Mapping of the Covid-19 Vaccine Uptake Determinants From Mining Twitter Data

Opinion polls on vaccine uptake clearly show that Covid-19 vaccine hesitancy is increasing worldwide. Thus, reaching herd immunity not only depends on the efficacy of the vaccine itself, but also on overcoming this hesitancy of uptake in the population. In this study, we revealed the determinants regarding vaccination directly from people’s opinions on Twitter, based on the framework of the 6As taxonomy. Covid-19 vaccine acceptance depends mostly on the characteristics of new vaccines (i.e. their safety, side effects, effectiveness, etc.), and the national vaccination strategy (i.e. immunization schedules, quantities of vaccination points and their localization, etc.), which should focus on increasing citizens’ awareness, among various other factors. The results of this study point to areas for potentially improving mass campaigns of Covid-19 immunization to increase vaccine uptake and its coverage and also provide insight into possible directions of future research.


I. INTRODUCTION
According to current knowledge, mass vaccination is the only way to contain the spread of the SARS-CoV-2 virus, the cause of the Covid-19 pandemic. To bring this pandemic to an end, a large proportion of the world needs to be immune to the SARS-CoV-2 virus. Herd immunity is a key concept for pandemic control and its extinction [9]. However, to achieve herd immunity and cut the transmission chain, using a vaccine with a claimed 95% efficacy, we need to vaccinate at least 63% to 76% of the population [7]. This required vaccine coverage is certainly very high, and may not be easily attained for many reasons. This is a huge challenge not only for pharmaceutical companies and finite healthcare resources, but also for government agencies and regulatory authorities [8], [9], [31].
Reference [10] highlighted the role of vaccination programmes, which must be effective and widely adopted. The observed poor uptake of vaccines in the population makes it difficult to limit the negative impact of Covid-19 on health worldwide. Statistics show that the percentage of citizens who have received at least one dose of the vaccine in the European Union (EU) is around 50% [6]. Some countries exceed this average, such as Germany -53%, and Finlandalmost 60%; however, vaccination rates are significantly off The associate editor coordinating the review of this manuscript and approving it for publication was Derek Abbott . target. While, previously, the biggest problem with the vaccination program was low supply, today it is low demand. Many people do not want to be vaccinated.
Despite the fact that governments are taking a wide range of measures in response to the Covid-19 outbreak, effective ways to encourage citizens to vaccinate are hard to find. To achieve the goals of the vaccination policy, in addition to overcoming the logistical and supply challenges, it is extremely important to counteract the reluctance to vaccinate, which is steadily growing. Vaccine hesitancy is a complex issue driven by a mix of demographic, social and behavioral factors. Determinants concerning vaccine uptake are complex and context-specific, as they vary according to the time, place and severity of the disease and the vaccine characteristics [5].
Many reviews have focused on the classification of possible determinants of vaccine aversion and the wider uptake of different vaccines, for example, the uptake of the influenza vaccine by older people [3], the tetanus/diphtheria/polio vaccine for children [4] or childhood vaccines till ≤7 years of age [5]. In the face of the current Covid-19 pandemic, a pragmatic methodology (beyond questionnaire experiments) is needed to reach the main determinants of Covid-19 vaccine acceptance, which is lacking in the literature. For that reason, this study aims to fill this gap. The study is based on text data obtained from Twitter regarding vaccines in Poland. By applying (i) a taxonomy model of 5As, and VOLUME 9, 2021 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ (ii) a bottom-up approach during data analysis -the mining of tweets from the public discussion provided the topics, and finally, after further analysis, a set of key determinants of vaccine uptake was obtained, and the model was expanded with another dimension labeled Assurance, thus forming 6As. The proposed approach (i) examines the main determinants of vaccine uptake, (ii) identifies possible root causes of nonvaccination, (iii) outlines the relevance of the determinants for citizens' perceptions, and (iv) can support the subsequent design of robust and evidence-based interventions by governments. Reaching the main determinants of vaccine uptake can help with designing and targeting vaccination strategies, in order to gain extensive acceptance in the population. This is a key path to ensuring a fast liberation from the Covid-19 pandemic.
The main contribution of this paper is (i) the identification of an additional sixth dimension in the 5As taxonomy, labeled Assurance; (ii) a preliminary proof-of-concept of the 6As; (iii) a validation of the usability of textual data from public discussions in identifying and classifying the determinants of vaccine uptake; (iv) the development of a bottom-up methodology for the examined issues.
The remainder of this paper is structured as follows. First, we review the background and relevant literature in Section 2. Section 3 introduces the research methodology. Section 4 presents the empirical results obtained in the study, with a discussion of the findings and implications. Finally, Section 5 concludes the study.

II. THEORETICAL BACKGROUND AND RELATED STUDIES
Vaccine 'hesitancy' is an emerging term in the scientific literature and public discourse (i.a. social media) on vaccine decision-making and the determinants of vaccine uptake. The reasons behind decisions to refuse or delay vaccination are varied and context-specific, thus there is no single form that vaccine hesitancy takes [11]. According to [2], the acceptance and adherence to public health recommendations by the population depend largely on the way people perceive a threat. The study of [12] revealed a comprehensive list of concerns related to the Covid-19 immunization of people who do not wish to be vaccinated. Respondents most frequently reported: lack of proper testing of vaccines (74.1%), vaccine adverse effects (65.1%), lack of vaccine effectiveness (44.9%) and improper transport/storage of vaccines (14%). However, the results of campaigns to encourage vaccination are not only dependent on vaccine efficacy and safety. Effective communication campaigns are needed, based on transparency and focusing on restoring trust in authorities, the government and medical professionals [14]. According to [13], vaccine acceptance among the general public and healthcare workers plays a crucial role in the successful control of the pandemic. We can consider immunization programs to be effective when there are high rates of coverage and acceptance in the population [15]. To achieve this, detecting the determinants of Covid-19 vaccine acceptance is crucial.
Reference [16] distinguished the determinants of Covid-19 vaccine acceptance, based on textual data collected from Weibo, a crucial public opinion platform in China. The main determinants of Covid-19 vaccine acceptance in China included the price and side effects. In turn, the study of [17] aimed to assess the prevalence of the acceptance of the Covid-19 vaccine, and the determinants of this among people in Saudi Arabia. By usage of a questionnaire, the researchers found perceived risk and trust in the health system to be significant predictors of the uptake of the Covid-19 vaccine.
The work of [18] focused on examining Covid-19 vaccine acceptance rates in Russia. The study identified a wide range of factors associated with Covid-19 vaccine uptake, which were grouped into the following main areas: sociodemographic and health-related characteristics, cues to action, perceived benefits and barriers. When the vaccine was proven to be safe and effective, the rate of vaccine acceptance increased. Moreover, gender and income significantly influenced the acceptance rates. Whereas [19] examined the individual, communication, and social determinants associated with vaccine uptake. Their study identified ethnicity, risk perceptions, exposure to different media for Covid-19 news, party identification, and confidence in scientists as factors that would affect Covid-19 vaccine uptake.
A review of previous research on vaccine uptake (see: Table 1) indicates that this phenomenon is increasingly gaining academic attention. Facing the fast-paced dynamic of the coronavirus pandemic, researchers use the different environments to collect data and use a variety of methods for data analysis. The rapid and easily accessible environment of social media, here namely Twitter, is popular and very often used to gain international insights into public opinion on the Covid-19 vaccination. However, a lack of research dedicated to the usage of the 5As framework is clearly visible.
The studies previously referred to above provide evidence that vaccine uptake may be determined by a complex mix of demographic, social and behavioral factors. To order these factors, the present study was based on the 5As taxonomy according to [1]. They identified the determinants of vaccine uptake as 5As dimensions: Access, Affordability, Awareness, Acceptance and Activation. Determinants extracted from a systematic literature review had been assigned to each dimension, and this approach facilitated their understanding. Their study proved that the 5As taxonomy captured all the identified determinants of vaccine uptake. Therefore, in this study we decided to use this framework in our methodology to develop a structured classification.
A sixth dimension, labeled Assurance, was uncovered during the empirical stage of this study. Table 2 includes a definition for each of these six dimensions. By knowing the major determinants of vaccine uptake, actions can be better tailored to effectively improve the success rate of the vaccination program.

III. METHODOLOGY
This section provides the research methodology adopted in the current study. Section A aims to presents the method of data collection. Section B describes data analysis. Section C explains the bottom-up approach taken in the present study.

A. DATA COLLECTION AND PREPARATION
The starting point in the empirical part of the study were textual data obtained from Twitter. Discussions between users on Twitter, which constitute opinions, insights and comments on vaccines, are valuable material that, after appropriate processing, will provide new knowledge. A scraping of Twitter data was conducted via QDA Miner software, using the keywords: , with the period between 1 st to 30 th May 2021. This query was designed to obtain a broad spectrum of data from discussions among Twitter users about vaccinations and vaccines. We collected in total 125 495 tweets only in Polish. The Polish language is so unique that it is not used outside of Poland. The assumption of focusing only on Polish tweets was aimed at: (i) selecting only one country for evaluation as a case study; (ii) providing access to discussions regarding homogeneous government regulations on vaccination; and (iii) guaranteeing the relative universality of the results for other European Union countries, given that Poland is also a member. After collecting the data, we performed the pre-processing steps. Tweets in a language other than Polish were deleted, duplicate or empty tweets were removed, and finally, we obtained a set of 105 849 tweets ready for further data analysis.

B. DATA ANALYSIS
First, topic modeling was performed to extract the latent topics in the tweet data using the QDA Miner software. A 33-topic model was found to be optimal in terms of the average semantic coherence of the model. As a result of this phase, we obtained topics, described by top-weighted keywords. Next, an iterative process of topic labeling was performed.
Second, we employed coding to identify relevant interactions between the topics and then aggregate them into higher-order concepts (categories of determinants). The topics were coded and classified under each dimension of the As framework. For example, the tweet extract ''I came for a vaccination, but it is a pity that the vaccines did not come '' 3 was classified as evidence of the topic concerning problems with delays in Covid-19 vaccine deliveries. When there are problems with the supply of vaccines, people who want to be vaccinated generally have a problem with vaccine uptake. Therefore, this topic was included in the Access dimension. Finally, as a result of this phase, we obtained 17 determinants. Then, each determinant was categorized as a representative of Access, Affordability, Awareness, Acceptance, Activation or other using the definitions of each dimension according to Table 1.

C. THE BOTTOM-UP APPROACH
The methodology developed for this study is presented in Figure 1. The activities performed, and the methods and software used at each stage of the bottom-up approach are discussed in more detail below.

1) DATA COLLECTION
Involves sourcing relevant data according to a chosen set of keywords and a defined time period. For this study they have already been determined in Section A. Data collection was conducted using the commercial software QDA Miner, which is part of the ProSuite software [45]. The ProSuite program provides advanced tools for a thorough analysis of data and consists of the following modules: (i) QDA Miner for qualitative data analysis, (ii) WordStat intended for content analysis and text mining, (iii) SimStat designed for statistical analysis. It also offers the option of scraping tweets. In other words, data extraction from Twitter was automated with QDA Miner. In total, 125 495 tweets were collected in this phase. The following information for each tweet was downloaded: (i) tweet full text, (ii) the numbers of favorites and retweets, (iii) user geolocation; (iv) user description/selfcreated profile, (v) tweet date and hour. In order to check whether the data are balanced, we divided all tweets into subsets (covering a period of 7 days) to identify tweets in the material in terms of a place and date of publication. The content in each subset was then compared to see if the data were evenly distributed. This experiment proved that the data was well balanced. It should be stressed that the research material collected at this stage is represented by unorganized data, with colloquial language, slang, abbreviations or extensions, etc. Thus, the subsequent stage of preparing the data is needed.

2) TEXT PREPARATION
Consists of the following tasks: (a) parsing, which means analyzing data and breaking them down into smaller blocks, which separately can be easily interpreted and managed; and (b) preprocessing, also called text cleaning of data, which includes the following jobs: (i) tokenization, where the words 3 In Polish: ''Przyjechałem na szczepienie, ale szkoda że szczepionki nie przyjechały '' are transformed from the text into structured sets of elements called tokens; (ii) compiling a stop word list, where the words which have low informative value or are semantically insignificant (e.g. and, a, or, the) are eliminated; and (iii) stemming, where the words are reduced to their basic form, i.e. word stems are identified. At this phase, we used the WordStat software. We also detected the language of the tweets and retained only tweets in Polish resulting in a dataset with 105 849 tweet documents for further analysis.

3) TOPIC MODELING
is a method for finding a group of words (i.e topic) from a collection of documents. This is a way to achieve recurring patterns of words in textual data. There are many techniques possible to obtain topic models (e.g. the Latent Dirichlet Allocation, LDA). However, ours was based on an algorithm implemented in the WordStat software. Unsupervised learning was chosen because it is commonly used and allows us to conduct exploratory analyses of large text data in social science research [47]. As a result of topic modeling with the usage of the WordStat software, 33 topics, described by top-weighted keywords, were obtained. Next, an iterative process of topic labeling was performed: (i) topics were labeled to create the first version of labels based on the keywords with the greatest weight, (ii) names of labels were polished through in-depth reading of the most representative topic tweets, and (iii) the final set of topic labels was created. Similar to [47], [49] and [52], our thematic approach relied on human interpretation. Thus, this approach could be significantly influenced by personal understanding of the topics and a variety of biases. The results of this stage are part of the supplementary material: Table B. Next, the proportions of occurring topics were calculated as a percentage (TP, %).

4) AGGREGATION OF TOPICS INTO DETERMINANT CONCEPTS
As a result of an in-depth analysis of textual material, by aggregating topics we created 17 determinants from 33 topics representing some kind of problem. It was assumed that each problem/topic could be linked by several determinants. So-called card sorting [53], which means that each topic written on an individual card was assigned to a logically coherent group, was used for creating a determinant. Then the obtained data were entered into the table. The results were presented in the supplementary material: Table C and  Table D. Two determinants outside the 5As framework were revealed at this stage. These were referred to as Protection and Insurance. A similar type of topic classification, not into determinants but overriding themes, was done in the works [47], [52] and [49]. There are many approaches for extracting knowledge from a short text (tweets). A comparison of selected research approaches can be traced in Table 3.

5) LINKING DETERMINANTS WITH 6As
Having applied the method used in the previous stage, 17 determinants were assigned to suitable dimensions of the 5As model. This analysis resulted in discovering an additional dimension, which was labeled Assurance. Thus, the research extended the model to 6As. The main topics (including the determinants of vaccine uptake emerging from the tweet topics) along with examples of comments were included in Table 5 in Appendix. By following the steps presented in Fig. 1, it is possible to access relevant knowledge and discover hot threads raised in social media discussions regarding the Covid-19 vaccination. This, in turn, provides a good basis for designing governmental guidelines for improving vaccination policies and increasing their effectiveness.

IV. RESULTS AND DISCUSSION
A set of 33 topics was extracted from the large text dataset representing tweets on the topic of the Covid-19 vaccination.
In the next phase of the study, a total of 17 determinants influencing vaccine uptake were identified. They are included in Table 4.
Moreover, the list of topics, extended by sample comments providing evidence for each identified factor, is presented in Table 5 (in Appendix). The calculation of topic proportions (TP%) made it possible to calculate the share of each As dimension (Fig. 2).
The results of this study show that Covid-19 vaccine uptake mostly depends on the dimensions defined as Awareness (39.4 %) and Access (27.3 %) to the vaccine. Awareness covers the availability of a wide range of actual and detailed information regarding vaccines in the population, such as immunization schedules, vaccine side effects, safety and efficacy. Whereas Access is linked to the organization of the national vaccination strategy in terms of the following factors: problems with scheduling vaccinations and long queues, delays in vaccine deliveries, poor organization of vaccinations, too few vaccination points, and localization problems, e.g. too far from home. These findings are consistent with the study of [20], who tested Covid-19 vaccine hesitancy in a representative working-age population in France. Their survey experiment showed that detailed knowledge regarding new vaccine characteristics and the national vaccination strategy determine Covid-19 vaccine uptake. The percentage share of all factors identified under each of the 6As dimensions is presented in Fig. 3.
The following subsections summarize the evidence identified for each dimension of the 6As framework.

A. ACCESS FACTORS ASSOCIATED WITH VACCINE UPTAKE
According to the WHO's guidelines, a COVID-19 vaccine allocation strategy should ensure that vaccines are free at the medical point of service, are allocated transparently, and with a participatory prioritization process. Due to the this, vaccines in EU countries are free of charge, so a determinant related to the price was not included in the Access group. However, the role of access on vaccine uptake was reflected in obstacles concerning scheduling vaccinations, long queues, and delays in vaccine deliveries. These problems, highlighted in the Twitter discussions, related to the improper organization of many steps in the immunization process, are major barriers to convenient access. Thus, they need urgent improvement and reinforcement.
Another group constitutes unclear procedures and regulations. Many problems were reported in the discussions, such as inconsideration of people from risk groups, exclusion of immobile and non-digital groups, and poor organization of vaccinations for the partially disabled, all of which significantly hinder access to vaccination. The location of vaccination points also had an impact on uptake. Prior studies showed that the organization of vaccinations with convenient access, e.g. in a workplace [22] or at a school [23], results in increased vaccine uptake.
The inclusion of help and facilities from the government is also an important determinant of convenient access to immunization. The analysis of the tweets revealed that, especially in the context of elderly people, there is a lack of assistance with registration and reaching the vaccination points. Mobile home vaccination teams would be a good solution.

B. AFFORDABILITY FACTORS ASSOCIATED WITH VACCINE UPTAKE
The affordability factors identified in the present study consist of two main groups of determinants. First, is the price of additional services, which concerns a payment, e.g. assistance with registration and reaching the vaccination points. This is especially true for elderly or disabled people who need the support of third parties to undergo the vaccination procedure. Not everyone can count on free support from their family members. This follows indirectly from the study [40], which found that seniors who lived alone had a lower likelihood of having received the vaccine than those who lived with others. Some have to pay for the help of an assistant in this process.
The second determinant is time cost, which is influenced by the lack of clear rules for the vaccination procedure. Twitter comments identify time losses resulting from unclear laws and regulations. An example of such a tweet is: ''@Szczepimysie Hello. Where should my friend who is allergic and had an anaphylactic shock, register in Pabianice? She was already registered for today and went to be immunized but was refused vaccination due to risk''. 4 Prior studies 4 In Polish: ''@szczepimysie Witam. W jakim miejscu w Pabianicach ma się zarejestrować moja znajoma, która jest alergiczką i miała kiedyś wstrząs anafilaktyczny. Była już zarejestrowana na dzisiaj i poszła na szczepienie ale odmówiono jej szczepienia ze względu na ryzyko''. revealed that time cost was a significant predictor of MMR (measles, mumps and rubella) non-vaccination in university students [24], and was a declared disincentive to receive vaccinations in 22% of health professionals surveyed [25].

C. AWARENESS FACTORS ASSOCIATED WITH VACCINE UPTAKE
As mentioned earlier, the determinant group belonging to the Awareness dimension covers the largest range (39.4%) in the entire As framework. It groups several threads covered in tweets, constituting 'hot' topics. Four determinants are included in Awareness. First, for increased vaccine uptake, people value the availability of actual information. A study of tweets revealed that the continuous volatility and inconsistency of information, the low quality of shared statistical data posted on the administration portal, as well as the lack of transparency of information from the government are factors that need improvement to increase vaccination coverage. The research of [42] stated that respondents reporting higher levels of trust in information from government sources were more likely to be vaccinated.
Second, detailed knowledge about vaccines plays a crucial role. This is in line with the work of [26], who found that more knowledge regarding vaccines improved uptake among health professionals. Moreover, according to the study of [25], people who were given more information concerning personal benefits and risks were more likely to be vaccinated.
Third, another diagnosed determinant is consideration of the vaccination and its side effects. This determinant was also identified in the research of [1] and [27]. The main topics on Twitter concerned fear caused by the increased number of deaths after vaccination, and captured the health risks vs. the usefulness of vaccination. Our findings are similar to the study of [22], who proved that the main reasons given for not receiving the vaccine were the belief that it had significant side effects, and concerns about its effectiveness.
Finally, the last factor was the awareness of the vaccination schedule. Lack of knowledge in this area is an obvious factor contributing to non-vaccination. Thus, thorough information campaigns are necessary so that people do not have to undertake a long search for where to go and at what times to get vaccinated. This is in line with [32], who pointed to an important factor being campaigns, which support people in gaining proper information and help build effective community engagement, and local vaccine acceptability and confidence.
In addition, we found that inconsistent risk messages in terms of the Covid-19 vaccine safety and efficacy from officials, public health experts and individuals, which were expressed in mass media, may contribute to a decrease in the acceptance of vaccination, due to a decline of confidence. This is consistent with the study of [21], who found distrust in vaccine safety to be a crucial determinant of Covid-19 vaccine hesitancy. Twitter users often expressed opinions about vaccine safety and questioned its effectiveness due not only to vaccine novelty, but also other factors (Fig. 4).
There is agreement with many prior studies [2], [25], [26] and [28] that efficacy and safety concerns, including side effects associated with vaccination, can have hugely detrimental effects on the uptake.

E. ACTIVATION FACTORS ASSOCIATED WITH VACCINE UPTAKE
Activation refers to the actions taken that will make individuals more likely to receive vaccines. Three types of incentive techniques have been identified to stimulate activation: (i) prompts and reminders, (ii) workplace policies, (iii) incentives. The first group included two topics with negative sentiment. The need for direct (or telephone) contact especially with seniors regarding vaccination was pointed out, as this group is constantly overlooked in government programs due to digital exclusion [34], [44]. This result is also consistent with [40], who revealed that receiving a reminder from a doctor (67.7%) was an important influence on accepting a vaccination. According to [22], providing reminders to staff in aged care facilities significantly increased influenza vaccine uptake. Thus, sending reminders about vaccination terms to people is a good idea, and according to [33], for the elderly generation, also in the form of a personal letter. In addition, another theme of negative sentiment was the lack of contextualization advertising, best represented by the tweet: ''The vaccine isn't yogurt, but that's a bit how it's advertised??''. 5 In this area, an important element for improvement is the creation of thoughtful advertising. To support an effective launch of new Covid-19 vaccines, a government needs to understand the community's concerns, and design such advertising strategies that will neutralize them, and eventually encourage vaccine uptake. Since ''one size does not fit all'', the work of [41] recommended avoiding generic messages and instead, considering the different emotional states of the community in tailored vaccine communication efforts.
Another determinant, labeled as Workplace policies, included the idea of compulsory vaccination, especially in certain professions (e.g. compulsory vaccination for all medical personnel and teachers). The tone of the tweets reflected the split of opinions on mandatory vaccination from acceptance to outright rejection of such a proposal. Examples were shared of forced vaccination by some employers, and the legal implications of this approach were discussed. The study of [38] suggests that obligatory mandates of the Covid-19 vaccination may be ineffective or, worse still, induce a backlash. In turn, the research of [42] reported that 48.1% of respondents would accept their employer's recommendation to vaccinate. They also claimed an attentive balance is required between educating the public about the necessity for universal vaccine coverage and avoiding any suggestion of coercion.
Finally, the last group of determinants, called Incentives, covered such encouragements as lotteries, Covid certificates, and the development of incentive measures for vaccination (e.g. a discount code to get to the vaccination point). When planning vaccination policies, it is worth considering in-depth the strategy for introducing incentives, as the study of [35] found that financial incentives failed to increase vaccination willingness across income levels. Moreover, [36] claimed that payment for vaccination is morally suspect, likely unnecessary, and may be counterproductive. Similarly, [39] argued that financial incentives are likely to discourage vaccination (particularly among those most concerned about adverse effects), and instead, contingent nonfinancial incentives are the desired approach.

F. ASSURANCE FACTORS ASSOCIATED WITH VACCINE UPTAKE
A few topics mentioned factors associated with vaccine uptake which were not anticipated by the 5As taxonomy, triggering a sixth dimension, which we labeled Assurance.     Three main themes emerged in this dimension: (i) discrimination against people who are not vaccinated, (ii) lack of insurance for severe vaccine adverse reactions, (iii) the need for preliminary medical tests before vaccination. The first of these created the Protection determinant, which includes comments presenting a wide range of discrimination against unvaccinated people (e.g. a curfew and travel ban for the unvaccinated, etc.). According to the public health principle of least harm to achieve a public health goal, policymakers should implement the option that least impairs individual liberties [43]. The next two topics were labeled Insurance. In this group of tweets, there were threads related to the lack of compensation in the case of death related to the Covid-19 vaccination, and insurance in the event of vaccine complications. The necessity of testing people before the vaccination itself was also indicated to diagnose possible contraindications and eliminate post-vaccination complications.
Taking action in the scope described above would certainly increase confidence and contribute to increased vaccine uptake in the population. [37] examined whether compensation can significantly increase Covid-19 vaccine demand. The results showed that, for vaccines, compensation needs to be high enough because low compensation can backfire.

V. CONCLUSION
The goal of this study was to determine whether the five dimensions (5As) of Access, Affordability, Awareness, Acceptance and Activation could correctly cover and organize all the determinants identified from tweets regarding Covid-19 vaccine uptake. This study proved: (i) the existence of a further sixth dimension, labeled Assurance; (ii) a preliminary proof-of-concept of the 6As; (iii) the usability and importance of textual data from public discussions in identifying and classifying the different determinants of vaccine uptake. Besides the above-mentioned contributions of this research, another added value to the theory and literature is also the development of the bottom-up methodology used during data analysis.
The empirical part of the present study showed that opinions expressed on social media, i.e. Twitter, constitute a valuable source of data. Knowledge hidden in this information and the discovered relationships should help design immunization campaigns in such a way as to fulfil the suggested needs of citizens and allay their fears as well. Policymakers need to design a well-researched immunization strategy to remove vaccination obstacles, false rumors, and misconceptions regarding the Covid-19 vaccines. Thus, the knowledge of determinants influencing Covid-19 vaccine acceptance can help to create communication strategies that are much needed to strengthen trust in government and health authorities. The study recognized that those interested in vaccination pay the greatest attention to the determinants in the area of Awareness and Acceptance. For this reason, the promotion of broad and detailed information regarding the vaccines and their side effects, safety and efficacy becomes a key direction in favor of Covid-19 vaccine uptake.
In summary, knowledge about why people avoid the Covid-19 vaccination and which problems could act as obstacles during the immunization process may help government agencies, officials, and other authorities to (i) develop guidance for policies of immunization programs, (ii) create preventative measures against vaccine avoidance, (iii) increase public information campaigns designed to raise confidence in the effectiveness and safety of the vaccine, and finally (iv) design more tailored activities to increase the overall level of vaccine uptake in the population.
However, the present study bears several limitations. First, this research focuses on the discussion via the Twitter platform and includes a short data retrieval period. Data that were collected and reported here are only a snapshot taken at an arbitrarily chosen point in time. These data were scraped in a highly changing environment of social media, with dynamic daily volatility in the perceived threat of the Covid-19 disease and issues of vaccines. Second, the study was narrowed down to only one country. Therefore, a generalization of results is difficult and it can be assumed that other threads may appear on social media discussions depending on the temporal and geographical scope of the study. Third, the study deliberately omitted the performance of a sentiment analysis of tweet data as this was not included in the purpose of the paper. In future, it is worth focusing on a task categorizing tweets for each topic into negative, neutral and positive.
Nevertheless, the 6As taxonomy successfully captured all the determinants of Covid-19 vaccine uptake. Thus, future research may use this taxonomy to structure, classify and compare the significance of each of the 6As in explaining the immunization gap for different vaccines.
In future research, a literature review could also be conducted to reveal current implementation strategies for Covid-19 vaccine promotion and to map them to the 6As framework identified in this study in order to determine gaps in recent research. APPENDIX See Table 5.