 vaccine discussion [121,122], and antivaccine sentiment was especially salient in the political right cluster [122]. Health care professionals had an essential role in supporting vaccine activities, and the highest active professional groups were pharmacists, nurses, physicians, and psychologists [125].

Seven studies assessed the reliability and quality of information about COVID-19 vaccination on social media according to the checklist and quality criteria for health information [125-131]. Reliability and credibility were assessed by the modified Health on the Net Foundation Code of Conduct (HONCode), and quality and reliability were evaluated according to DISCERN criteria. Studies showed that most videos were of high quality with good integrity, comprehensibility, relevance, depth, and accuracy of the information provided, and videos with factual information were higher in quality than those with nonfactual information [127,128,131]. Sources of high-quality videos were pharmaceutical companies, pharmacists, society organizations, and academics, while news provided a high percentage of low-quality videos [128]. COVID-19 vaccination FAQ websites provided quality information, but more effort should be taken to make the content more readable and to update the content [130]. The reputation, expertise, and presentation qualities of the authors were the main criteria for evaluating their credibility, while the credibility of antivaxxers as creators of vaccine-related information was very low [126].

Nine articles [133-141] explored the applications of digital technologies in the strategy of COVID-19 vaccination. Machine learning algorithms were mainly applied to this field to help governments formulate better vaccination strategies in different scenarios. These digital solutions supported and complemented previous technologies, and the main barriers to their promotion lay in their data sources.

Four articles [135,136,138,139] tested COVID-19 mitigation strategies and the best vaccination criteria to minimize the number of infections and deaths. The remaining 5 studies [133,134,137,140,141] focused on priority groups and areas for COVID-19 vaccination. The unified Hierarchical Priority Classification-XGBoost model represented a significant improvement in predicting priorities for COVID-19 vaccination [133]. The Susceptible-Infected-Recovered model and a disease propagation graph simulated the optimum distribution of COVID-19 vaccines based on contact tracing data from cellular networks and Bluetooth signals [140]. Spatial artificial intelligence and satellite imagery were used to identify the location of vulnerable populations and target vaccination populations [141].

Twenty-two articles [142-163] involved the applications of digital technologies in the distribution and delivery of COVID-19 vaccines. Blockchain and algorithms were the leading technologies used in this field to optimize the distribution of COVID-19 vaccines. The prominent innovation feature was to support the existing vaccine distribution system, and the biggest promotion obstacles were blockchain-related technical issues, health care infrastructure issues, and data source issues.

The current platforms failed to meet the various storage and delivery conditions for different brands of COVID-19 vaccines [197-199], and blockchain was increasingly used to monitor the whole process from production to delivery, and to ensure vaccine safety in the supply chain. The blockchain applications in this field included Ethereum [155,156] and smart contracts [156-161], which could manage the data on COVID-19 vaccine distribution and automate traceability. The Ethereum solution was a decentralized application platform built on blockchain technology at a low cost, and smart contracts were automatically executable and code-based transactions on the blockchain, which were secure enough to avoid possible attacks and vulnerabilities [155]. Blockchain combined with unmanned aerial vehicle communication networks can ensure transparency of the vaccine supply chain and mitigate security attacks [157,161]. Additionally, 11 studies [142-152] utilized algorithms, such as LSTM, machine learning regression, and artificial neural networks, to design optimal vaccine distribution strategies, monitor vaccine storage temperature, and address supply chain issues.

Six articles [164-169] applied algorithms in the model prediction of COVID-19 vaccination. Five of them focused on the prediction of vaccination progress and vaccination coverage from a specific country to the global context [165-169]. These algorithms mainly complemented and supported existing algorithms, with no obvious barriers to promotion. Besides the prediction of vaccine coverage, machine learning algorithms were used to analyze data from the vaccine adverse event reporting system to predict the safety of different COVID-19 vaccines across age groups [164].

Eleven articles [170-180] aimed to improve the efficiency and quality of COVID-19 vaccination services through algorithms and mHealth apps, as a support or complement to current vaccination services. Technical issues, such as operator difficulty in adopting these technologies, hindered the promotion of these technologies.

During the pandemic, drive-through clinics had been proposed as one of the effective approaches for temporary mass COVID-19 vaccination [200,201]. Machine learning models can help quickly assess the potential output of and design a smart parking system for drive-through vaccination clinics [170,172]. Digital technologies were also used to examine the accessibility of vaccine registration websites to ensure that the disabled can independently schedule vaccination appointments [179], to develop a multilingual app to facilize people with limited local language skills [174], to schedule people at more suitable vaccination centers [171], to remind about the next vaccination date [176], and to provide personalized emails/messages for vaccination promotion [178].

Thirteen articles [181-193] applied blockchain, mHealth, algorithms, and biometric technologies for COVID-19 vaccination certification. Eight of these studies were considered innovative, but some digital solutions had technical barriers, transaction costs, and ethical barriers to promotion. Many countries promoted vaccination certification to enable individuals to return to normal life [202,203]. The COVID-19 vaccination certificate built on blockchain technology had the advantages of decentralization, interoperability, security, transparency, and antitampering. Biometric technologies, such as face recognition [181,183] and iris recognition [193], were used to identify vaccination status, where deep learning algorithms like the Yolov5 model and convolutional neural networks can help perform this process. However, the promotion of biometric technologies may face ethical and data privacy issues. An artificial intelligence bot was also developed to detect fake vaccine certificates [182].

Three articles [194-196] focused on postvaccination surveillance. Two studies monitored adverse events following immunization through user-initiated reports in mobile apps (Respon and vaxEffect@UniMiB) [195,196], and another study established a dynamic monitoring model on COVID-19 vaccine effectiveness through health code blockchain [194]. As an innovative digital solution, technology realization was the main barrier to rollout.