2.1. Social Media and Politics

Various research papers observe that media and political information or news influence public opinion widely. Because of this, social media has become an increasingly important topic among researchers. Social media analysis is essential for politicians because it provides valuable insights into public opinion tracking and measuring. Candidates create systems for monitoring and evaluating online activity, such as communication on Facebook, Instagram, and Twitter, and compare it with their conventional ones. Understanding public opinion, identifying key issues that are important to their constituents, targeting specific groups of voters based on their interests and behaviors, and measuring the effectiveness of their campaigns to help politicians respond quickly to crises and manage their reputations are some reasons social media analysis is essential for politicians (Mameli et al., 2022).

Meduru et al. (2017) realised that there is a connection gap between citizens and their government. As a result of this communication gap, policy decisions implemented directly or indirectly by the government may fail. Further research suggests using the Vader Lexicon to analyze opinions using Twitter feeds for political analysis. The use of Twitter to frame the COVID-19 virus was investigated using four methods by leaders of the “Five Eyes” intelligence-sharing group, which includes the United States, the United Kingdom (UK), Australia, New Zealand, and Canada. It has been suggested that implementing a new algorithm utilizing the benefits of several algorithms can increase the accuracy of the algorithms so that they can be effectively used in forecasting.

A sentiment-based classifier using machine learning and a “Fuzzy Set and Rough Set approach” for predicting the results of the 2019 presidential election of Sri Lanka, using tweets from September 1 to November 10, 2019 (two months before the election) for both “Gotabaya Rajapaksa” and “Sajith Premadasa,” was also developed by Perera and Karunanayaka (2022). This study reveals flaws and limits in sentiment-based classifiers regarding political tweets associated with this 2019 presidential election case study. It has been suggested that implementing a new algorithm utilizing the benefits of several algorithms can increase their accuracy, so that they can be effectively used in forecasting. Howard and Parks (2012) suggest that communication researchers are uniquely positioned to analyze the relationship between social media and political transformation in terms of both causes and effects. In Bermingham and Smeaton (2011), when considering the Irish General Election, the case study to investigate the potential of modelling political sentiment through social media data mining, discovered that social analytics using both volume-based measures and sentiment analysis is predictive. Several observations are related to monitoring public sentiment during an election campaign, such as examining a variety of sample sizes, periods, and methods of qualitatively exploring the underlying factors. This could involves investigating how politicians’ mistakes on social media platforms affected people’s opinions of Twitter communication and their responses to other politicians’ tweets (Lee et al., 2020). These results highlighted that political gaffes made by politicians on social media may, ironically, improve the success of their efforts by creating a sense of genuine dialogue.

In a study of presidential elections in Egypt, users from different cultures and backgrounds on social networks such as Facebook and Twitter focused on a large number of textual comments that reflected their views in various situations (Elghazaly et al., 2016). The study used support vector machine (SVM) and Nave Bayesian to classify the Arabic text. The comparison showed that the Nave Bayesian technique has the best accuracy and lowest error rate. The study by Bose et al. (2019) further demonstrate how data twitter can be used and how the national research council Emotion Lexicon can be used to analyze politics. According to the 14th Gujarat Legislative Assembly elections of 2017 in India, a dataset of Twitter messages related to the election was used to predict the chances of party victory. Then, the sentiment was analyzed as positive, negative, and neutral, and the results were summarized.

2.2. COVID-19 Pandemic and Politics

The COVID-19 pandemic has had a significant impact on politics around the world. In some countries, the pandemic has led to political instability and unrest, mainly because governments have been slow to respond to the crisis, and the lack of management and information received by the health sector has led to social unrest. Summarizing these facts, politics has been tested through several variable aspects during the epidemic period. Palmer (2021) investigates the political attitudes and behavior of residents of states in the United States. The data set consists of demographic factors on education, race, age, and social trust, with tested policy specifications, social trust, and relationship with policy outcomes. Statistical diagnostic methods were used to examine outliers’ distribution and the possible influence and impact to uncover essential failings within all models. The analyzes’ assumptions have been meticulously interpreted using data.

Rieger and Wang (2022) examined data from 57 countries between March 20 and April 22, 2020 on how various factors such as education, actual stringency level, media freedom, and death rate affect judgments of government policy responsiveness and how people perceive government responses. Survey evidence shows that conspiracy theorists tend to take government counterattacks to be harsh. Purushothaman and Moolakkattu (2021) look at previous studies, press reports, and press releases by government agencies to gather the necessary data from India. The research questions investigated in their article are government response to the pandemic, containment measures, government consolidation of power and international status-seeking, and foreign policy related to the impact of the pandemic and vaccination diplomacy. Montiel et al. (2021) have mapped the political rhetoric of national leaders and characterized global changes during the COVID-19 pandemic. For this, the main rhetorical stories in the publicly available speeches of 26 countries were used as a sample. Lipscy (2020) proposes a broad investigation centered on causes, reactions, and transformations. The COVID-19 pandemic hence highlights the need for more scholarly attention to the political crisis. International relations have focused on military crises, neglecting non-militarized crises such as pandemics. To understand crises, researchers must examine the threat, uncertainty, and time pressure that impact political and economic outcomes. Another paper discusses predicting responses using political-related blogs as primary data. Yano et al., (2009) examined 40 blog sites from November 2007 to October 2008, and the latent Dirichlet allocation (LDA) topic model was used to generate the topics. This provides a qualitative discussion model on a novel comment prediction task, which uses models to predict which blog users will comment on a given post.

2.3. Methodological Review

Different methods are utilized for evaluating social media data and political facts. To evaluate public opinion on the central UK government during the first phase of the COVID-19 epidemic, structural topic modeling was used to extract topics from 4,000 free-text survey responses collected between October 14 and November 26, 2020 (Wright et al., 2022). Overall, they classified answers as positive or negative using sentiment analysis. Specifically, Bogdanowicz and Guan (2022) identified the 12 most popular topics in our dataset from April 3 to April 13, 2020, such as politics, health care, community, and economy, and discussed their growth and changes over time. While macro-level growth has been experienced over time, it is clear that micro-level changes in topic composition have also been exhibited. The research by Rathnayaka and Rupasingha (2023) outlined the sentiment analysis on Twitter data, focusing on the COVID-19 pandemic and political changes’ impacts on people’s lives. Their work employed both machine learning and deep learning techniques, specifically SVM, artificial neural network, and long short-term memory for the categorization of tweets into positive, negative, and neutral sentiments.

Guo et al. (2016) examine two big data text analysis methods: lexicon-based analysis, the most popular automated analysis approach in social science research, and unsupervised topic modeling (LDA analysis). Analyzing 77 million tweets about the 2012 United States presidential election, the study provides a starting point for scholars to evaluate the effectiveness and validity of journalism and various computer-based methods. Notably, the study by Guo et al. (2016) revealed that the two methodologies significantly produced different results. The LDA-based analysis, in particular, was able to read more tweets and reveal deeper subtleties of the conversation. The LDA-based analysis was also determined to be more valid than the dictionary-based technique based on human coder ratings. The LDA-based analysis is cost-effective since it requires little human labor. This demonstrates that, in general, LDA-based analysis outperforms other computer-based approaches.

Olbrich and Yamshchikov (2018) have chosen probabilistic LDA topic modeling to develop political space, which is widely used in NLP, with a particular focus on developing unsupervised text analysis methods. As LDA can generate meaningful political spaces, new dimensions of political discourse can be identified. Because of the qualitative nature of working with textual data, it is evident that the evaluation of the LDA model’s findings is both quantitative and qualitative. This paper introduces probabilistic latent semantic analysis (PLSA) and LDA as Twitter user opinion extraction methods. It is unrelated to the COVID-19 pandemic and politics, but provides a comprehensive explanation regarding topic formatting (Gandhi, 2021). It utilizes Dirichlet priors for the document-topic and word-topic distributions, allowing for better generalization.

Probabilistic topic models have recently received much attention because of their performance in various text-mining tasks such as retrieval, summarization, categorization, and clustering (Lu et al., 2011). In recent years, probabilistic topic modeling has been frequently applied to research topic modeling to focus on finding hidden topics in large documents.

The comparison summary with previous studies is discussed in Table 1 (Bermingham & Smeaton, 2011; Bose et al., 2019; Elghazaly et al., 2016; Lee et al., 2020; Montiel et al., 2021; Palmer, 2021; Purushothaman & Moolakkattu, 2021; Rathnayaka & Rupasingha, 2023; Rieger & Wang, 2022; Wright et al., 2022; Yano et al., 2009).

2.4. Research Gap

Considering the COVID-19 pandemic, we determine the most frequently discussed political themes on Twitter. The critical point is that we could not find any studies focused on the COVID-19 pandemic and political changes that applied a topic modeling approach to extract the relevant topics. The proposed approach is mainly targeted to fill this research gap. Some other relevant research is limited to one country and uses relatively small data sets. When considering the source of data acquisition, some studies collected data from weblogs, Google forms, and polls conducted in different countries. According to the literature, using public opinion platforms such as social media to get accurate ideas is better. It provides opportunities for learning about and comprehending people, their potential, hobbies, and viewpoints.

Due to the ease of obtaining and accessing data, attention has been drawn to Twitter. Hence, the study’s main focus was on the analysis of tweets using standard topic modeling methods. It is capable of correctly studying changes in politics during the COVID-19 epidemic. The literature-identified LDA and PLSA topic models were more efficient and accurate for the proposed approach. As a result, the entire globe has become the subject of research, obtaining data far more extensively than with prior data collection methods. LDA and PLSA algorithms are projected to discover these study fields.

3.1. Data Collection

Twitter social media was used to collect data relevant to the world’s political changes during the COVID-19 pandemic. Tweets were collected during March 2020 and December 2021, the peak period of the COVID-19 pandemic, and 10,658 tweets were collected through the Twitter application programming interface. First, a Twitter application development account was registered to scratch tweets from Twitter. In the Postman workspace, we made a fork through workspace and sent a fetch request to search the entire archive by providing values for several variables (query=COVID-19, political changes, date=20200301000, toDate=20200630000, and maxResults=100). Comma separated values formatting was used in the data set. Fig. 3 shows a portion of the data set. The keywords used are “COVID-19,” “Political changes,” and “Government decisions.” Data were downloaded in JavaScript Object Notation format by sending a fetch request. Our keywords were chosen to reflect a global focus rather than being specific to any particular country or region. However, the collection process was primarily focused on English-language tweets.

Fig. 3

A portion of the data set.

3.2. Data Pre-Process

Data pre-processing is modifying or removing data before use to ensure or improve performance. This phase of the data mining process is essential. The data was pre-processed using Python libraries with NLP techniques to prepare the data suitable for the analysis. First, we import some required libraries such as “re, pandas, numpy, gensim, spaCy, pyLDAvis.” These modules are available for performing text preprocessing, topic modeling, visualisation, and plotting tasks in the subsequent code. For example, spaCy and Natural Language Toolkit (NLTK) are NLP libraries that provide various functionalities, such as tokenization, part-of-speech (POS) tagging, and dependency parsing. Gensim is a popular Python library for topic modeling analysis. The NumPy library provides support for mathematical operations on arrays and matrices. pyLDAvis is used for visualization. After this, the following mandatory steps are proceeding.

3.3. Feature Extraction

Feature extraction in the LDA topic model is the process of identifying the most important words or phrases for a better understanding of the context (Blei et al., 2003). After cleaning and normalizing the data, we need to convert it into numerical representations for modeling: that is, converting unstructured data into a structured numerical format. This study used Bow to convert into a numerical representation. The bag-of-words model is a way of representing modeling text with algorithms. The code “[id2word.doc2bow (text) for text in texts]” represents this process. The “doc2bow” counts the occurrence frequencies of each distinct word, and “id2word” maps the word_id to a unique word in the corpus using the Gensim library.

3.4. Applying Topic Models

To identify the key themes and topics about political behavior that occurred during the COVID-19 pandemic, this study used LDA and PLSA topic modeling analysis. PLSA and LDA are both topic modeling algorithms used to analyze large collections of documents. The analysis will highlight the most significant changes in political speech as well as people’s opinions on these changes. Here, we examine which method of analysis provided the most accurate information.

The existing techniques, LDA and PLSA, were used to obtain more accurate results effectively and efficiently. The literature-identified LDA and PLSA topic models were more efficient and accurate than other methods for the proposed approach. They are successful methods, and each document is viewed as a mix of multiple distinct topics. The advantage of the LDA technique is that one does not have to know in advance what the topics will look like. LDA automatically provides topics based on patterns in the data. PLSA provides a clear probabilistic interpretation of the relationship between documents and topics. The proposed approach primarily relies on such established techniques such as LDA and PLSA when considering these advantages.

3.4.1. LDA Topic Modeling

LDA is a topic modeling technique that represents the relationship between multiple documents in a corpus. This technique allows a set of observations to be explained by unobserved groups that define why some parts of the data are similar (Alamoodi et al., 2022). It uses a Bayesian approach and assumes a Dirichlet prior over the topic distributions. Fig. 4 shows the algorithm.

 

Fig. 4

Latent Dirichlet allocation algorithm. M, number of documents in the corpus; N, number of words in the document; α, Dirichlet parameter (controls per-document topic distribution); θ_m, document topic distribution; Z_m,n, word topic assignment; W_m,n, observed word; φ_k, topic-word distribution; β, Dirichlet parameter (controls per-topic word distribution).

 

The variable names are defined as,

M is the number of documents in the corpus.

N is the number of words in the document.

K is the number of topics.

α is Dirichlet parameter (controls per-document topic distribution).

θ_m is document topic distribution.

Z_m,n is word topic assignment.

W_m,n is observed word.

φ_k is topic-word distribution.

β is Dirichlet parameter (controls per-topic word distribution).

LDA estimates the proportion of words in a document assigned to a certain topic, and then finds the fraction of words assigned to a topic across all texts. These themes are then qualitatively examined to validate their content relevance.

3.4.2. PLSA Topic Modeling

PLSA is a generative model that assumes each document is formed from a mixture of latent themes. Each word in a document is generated with a specific probability by one of these topics. PLSA aims to estimate the probability distributions for each document’s subjects and terms. It is a topic modeling technique for data analysis with applications in information retrieval and filtering, NLP, machine learning from text, and other fields. The PLSA algorithm generated hidden topics based on the importance of words and the similarity of topics in a document (Khotimah & Sarno, 2019). Fig. 5 shows the algorithm.

 

Fig. 5

Probabilistic latent semantic analysis algorithm. d, number of documents in the corpus; w, number of words in the document; z, number of topics; P(d), probability of selecting a specific document (d) from the entire document collection; P(z/d), topic (z) is present in the selected document (d) with probability; P(w/d) is word (w) is drawn from the topic (z) with probability in a given topic.

 

The variable names are defined as,

d is number of documents in the corpus.

w is number of words in the document.

z is number of topics.

P(z/d) is topic (z) is present in the selected document (d) with probability.

P(w/d) is word (w) is drawn from the topic (z) with probability in a given topic.

4.1. LDA Visualization Analysis

The LDA visualization analysis used the “LDAvis” R package (Sievert & Shirley, 2014) for the visualization, which is intended to explore LDA topic models by varying the weights of top-ranked topic words. Inter-topic distance map is a web-based interactive visual representation of topics computed using LDA. This is the most often used visualizing library.

The topics are plotted on an inter-topic distance map by default; the x and y axes here are labeled “PC1” and “PC2” (principal components 1 and 2). Fig. 6 shows that 15 topics were chosen, and bubbles represented the topics with their frequencies. The size of each bubble indicates the frequency of that topic, and the distance between the bubbles reflects the semantic similarity between topics. The topic format will be successful if it uses large, evenly-spaced bubbles that do not overlap. Fig. 6 shows that topics 2, 7, 9, and 15 are clustered independently, and others overlap event topics. In considering overlapping topics, some keywords are commonly used for each overlapping topics. For example, “covid” and “government” keywords are common to topics no. 10, 11, and 14. However, those clusters cover specific topics, and they are clusters of mutually distinct topics.

Fig. 6

Inter-topic distance map (topic bubbles).

Fig. 7 shows the 30 most relevant terms and ranks them in order of relevance from top to bottom and fills the bar chart. Considering the keywords shown in LDAvis, the words with the highest frequency are “good,” “decisions,” “make,” “look,” “covid,” “trackle,” “people,” “inquiry,” “add,” and “vaccine.”

Fig. 7

Most salient terms.

Fig. 8 (Sievert & Shirley, 2014) shows the interaction of left and right sides in the inter-topic distance map. For example, when topic number 10 is selected in the right panel, the left panel terms with their frequencies are represented by red bars. Primarily, LDAvis enables one to choose a topic and display the most relevant keywords for that topic. Relevance is defined by λ, which is the weight assigned to the probability of the term in a topic relative to its lift. The value is intended between 0 and 1. When λ=1, terms are ranked according to their probabilities inside the topic, whereas when λ=0, terms are re-ranked solely according to their lift; Sievert and Shirley (2014) realized that λ=0.6 was the optimum interpretable result.

Fig. 8

How the inter-topic distance map is displayed when the topic is selected. 1. Saliency(term w)=frequency(w)×[sum_t p(t|w)×log(p(t|w)/p(t))] for topics t. 2. Relevance(term w|topic t)=λ×p(w|t)+(1-λ)×p(w|t)/p(w); data from the article of Sievert & Shirley (2014).

4.2. LDA Topic Analysis

LDA assumes that each document is composed of a distribution of topics, and each topic is composed of a distribution of words. The task of this LDA algorithm is to build the hidden topic (“latent”). Table 3 presents the topics and keywords of tweets in each topic produced from 10,792 tweets. When considering Table 3, the study identified the five key themes and 15 topics about political behavior that occurred during the COVID-19 pandemic. “No” in Table 3 represents the relevant cluster numbers in the Inter-topic distance map in Fig. 6.

As shown in the following Table 3, we defined and interpreted the five theme findings. These five themes are defined based on the results of the inter-topic distance map in Fig. 8 (Sievert & Shirley, 2014).

In Fig. 8 (Sievert & Shirley, 2014),

Fig. 8 (Sievert & Shirley, 2014) visualizes the relevant set of 30 keywords for each topic. Among the 30 keywords, the lengthy red bars at the top show the top keywords for the selected topic [in Fig. 8 (Sievert & Shirley, 2014), selected topic 10]. Based on these top keywords, we manually decided on the name for that topic based on expert knowledge and feedback.

As shown in Fig. 8 (Sievert & Shirley, 2014), we extracted 15 topics, and considering all the extracted 15 topics, we divided them into five themes manually based on expert knowledge and feedback. The decided themes are ‘Government Responses for Managing the COVID-19 Pandemic,’ ‘Government Decisions for COVID-19,’ ‘Public Response to Government Measures for COVID-19,’ ‘Social Influence,’ and ‘Vaccination.’ Considering the visualization of LDA and PLSA, as well as feedback from experts and knowledge from relevant authorities and government agencies, we decided that the themes are accurate and meaningful.

4.3. PLSA Topic Analysis

Considering the topic of the PLAS model, the model we generated 25 topics, and Table 4 shows a list of topics along with their corresponding probabilities in order from highest to lowest. Each topic represents a specific theme or concept, and the probability indicates the likelihood of that topic being present in a given dataset or document. These topics do not give an exact meaning to the dataset. However, this is based solely on the given word probabilities.

The most frequent topics are “stay,” “covid,” and “news,” highlighting the section on health during the COVID-19 pandemic. “Transparency,” “regulation,” “procurement,” “conflict,” and “politics” are keywords that refer to the political content of this data set. Others do not convey a specific meaning relating to our objectives. Most of them are verbs that are related to health protection. This means a significant portion of the analyzed data includes health concerns.

4.4. Word Cloud Visualization

The word cloud library is used here to illustrate the data visualization of pre-processed text from the data set. Words are arranged in no particular order, and the library randomly assigns colors to each word and visualizes word frequency and importance within a given text. Large words appear more frequently in the dataset, while smaller ones appear less often.

An overview of the 15 topics that the study generated is represented in Fig. 9. The size of each word reflects its frequency or importance within the corresponding topic. For example, the prominent words might include “covid,” “suggest,” “say,” “good,” “think,” “go,” “country,” “give,” “trust,” “due,” and “change.” The word cloud provides an overview but does not convey detailed insights.

Fig. 9

Word cloud visualization in latent Dirichlet allocation.

According to Fig. 10, “government,” “covid,” “good,” “people,” “say,” “public,” “health,” and “decision” words are typically the larger words and are prominently shown rather than, “make,” “change,” “go,” “failure,” “tackle,” “look,” “demand,” “globally,” and “vaccine” words are prominently shown. Word clouds may not provide detailed information on word relationships, and they may not provide deep meaning to the dataset.

Fig. 10

Word cloud visualization of the dataset.

4.5. Analyzing the Results of Topic Modeling

Topic model evaluation determines how well a topic model performs its intended function. It is important to determine how well it performs the task for which it was created. When the topic model is used for a quantifiable goal, like classification, it is relatively easy to determine how effective the model is. However, evaluation is more challenging if the model is applied to a more qualitative task, such as examining the semantic themes in an unstructured corpus. Here, we evaluate the topic models using the coherence and Kullback-Leibler divergence (KL) scores.

4.5.1. Coherence Score

By calculating the degree of semantic similarity between high scoring terms in the topic, Topic Coherence measures the score of a single topic (Kapadia, 2019). These metrics help distinguish themes that can be understood semantically from topics that are the result of statistical inference. Among the different coherence measures, this study is based on the C_v measure. This measure builds word content vectors based on their co-occurrences and then determines the score using cosine similarity and normalized pointwise mutual information. The range from 0 to 1 is the C_v coherence score. Fig. 11 shows a graphical representation of the measurement of coherence scores. Higher scores of coherences are always better. This study identified that the optimal number of topics is 15, and its coherence value is 0.37.

 

Fig. 11

Coherence score.

 

4.5.2. Kullback-Leibler Divergence Score

The KL score, often known as the KL score (Heimel, n.d.), measures how different two probability distributions are from one another. KL divergence is a non-negative measure of the difference between two different probability distributions. When approximating a distribution, we can try KL divergence to reduce information loss (Avanesov & Kozlov, 2014).

The KL divergence is on a downward trend in Fig. 12, and the method is getting closer to its goal of minimizing the variance between the distributions. PLSA is hard to train, and this model is based on an extensive collection of documents. Therefore, we can use KL divergence to improve the time efficiency of the training algorithm. The curve’s points represent the KL divergence values at particular iterations. Based on the lowest KL divergence attained, the optimal model can be chosen based on the KL divergence’s stabilization. For instance, the KL divergence is 5.4 in the first iteration (y=0). Indicating that the distributions are becoming more similar or aligned, the KL divergence reduces throughout the course of repetitions.

 

Fig. 12

Kullback-Leibler divergence score graph.

 

Here, the coherence and KL scores are used to measure the “optimal number of topics” for the topic models. In this case, we selected 15 topics belonging to the values with the highest coherence score probability for the LDA, and the lowest score iteration was used, and 25 topics were selected in PLSA.

4.6. Discussion

When discussing the significance of our research findings in comparison to previous studies, the key point is that we were unable to find any prior research that specifically focused on the political changes during the COVID-19 pandemic using a topic modeling approach to extract relevant topics. However, when considering some relevant studies, they are limited to one country and use relatively small data sets. Also, those studies used some Google forms, interviews, discussions, weblogs, or poll data instead of social media data. This proposed approach filled this research gap. The proposed approach used global data targeting the whole world; the data set is relatively large and successfully utilized both LDA and PLSA results for comparison. Moreover, Twitter data was more reliable for this kind of approach since people share their opinions on the platform freely and publicly. Therefore, when comparing the existing studies, the results of the proposed approach are significant and valuable for making more accurate decisions, ideas, and solutions for any relevant issues.

In this study, we extracted the main topics relevant to political changes during COVID-19 using a topic modeling approach. The topics that were identified as being related to political changes during the COVID-19 pandemic are explained in Table 3 and Table 4. As a result, 15 topics under five themes were identified using LDA (Table 3), and 25 topics were extracted through PLSA (Table 4). In order to identify topics, we used the most relevant keywords from each topic. For instance, consider topic no. 1, ‘The travel mandates’ in Table 3, where the keyword “travel” pertains to transportation. The keywords “government,” “follow,” “information,” “mandate,” and “guideline” act as a reminder of the rules and guidances to follow during the pandemic situation. On the other hand, the words “let” and “allow” imply restrictions on travelling. These kinds of suggestions are based on the topics and keywords in Table 3 and Table 4.

When comparing topic modeling approaches, LDA provided more accurate and coherent topics than PLSA. LDA is a valuable tool for comprehending the main discussions and trends surrounding political proceedings for health during the pandemic, as it provides a set of topics represented by their most probable words and their distribution across documents.

Important points excerpted from the Twitter data are that political authorities had to deal with many challenges and concerns brought on by the COVID-19 pandemic, and the focus was on the healthcare industry, which is grappling with the massive issues caused by the virus. The adoption of measures to prevent the spread of the virus and reduce its possible effects on individuals and communities depended heavily on political authorities. Addressing several issues, including virus detection and testing, manipulating vaccines, and healthcare infrastructure, has been given priority. In the discussion of the control of the spread of the virus, themes were cited along two lines: the effort to control the spread and the decision to control the spread of the virus. However, the adoption of decisions to prevent the spread of the virus and reduce its possible effects on individuals and communities depended heavily on political authorities.

Another concern is media influence and their leading actions to improve public health, affection of the economic and education sectors, lockdowns and travel restrictions, and how those issues have changed people’s lifestyles, which frequently came into the discussion. Many users expressed doubts about the necessity and efficacy of these policies, as well as worried about how they might affect the economy and personal freedoms. Users discussed news, facts, and personal experiences about vaccination programs, marketing tactics, and vaccine reluctance. They also conferred with the experts to give their opinions about the significance of vaccination.

The online discussion regarding government decision-making regarding the process of vaccination highlighted the difficulty of managing a public health emergency and the significance of incorporating the general public into the decision-making process. Accordingly, those discussions were highlighted during the pandemic period and the proposed approach successfully extracted and finalized them under the different topics and themes.