Sentimental and Time Series Study of Coronavirus
Immunization Tweets Using VADER
Vishal Kumar Goar,
1,*
Nagendra Singh Yadav
2
and Manoj Kuri
1
1
Engineering College Bikaner, Bikaner, Rajasthan, 334004, India
2
Bikaner Technical University, Bikaner, Rajasthan, 334004, India
*Email: vishalgoar@gmail.com (V. K. Goar)
Abstract
A suitable platform for sentiment analysis of people is one of the hidden advantages of social channels. This has led
to drawing the focus of various research communities and hence sentimental study has gained much awareness in
recent years. Among the available options, Twitter happens to be the most accepted of all functional platforms.
Identifying the well-defined methodology or technique for sentimental study related to data available on Twitter
concerns the selection of an eligible set of data and such study of results is the prime focus of our research. In this
research, there is an analysis of public sentiments expressed in the Twitter database regarding the Coronavirus
disease (COVID-19) vaccine. With a flood of information-carrying myth and reality about COVID-19 vaccine vegetation
of uncertainties, the component of excitement and fear started growing across the globe. The polarity of sentiments
that could be of any type i.e. neutral, positive, negative when identified on a time scale generates trend analysis for
a suitable approach. After capturing public thoughts, opinions and feelings systematic literature review is performed
and an investigational prototype is generated in order to scatter the sentiments on the inspected data & recognize
the everyday sentiment over the span of the timeline. Documentation of fluctuations in daily sentiments is shown
through time series analysis. This research reflects the set of data related to tweets captured from September 21 -
March 22. As per our findings, the Valence Aware Dictionary and sentiment Reasoner (VADER) sentiment analyzer is
the best and most effective model to get optimal results from the collected sentiments, and the polarity score is
recorded over some time. This research enhances the interpretation of the public’s point of view on coronavirus
immunization and helps them focus on removing COVID-19 from the rest of the world.
Keywords: Coronavirus immunization; Sentimental study; Time series analysis; social media; VADER; COVID-19.
1. Introduction
Coronavirus outbreak has introduced good-sized interest to the healthcare area these days, and it has caused the
replacement of the idea of protection with each element of our existence. Social distancing is a successful practice for
lowering the growth of Coronavirus disease (COVID-19).
[1]
Protection course of action which includes the adoption
of masks, washing palms at several time intervals, and staying cautious concerning intimacy is presently essential. But
those could be the handiest lessen the growth of COVID-19 instead of removing it. With sanctioning of COVID-19
vaccines by renowned pharmaceutical giants such as Pfizer or BioNTech, Moderna, Oxford or AstraZeneca, Covaxin,
and Sputnik V, a component of relief was observed across the world. But soon myths and facts started floating about
the whole vaccination process on social media platforms which provoked some people to remain hesitant about
receiving a vaccine for COVID-19. World Health Organization (WHO) also admitted it by stating this became one of
the biggest threats to global health in 2019.
Nowadays, social media forms i.e. Instagram, Twitter, Facebook, and YouTube have become integral parts of everyday
lives. This has become a valuable resource referred to as social data. Events that happen in everyday life are shared
willingly on media platforms, and anyone is free to write comments and suggestions. People discuss and give their
thoughts about these events. Furthermore, social forums are extensive sources of facts for upcoming and trendy
businesses to get a feel of public perception and obtain reviews about the products they manufacture. A lot of facts
regarding the coronavirus vaccine are available on various social forums. Compared to other social forums, Twitter is
found to be the first pick when comes to information because it provides ample information that is suitable for time
series sentimental analysis.
[2]
Twitter is a well-known microblogging utility that lets a user share and illuminates real-time messages called tweets.
Microblogging services today have become eminent and consistently used platforms. Extraction of data is a
challenging task as there is a use of informal language, non-textual content, dialects, acronyms, multiple punctuation
marks, and emotions used to express their sentiments.
[3]
Tweets obtained from Twitter enable investigators to capture
a large variety of content, thereby giving freedom to gaining insights into early feedback plans of action. There is a
categorization of trending tweets and they are classified into collective categories i.e Tech, News, and sports. Twitter
also uses distinctive features i.e. ashtags, tags using @, emoji, and Hyperlinks.
In the modern era of a data-driven environment, Sentiment analysis has opted as one of the in-demand fact-finding
subjects in the area of NLP (Natural Language Processing) which in turn is closely associated with artificial
intelligence. Some uses of sentimental analysis can be discovered in news articles & product reviews.
[4]
The results of
sentimental findings are implemented in public market investigation and decision-making. In our research, to execute
sentimental analysis, we have considered a set of data captured from Twitter API alongside a tweepy python package
which is required to predict the sentiments from the data.
[5]
In this research, the sentimental analysis technique was put into the collected data and a comprehensive description is
stated. A literature study put forward that several investigators are operational for sentimental analysis on Twitter. In
extension to those research works, our research explains the best suitable way for performing the sentiment analysis
on the Twitter data and time-based analysis on the Twitter trends over the timeline of the COVID-19 vaccine. Sentiment
analysis (SA) is a knowledgeable system of extricating a person's emotions and feelings. It’s far mostly pursued
domains of NLP (Natural Language Processing).
[6]
The Time-based evaluation is a sequence of observations collected
in consistent periods which means emerging models to evaluate the observed time series. In this research, the VADER
(Valence Aware Dictionary and sentiment Reasoner) assesses tweet polarity & classifies tweets with the help of multi-
class sentiment analysis.
[7]
2. Literature review
Alhaji et al. performed their research work with the help of an ML (machine learning model) i.e. Naive Bayes to
perform sentimental analysis on tweets in the Arabic language using Python's NLTK library.
[8]
The hashtag's tweets
were associated with seven government-urged public health initiatives. A huge number of 53,127 tweets were examined
in this study. The number of tweets reflecting positive sentimental analysis was greater than negative ones.
Kaur and Sharma after collecting relevant tweets from Twitter API, thoroughly examined the sentiments related to
both disease and virus of COVID-19.
[9]
They employed ML (machine learning) methods or processes to discover
sentimental emotions in this study. The NLTK library was utilized to accomplish the preprocessing and the text blob
data sets were utilized for Twitter investigation. Various visualizations were used to project the exciting end results
sentiments. In comparison to this research, they implied the ML methodologies to identify products for sentiment
investigation. In addition, we utilized the lexical-based technique for sentiment analysis and performed time series
analysis in our research.
Tweets connected to #corona-virus, according to Prabhakar Kaila et al., were appropriate for applying and evaluating
sentiment analysis of COVID-19.
[10]
They investigated the information acquired in the record named matrix from the
data sets using the LDA (Latent Dirichlet Allocation) technique. Using LDA approaches, a tremendous amount of
information on the COVID-19 infected paramedic was revealed, including positive sentiments such as trust and
negative sentiments such as dread.
[11]
Gilbert et al. developed VADER, a directive enabled sentimental analysis tool which is best fit for sentimental analysis
related to social media.
[12]
SentiWordNet, ANEW (Affective Norms for English Words), the General Inquirer, LIWC
(Linguistic Inquiry & Word Count), and ML techniques which depend on Naive Bayes, Maximum Entropy and SVM
(Support Vector Machine) algorithms were compared to its efficiency for 11 typical state-of-the-art benchmarks. The
development, endorsement, and testing of VADER were identified in the research study.
[13]
To diagnose the sentimental
lexicon utilized in the social domain, the investigators equipped quantitative & qualitative techniques. Findings show
that VADER enhanced the potentials advantages related to LIWC. VADER distinguished itself when compared to
LIWC by being more attentive to social media sentiment expressions.
Medford et al. used the dataset of coronavirus hashtags to look for specific tweets for 2 weeks. i.e. Jan 14 - Jan 28,
2020.
[14]
Application Programming Interface captures the tweet and stores it in the form of plain text in most cases.
This study uncovers and analyses connected frequency terms i.e. vaccination, and infection preventive techniques. The
sentimental study was utilized to assess the sentimental state and dominating sentiment of each tweet. Lastly, with the
help of an unsupervised ML technique, significant themes in tweets are carefully analyzed and discussed over time.
Cherish Kay Pastor et al. express the thoughts and feelings of Filipinos as a result of the intense society quarantine
imposed by the COVID-19 Pandemic, particularly in Luzon.
[15]
Based on the users' tweets, the researcher also
investigates harsh community quarantine and other Pandemic repercussions on current life. To acquire a better sense
of user attitudes from extracted tweets, the Natural Language Processing methodology is frequently employed. The
collected opinions are the data examined in this process.
[16]
In this study, AD Dubey, A. D et al., collected and analyzed tweets from a total of twelve states within a specified time
frame. The tweets were captured from March 11 - March 31, 2022. The purpose of this research is to observe people's
reactions to disease outbreaks in these countries.
[17]
A careful task of pre-processing with the removal of irrelevant information from tweets is performed for a productive
outcome. A ray of hope with positive thinking is observed in these societies, but the sign of grief and pain also floated
among them. Mainly four states of the European continent believe they cannot trust the situation due to the effect of
this pandemic on the huge population.
[18]
Looking at previous studies most researchers used Python's NLTK package and the Twitter API to extract corona-
virus-related tweets.
[19]
Both machine learning approaches and VADER sentiment analysis approaches were
implemented to perform sentiment analysis.
[20]
Other methods, such as LDA (Latent Dirichlet Allocation), were also
used. In this thesis, as per a systematic literature review, we have used the VADER sentiment analyzer to perform
sentiment analysis using NLTK python’s library. Twitter API is utilized to capture the dataset with the help of Twitter.
Time series analysis is conducted for the study of daily sentiments of the people and also to find out the per day tweet
counts.
[21]
3. Methodology
In this study, we used two research methods that are systematic literature review and an experiment method. Starting
with the literature review we carefully analyzed the data and choose the approach based on the results. Followed by
this research questions were experimented with in which the distribution of sentiments was determined.
Adhering to Marcus Gustafsson and Eric Gilbert's guidelines a systematic literature study was conducted to address
RQ1. Several steps were taken to identify appropriate approaches for sentiment analysis. These steps are abbreviated
as ACTION:
1.An Identification of the keywords: Keywords identified in this process are: sentiment analysis, time-based analysis,
and COVID-19 vaccine.
2.Create the search strings: The search string is developed by choosing significant keywords from the keywords
mentioned earlier.
3.Trace the literature: Using a search string various digital database platforms were searched like Diva, Google Scholar,
IEEE, and Research Gate.
4.Inclusion and Exclusion criteria for selection: For better results inclusion and exclusion criteria are applied to the
collected literature. Inclusion criteria are Articles & papers written only in English that too with approaches to
sentiment analysis. Exclusion criteria involve articles with inadequate information.
5. Organize, Evaluate and select the literature: After exercising the inclusion and exclusion criteria, the improvement
is done by meticulously assessing and selecting the collected literature.
6. Nutshell the concluded literature: Here outline of overall findings with representation for analysis is executed.
3.1 Experiment
Now it’s time to develop a model for assorting sentiments and evaluating RQ2 to predict the arrangement of daily
sentiments over a time series. This process is carried out by an experiment. A series of steps adopted in this process
are as follows:
3.2 Preparations for software environment
The development of this model progressed the usage related to Python. The models related to machine learning in this
experiment were developed by using the following Python libraries:
• Python V.3.9: Python is a scripting language that is interpreted, interactive, and object-oriented. It is very legible
and has fewer syntactical constructions than other programming languages.
• NLTK V.3.6.2: A Python package is known for working with human language data and providing a straightforward
interface to lexical resources like WordNet and text processing libraries. These lexical resources are used to
accomplish categorization, tokenization, stemming, parsing, tagging, and semantic reasoning.
• Pandas V.1.0.1: Pandas is a Python module that works with data structures and functions as a data analysis tool.
Pandas perform the entire data analysis pipeline in Python, eliminating the need to use a more domain-specific
language like R.
• Tweepy V.3.10.0: A Python package that connects to the Twitter API and obtains tweets from the platform. This is
used to directly stream real-time tweets from Twitter.
• NumPy V.1.18.1: A fundamental Python computing package that extends the scalability of multi-dimensional arrays
and matrices by providing a large number of high-level computational operations.
• Scikit-learn V.0.22.1: A straightforward and efficient tool for data mining and analysis.
• Matplotlib V.3.1.3: This Python package creates plots, histograms, power spectra, and bar charts, among other
things. In this study, the matplotlib.pyplot package is utilized to plot the measurements.
3.3 Collection of data
As per the basic requirement of this study social media i.e. Twitter has been selected to gather the data sets. We have
described the each of the steps for the execution of the entire work in sequential order:
In the first step, we validate the relationship between Python and Twitter Microblog. Twitter makes its data available
through public APIs which may be accessed via URLs. Python includes a tweepy package that allows accessing
Twitter's data via the API. Calling required libraries, such as Tweepy, is the primary step in this operation. Alphabetic
characters were collected in form of tweets from Twitter. A lot of emotional signs like a laugh, sadness, and even
emojis to express feelings are also included by the users. The data collection is exercised for seven days, and each
day's data is stored in different CSV files. Targeted information was the content & each of the tweets was associated
with the timestamps. The Prime work was to capture the tweets and pass on the tweets to a function that delivers the
sentimental investigation with the help of python's library. Extracting Twitter data from publicly available raw tweets
in a real-time situation is the method used in this process. To collect the data Twitter API was used. Twitter API enables
users to download tweets officially from a user account and save the tweets in a suitable file format. A total of 7,313
tweets were collected which were concerning the COVID-19 vaccine published on Twitter's public message board.
Keywords such as #Pfizer & BioNTech vaccine, #corona vaccine 2020, and #COVID-19 vaccine were used to retrieve
tweets. This is how the management of the most relevant tweets took place.
[22]
3.4 Data overview
As shown in Fig. 2 dataset is extracted consisting of various fields. The various areas like user details and activities are
described here. With 7,783 tweets 16 fields in total are focused. The fields are the user’s name, id, location, description,
followers, friends, favorites, likes, dislikes, verified, created, date, text, hashtag, source, and re-tweets. The important
fields like user_id, user_name, date, text, and hashtags, are majorly required and engaged in analyzing the data for the
sentiment analysis.
3.5 Data pre-processing
As we know that on Twitter, a tweet is a micro-blog message with a limit of 140 characters only. The maximum number
of tweets encompasses i.e. embed URL, plain text, photos, username, and emotions. Miswrite are commonly observed
in them. An unstructured data on COVID-19 is captured with the help of Twitter & later exposed to text cleaning with
screening, filtrate, and lastly, classification in this operation is what this study focuses on.
[23]
This is the reason we
performed a series of pre-processing steps to eradicate irrelevant information from the tweets. For analyzing the text we
needed to remove slang words, HTML characters, stop words, punctuations, URLs, etc.
[24]
For improved accuracy
splitting of attached words is also performed for cleansing.
[24]
The rationale for this is that the cleaner the data is, the
better it is for mining and feature extraction. All duplicate tweets and retweets were deleted from the last illustration
of 14,500 tweets. Each and every tweet was parsed to deliver the core message. The Natural Language Toolkit (NLTK)
of Python was utilized to pre-process this data. To begin, use python to detect and remove specific characters in tweets,
i.e. URLs ("http://url"), retweets, user mentions & inappropriate punctuations. The hashtag (#) frequently describes
the subject of tweets & includes useful information relevant to the tweet's topic, they’re included in the tweet, but the
"#" symbol has been removed.
[25]
cleaned_tweets = [] for tweet in tweets:
# String search - remove searched substring from string # RE for links: r'http\S+
# RE for @mentions: @[A-Za-z0-9]
cleaned_tweet = re.sub(r”http\S+|@[A-Za-z0-9]+”, ““, tweet[0]) # Store in a new list of lists with cleaned tweets
cleaned_tweets.append([cleaned_tweet, tweet[1]])
The tweets were then converted to lowercase, and stop words (words with no essence i.e. is, he, they) were removed.
Such tweets were then separated into separate words, then stemmed using the Porter stemmer. The dataset was ready
for sentiment categorization after these pre-processing steps.
[3]
3.6 Analysis of Tweet sentiment
The attitudes conveyed inside the tweets were categorized with the utilization of the VADER Sentimental Analyzer on
the dataset. In order to categorize our data set, we first constructed a sentiment intensity analyzer (SIA). The feelings
were then determined using the polarity scores approach. The already processed tweets were then categorized as in
sentiments, or compounds by utilizing the VADER Sentimental Analyzer. The compound worth is a useful statistic for
the scalability related to sentiment in a tweet. The compound score is measured by multiplying the valence ratings
related to every term in the lexicon, which is later updated as per the guidelines & standardized to a range of -1 to +1.
The threshold values divide tweets into good, negative, and neutral categories.
[3,12]
Refer to "(1)" for typical threshold
values utilized in our study:
Classification of sentiments:
• Positive sentiments: compound value > 0.000001, assign score = 1
• Neutral sentiments: (compound value > -0.000001) and (compound value < 0.000001), assign score =0
• Negative sentiments: compound value < -0.0000001, assign score = -1
3.7 The KDE distribution for analyzed data
Tweets are separated based on their compound value. The tweet is categorized as a positive tweet when the compound
value is more than the threshold level & as a negative tweet if the compound value is smaller than the threshold level.
In the rest of the situations, it was seen as neutral. As a result, the three categories were created based on emotional
values. Determination of the length of the model input is by the sentiment value, which is essential for model growth.
Followed by this, a summary distribution of all sentiments is also offered by us. Kernel density calculations will be
implemented first before the distribution is projected.
While implementing the KDE graph, the Seaborn (Python data visualization) package founded on Matplotlib furnishes
a high-end interface for generating KDE graphics.
[26,16]
Then, depending on emotion values, the CDF (Cumulative
Distribution Function) is used to observe significant changes in the strength of sensations in data. It gives you the
percent of the normal distribution function that is less than or equal to the random variable you gave. As a result, the
CDF of the standard normal distribution divides overall feelings into sentiments i.e. neutral, negative, and positive
categories built on sentiment values & density.
3.8 Sentiments in word cloud
The frequently recurring set of words in the above-distributed sentiments are found in this study, which includes both
positive and negative sentiments about the tweets. The comments are displayed as a word cloud with a set of sentence
probabilities, which helps to highlight the most often referenced words in the reviews. The word cloud shows the
words which are more likely to appear in the sentence. For each of the leading positive and negative sentiments, a
word cloud is constructed using the 'Word Cloud' packages.
[27]
3.9 Allocation of daily sentiments over each partition of the time series analysis
A time-series overview of daily Twitter volume is used to break the sample timeline into smaller time intervals. Peaks
in Twitter activity are discovered using time series analysis to show the underlying work process over time. This type
of research uses continuous data as feedback to detect changes in situational information about a topic across time.
This method of describing real-time events has been applied to a range of sectors, including economics, the
environment, science, and medicine. To figure out where and when the changes happened, we employed a variety of
methods, including autocorrelation and seasonal decomposition of attitudes. To create independent time series, we
exploited both rapid variations in relative volume and occurrences.
To begin, we divide the daily sentiments into three division periods and distribute them across the timeline for each
partition, measuring the mean & SD (Standard deviation) related to positive & negative sentiments. After separating
these tweets, we develop a model to show the SD and mean for positive and negative attitudes.
3.10 Decomposition of sentiments into systematic components and autocorrelation analysis
To reduce the lags in the built-in model, we employ autocorrelation analysis. The Pandas Series was used in the project.
The Pearson correlation coefficient value is returned by the autocorrelation function (Pandas.Series.autocorr). The
Pearson correlation coefficient is a representation of two variables' linear correlation. The Pearson correlation
coefficient ranges from -1 to 1, with 0 indicating that there is no linear link, >0 indicating a positive association, and
0 indicating a negative relationship. A positive correlation coefficient reflects that 2 variables are in motion in the alike
direction, whereas a negative correlation coefficient reflects that they’re in motion in opposing directions. To
differentiate the data, we utilised a lag=1 (or data(t) vs. data(t-1)) and a lag=2 (or data(t) vs. data(t-2) (t-2). The
autocorrelation plot was then utilized to measure the values of the autocorrelation method (AFC) opposite to various
lag sizes. As the lag value grows larger, we compared fewer and fewer observations. The sum number of monitoring
(T) must be at least 50, & the highest lag value (k) must be less than or equal to T/k, according to the general rule. We
only considered the first 20 values of the AFC because we have 60 observations.
[28-30]
The data was then shown using time series decomposition. A time series could be divided into four dissimilar pieces
using this method: trend, seasonality, residue, and noise. The season_decompose () function, which returns a result
object, should be used. The result object provides an array that may be used to access the four-decomposition data.
[30]
3.11 Analysis of daily trend with events related to that particular date
Prediction of data is done after performing seasonal decomposition and autocorrelation analysis. We segregated our
dataset into “date”, “usernames", “text", and "hashtags” and also added an area as "count" (a routine counter). Finally
merged the data based on the date field to observe the daily analysis of the tweets in our data.
4. Results and discussion
4.1 Results of literature review
To answer RQ1 an SLR (Systematized Literature Review) is executed as reflected in Table 1. The goal is to identify
the most eligible system that accelerates perfect results of sentiment analysis.
Table 1: Results of the literature review.
Title
Findings
VADER: A Stingy Rule-founded
representation i.e. model for Sentimental
inspection of social media Text
A comparison of VADER Sentiment scores and 10 different extremely
popular sentiment analysis tools/techniques was measured that will give
the best performance in all metrics. VADER scored highest among all
with large datasets.
[12]
Sentimental inspection for Tweets in
Swedish
The typical method of sentiment analysis is briefly described in this
paper for evaluation. Classified training data is required when
employing a machine learning approach. The data will subsequently be
used to train an algorithm that will predict the ordering of unknown
data.
Machine learning techniques were explored and tested, which is time-
consuming given the scope of this paper. The VADER sentiment
analyzer was chosen instead.
[31]
Use of VADER and SVM for forecasting
customer reaction sentiment.
Although this research is in a different area, it has been taken into
account because it compares algorithm accuracy. VADER outperforms
machine learning algorithms and lexicon-based techniques such as
Support Vector Machines (SVMs) in terms of accuracy.
A Review of Social Media Posts from
UniCredit Bank in Europe: A Sentimental
inspection Approach
VADER has opted for sentimental inspection in this research since it
performs well on brief documents i.e. Tweets.
[32]
Broad research on Lexicon-founded
Methodologies for Sentimental inspection
This work pertains to a separate domain because it compares the
accuracy of lexicon-based techniques like VADER, Textblob, and
NLTK.
[33]
Hybrid procedure: naive Bayes &
sentimental VADER for inspecting the idea
of mobile unpack video comments
The sentiment analysis in this paper is done using a hybrid strategy that
combines VADER and naive Bayes approaches. The lexical method for
social media text used by Sentimental VADER has a positive impact on
the Naive Bayes classifier in identifying sentiments.
[34]
In the Systematic Literature Review, several publications were found on the sentiment analysis segment that utilized
machine learning and lexicon-founded methodologies (SLR). According to most articles featured a comparison of
machine learning and lexicon-founded techniques.
[12,33-35]
Twitter datasets demand a comparison of algorithms to find
the best one. The VADER is widely considered the most extensively used technique for obtaining the best possible
results for sentiment analysis classification.
4.2 Collected dataset using Twitter API
Table 2 displays a synopsis related to the data set obtained using the Twitter API. The collection of data includes the
following crucial fields: id, user name, date, text, and hashtags, which are all used to analyze the data for sentiment
analysis.
Table 2: Dataset overview.
S. No.
User_name
Date
Text
Hashtags
1
###### ###
20-
12-2020
06:06:4
4
Daikon paste could be used to treat a cytokine
storm, according to the same people.
#PfizerBioNTech https://t.co/xeHhIMg1kF
['PfizerBioNTech']
2
###### #### ######
12-
12-2020
20:17:1
9
Explain why we need vaccination to me again,
@BorisJohnson @MattHancock
#whereareallthesickpeople e #PfizerBioNTech
['whereareallthesickpeople'
, 'PfizerBioNTech']
3
######### # #######
12-
12-2020
20:04:2
9
There haven't been many sunny days in 2020,
but here are a few highlights:
1. #BidenHarris winning #Election2020…
['BidenHarris',
'Election2020']
4
#### ######
12-
12-2020
20:01:1
6
Covid vaccine; You getting it? #CovidVaccine
#covid19
#PfizerBioNTech #Moderna
['CovidVaccine', 'covid19',
'PfizerBioNTech', 'Moderna']
5
#########
###
12-
12-2020
19:30:3
3
#CovidVaccine
States will start getting
#COVID19Vaccine Monday, #US
['CovidVaccine',
'COVID19Vaccine', 'US',
'pakustv', 'NYC',
'Healthcare', 'GlobalGoals']
6
### ##### #########
Together we can win the battle against
#COVID19
[‘covid19’, ‘We4Vaccine’,
‘IndiaFightsCorona’,
‘LargestVaccinationDrive’]
4.3 Outcome of pre-processing the data
Table 3 summarizes the outcomes of the pre-processing procedures applied to the dataset. The number of words in
reviews and vocabulary was greatly reduced as a result of this method. Given this outcome, the pre-processing phase
was critical in assisting the researchers in cleaning up and removing extra words.
Table 3: Result after pre-processing the tweets.
Text
Tokenized
No_stopwords
Stemmed_porter
Stemmed_snowball
Lemmatized
0
The same
[same, folks,
[folks, said,
[folk, said,
[folk, said, daikon,
[folk, said,
folks said
said, daikon,
daikon, paste,
daikon, past,
past, could,
daikon,
daikon
paste, could,
could,
could,
treatcytokin...
paste, could,
paste could
trea...
treatcytok...
treatcytokin...
treatcytoki...
treatcytoki...
1
while the
[while, the,
[world, wrong,
[world, wrong,
[world, wrong, side,
[world,
world has
world, has,
side, history,
side, histori, year,
histori, year, hope,
wrong, side,
been on the
been, on,
year,
hope, bigg...
bigg...
history, year,
wrong side
the, wrong...
hopefully...
hopefully...
of...
2
Russian
[russian,
[russian,
[russian, vaccin,
[russian, vaccin,
[russian,
vaccine is
vaccine, is,
vaccine,
creat, last, 2, 4,
creat, last, 2, 4, year]
vaccine,
created to
created, to,
created, last, 2,
year]
created, last,
last 2 4
last, 2, 4...
4, years]
2, 4, year]
years
3
facts are
[facts, are,
[facts,
[fact, immut,
[fact, immut, senat,
[fact,
immutable
immutable,
immutable,
senat, even,
even, ethic, sturdi,
immutable,
senator
senator,
senator, even,
ethic, sturdi,
enou...
senator,
even when
even, when,
ethically, s...
enou...
even,
you re n...
y...
ethically, st...
4
explain to
[explain, to,
[explain,
[explain,
[explain, needvaccin]
[explain,
me again
me, again,
needvaccine]
needvaccin]
needvaccine]
why we
why, we,
need
needvaccine]
vaccine
4.4 Results obtained after using VADER
The findings of Twitter sentimental inspection utilizing the NLTK and VADER sentimental inspection tools are
described in this section. The VADER Sentiment Analyzer calculated the sentiment scores for each tweet as positive,
negative, neutral, or complex in Table 4.
Table 4: Sentimental outcome of tweets utilizing the Vader
After applying the thresholds indicated in Section 4.4, Table 5 illustrates the categorization of tweets i.e. favorable,
neutral, or negative. We utilized VADER to select the proper thresholds to directly classify tweets i.e. good, neutral,
or negative as indicated per Section 3.6
The overall sentiment score and polarity of each tweet are shown in Fig. 1. This is dependent on the scoring guidelines
and how tweets are classified as positive, negative, or neutral.
Table 5: Overall sentiment polarity for every tweet.
Tidy Tweet
Tidy
hashtags
Sentiment
Positive
Sentiment
Neutral
Sentiment
Negative
Sentiment
Number of words
Folk said daikon past could
treat
cytokinstor…
Positive
0.000001
1.000001
0.000001
8
World wrong side histori
year hope biggest vac…
Negative
0.109001
0.766001
0.125001
21
[{'compound': 0.1531,
'neg':0.000001,
'neu':0.000001,
'pos':1.000001,
'tweet': ‘folk said daikon past could treat cytokinstor...’},
{'compound': -0.5859,
'neg':0.125001,
'neu':0.766001,
'pos':0.109001,
'tweet': ‘world wrong side history year hope biggest vaccine’},
{'compound': 0.0,
'neg':0.000001,
'neu':1.000001,
'pos':0.000001,
'tweet': ‘explain need vaccine where are all the sick people’}]
Coronavirus
sputnikvastrazenecapfizerbi
ontec
Sputnik
astrazeneca
pfizerbiontec
hmoderna
Neutral
0.250001
0.750001
0.000001
9
Fact immut senatevenyour
ethic sturdy enough…
Neutral
0.000001
1.000001
0.000001
20
Explain need vaccin
Whereareallthesickpeopl
Neutral
0.000001
1.000001
0.000001
7
Overall sentiments are distributed into the three different classes i.e. negative, neutral, and positive aligning to their
sentiment values as reflected in Fig. 1, which presents a total number of tweets into three classes: neutral, positive, and
negative as per their sentiment values in the collected dataset. depending on the outcome displayed in Fig. 1, many
tweets in the collected data set demonstrated positive or neutral opinions regarding the COVID-19 vaccine.
Fig. 1: Overall sentiments distribution.
Although, as shown in Fig. 2, 28.2% of the tweets have shown a positive outlook, 18.6% of the tweets have shown a
negative outlook, and 53.2% of the tweets have shown neutral views. Because of the tiny number of tweets, the neutral
proportion was the highest among all other classifications, resulting in unreliable results. The utilization of a generic
lexicon to describe the Twitter data may have led to the belief that the threshold value may give numerous impartial
opinions.
Fig. 2: Doughnut-chart of sentiment classification distribution.
4.5 KDE distribution results for the analyzed data
Fig. 3 shows a KDE plot based on plot data, which shows the estimated distribution of each sentiment. Seaborn, a
Python data visualization toolkit founded on Matplotlib, furnishes a high-end interface related to implementing KDE
visuals. The normal distribution of the sentiments i.e. neutral, negative, and positive over the tweets as per the
sentimental values is shown in Fig. 5. The majority of sentiment values fall between -0.5 and 1.5. For the positive,
negative, and neutral values, we selected green, red, and orange colors, respectively. It's also evident that the majority
of people are indifferent. We can see from the graph below that the distribution of neutral sentiments is higher than the
distribution of positive and negative sentiments across tweets, and that most tweets do not resemble a more positive
or negative view of almost neutral.
Fig. 3: Normal distribution of sentiments across our tweets.
Fig. 4 shows the CDF of the standard normal distribution. The overall sentiments are distributed into positive, neutral,
and negative according to their sentiment values and density.
Fig. 4: CDF of sentiments across our tweets.
4.6 Sentiments results in word cloud
The trigram of 15 statements in Tables 6 and 7 begins with one of the top ten positive and negative tweet words. The
probability of the sentence will appear in a random 'extremely' negative tweet. Positive and negative connotations, as
well as degrees of positive and negatives, are assigned to the terms. The total sentiment of a sentence is calculated by
aggregating the words' sentiments. We may conclude from a few more tweets that it is frequently imperfect, but on
average, it reaches the proper findings.
Table 6: Trigram of 15 sentences one of the top ten positive tweets.
One of the top 10
words
2
nd
Word
3
rd
Word
Probability of sentence
0
Today
Thank
You
1.000000
1
Vaccine
Happy
Dr
1.000000
2
Vaccine
Technology
has
0.835690
3
vaccine
Reduces
the
1.000000
4
first
Vaccination
This
0.666667
5
good
Watched
another
1.000000
6
today
In
and
0.100000
7
so
Here
is
1.000000
8
dose
Done
amp
0.531250
9
vaccine
Safe
COVAX
1.000000
10
grate
To
stop
0.524390
11
vaccine
Grateful
if
1.000000
12
first
Dosage
on
0.500000
13
dose
Done
one
0.631250
14
vaccine
Canada
federal
0.620000
Table 7: Trigram of 15 sentences one of the top ten negative tweets.
One of the top 10
words
2
nd
word
3
rd
word
Probability of
sentence
0
vaccine
Sending
this
0.490678
1
Pfizer
BioNTech
Vaccines
0.125000
2
19
Live
Updates
0.210567
3
vaccine
In
kids
0.166667
4
vaccine
US
already
0.333333
5
covid
Vaccine
Neck
0.314925
6
Vaccine
Tomorrow
little
0.333333
7
people
Including
BAME
0.476557
8
vaccine
Of
course
0.266463
9
Vaccine
Of
his
0.500000
10
vaccine
To
be
0.400000
11
vaccine
Was
dev
0.271429
12
amp
2
nd
do
0.470000
13
The
Event
was
0.352545
14
Pfizer
Covid
Vaccine
0.242857
Fig. 5: Word cloud of the top positive and the negative sentiments.
Fig. 5 shows the most negative sentiments and the most positive sentiments by using the word cloud. In Tables 6 and
7, we used the random colorization scheme to color the terms according to the Probability of the Sentence.
4.7 Distribution of daily sentiments results over each division of the timeline
Table 8 displays the mean and standard deviation (SD) for positive and negative attitudes, separated into three partitions
to disperse daily sentiments along with the timeframe for each partition.
The attitudes are spread daily over each partition, as shown in Fig. 6, as the tweets convey positive and negative
sentiments that surge at different times. For example, the highest opposed end happened on December 14, which
represents the most negative attitudes, whereas the greatest positive sentiments occurred on December 23. However,
the amplitude of the surges decreased after these incidents, lasting only a few days. Besides, the standard deviation (σ)
trend line was consonant all over the duration, while the mean (μ) declined because of the lower number of tweets with
regard to the end of the era.
Table 8: Mean and the SD of the sentiments in each partition.
Partition_1_
Mean
Partition_2_
Mean
Partition_3_
Mean
Partition_
1_SD
Partition_
2_SD
Partition_ 3_SD
Positive
Sentiment
0.106981
0.111546
0.112899
0.154634
0.155414
0.159998
Negative
Sentiment
0.047555
0.051127
0.041657
0.104322
0.103279
0.098980
Fig. 6: Distribution of daily sentiments over the timeline of each partition.
The sentiments of the tweets do not meet statutory requirements in terms of non-constant mean and variance, as seen
in Fig. 6. We have tested our hypothesis on three partitions of our data in the above code cell. It implies that the data
has some patterns.
4.8 Results for autocorrelation analysis and the decomposition of sentiments into systematic components
Fig. 7 shows that the ACF values are within a 95% trust zone (constitute by the solid grey line). It ensures that our data
is free of autocorrelation for lags greater than 0.
Fig. 7: Autocorrelation of positive and negative sentiments.
The trend and seasonality information collected from the series appears to be reasonable in Fig. 8. The residuals are
also intriguing, revealing times in the series with strong variability trends.
Fig. 8: Decomposition of sentiments into trends, level, seasonality, and residuals.
4.9 Day-to-day trend analysis results with events related to that specific date
Fig. 9 depicts the implementation of time series analysis with a graph that reflects the no of tweets per day over the
dates. The X-axis represents the no of tweets every day, while the Y-axis represents the dates. The data is collected
over five months, with each day having a specific quantity of tweets. Assume that on September 15, 2021, there are
139 tweets each day. We gathered recent news updates by comparing them with normal news material and utilizing
trend analysis, which detects the peaks of Twitter activities. This time-based analysis has provided us with news
information. The following is the latest news: (1) The committee suggested acquiring up to 300 million extra quantities
of BioNTech-Pfizer vaccine, (2) The Vaccine was effective against a variant discovered in the UK, (3) Israeli study
finds Pfizer vaccine 85 percent effective after the first shot, and (4) The Presidency of Joe Biden began, which displays
the news on a specific day in these months.
Fig. 9: Day-to-day trend analysis results with events related to that specific date.
4.10 Discussion
RQ1: Which is the best possible way to get ideal results for sentimental analysis classification?
The outcome acquired provokes the suitability of a systematic literature review (SRL). With a non-pre-judgmental
approach, a simple comparison between different known machine learning and lexicon-based method has been
conducted and as an outcome, it is concluded that the lexicon-based method scores better in most of our systematic
literature review (SRL). A comparison model was recommended in various research papers.
[12,31,35]
Considering the
results related to SRL (Systematic Literature Review) in segment 4.1, a meticulous approach is chosen and that is the
VADER sentiment analyzer. VADER holds superiority as a conclusion of so far work. Hence this is the finest fitting
method to execute the sentimental analysis.
The results were further endorsed by the literature review results revealed in section 4.1, which classified training data
as essential for using a machine learning approach. An algorithm will then be taught on this data to figure out and
predict unidentified data classification. The analysis and testing of machine learning algorithms were extremely time-
consuming creating a loss of motivation and focus for better results. So, we choose the VADER approach as a
replacement for the machine learning method.
[31]
RQ2: How do we manage the everyday sentiments distribution on top of the timeline series?
A graph representing the allocation of daily sentiments over the timeline of each partition, as displayed in the findings
section 4.7 clarifies that the data is separated into three divisions primarily on the timeline of the COVID-19 vaccine.
So, in section 4.7, the daily sentiments are allocated over the timeline series of every partition based on the mean and
standard deviation (SD) values. However, when it comes to the model, there are some lags in the results. To correct
these lags, a literature review is conducted on some research. So, by considering the literature study from [7,28,29]
and [30], It is concluded that autocorrelation analysis and seasonal decomposition should be used to repair lags in
time-series models and to check for seasonal trends in our model.
To demonstrate the results that are committed to defining the trends, level, seasonality, and residuals to monitor
seasonal patterns of positive and negative sentiments and also to resolve the lags in the model, autocorrelation analysis,
and decomposition of sentiments are performed. Finally, based on the findings in section 4.8, we may infer that our
data is free of lags because there is a 95 percent confidence interval that confirms the same.
The results are presented in the form of a graph that displays the values. Finally, the results of the daily trend analysis
with events connected with certain dates are displayed in the 4.9 results section, which is a graph representing the
number of tweets each day across five months of Twitter data from 2020 to 2021. This process can be completed by
collecting five months' worth of tweets per day, as well as news and announcements from those months. By
implementing this, we were able to find out the facts at that particular point in time. Several strategies were employed
to discover any beneficial improvements and to make it easier to spot differences quickly.
6. Conclusion
A systematic literature review is undertaken in this study to determine the best possible strategy for performing
sentimental analysis on the Coronavirus vaccination. There was sufficient data to conclude that VADER is a suitable
method for sentimental analysis. As a result, the NLTK and the VADER analyzer were selected to perform a
sentimental analysis of 14,500 messages on Twitter, which uses a multi-classification technique to analyze tweets. To
express and reinforce sentiment intensity, VADER adopts grammatical and syntactical guidelines. The results reveal
that the KDE distribution for each sentiment is i.e. neutral, negative, or positive depending on their sentiment levels.
We may conclude from this study that humans response to sharing the sentiment on social media, especially on Twitter,
transposes every day. This information about the COVID-19 vaccine epidemic reveals how individuals, government
agencies, and social media outlets reported on the incident.
In terms of time-series analysis, we can infer that by calculating standard deviation and mean values, we discovered
various lags and patterns after executing the allocation of daily sentiments over each partition's timeline.
Autocorrelation analysis is used to correct lags in the data, and we may also uncover trends, levels, seasonality, and
residuals by analyzing the sentiments. The news on certain special days of our data has revealed more significant
results in daily trend analysis with events related to the particular day.
During the global outbreak of COVID-19, 140 million tweets were shared by people, organizations, and government
agencies through Twitter. On social media platforms such as Twitter and Facebook, content is often buried beneath the
noise, so extracting meaningful information from large amounts of noisy content is challenging, but once it is cleaned,
this data reveals human feelings and emotions as well as expressions and thoughts. Analyzing it carefully provides a
great deal of insight into the present moods, attitudes, and cultures of many human communities. In order to categorize
the tweets’ sentiment, three types were identified (positive, negative, and neutral).
In this study, the following contributions are made:
• The purpose of this work is to identify a transformation-based multi-depth analyzer tool for sentiment analysis of
tweets regarding the Coronavirus.
• Automated learning of features without being human-supervised by extracting concise sentiment information from
tweets.
• Present an expansive examination between existing ML and DL message grouping strategies and examine the given
gauge results. The proposed model beat on genuine datasets contrasted with all recently utilized strategies.
As social media tends to spread misinformation, health organizations need to develop reliable methods for detecting
Coronavirus precisely in order to prevent false information from spreading. In comparison to similar studies of the
same nature, the proposed approach performed very well on the given dataset and showed greater accuracy. The main
focus of this article was the creation of a new dataset, rather than the efficient classification of users’ sentiments. Hence,
we propose a VADER sentiment analyzer to categorize the user’s sentiments about COVID-19 based on their tweets.
This study presents a clever structure that utilizes data from social media for grasping the public way of behaving
during a significant troublesome occasion of the hundred years.
Conflict of Interest
There is no conflict of interest.
Supporting Information
Not applicable
Use of artificial intelligence (AI)-assisted technology for manuscript preparation
The authors confirm that there was no use of artificial intelligence (AI)-assisted technology for assisting in the writing
or editing of the manuscript and no images were manipulated using AI.
References
[1] S. Boon-It, Y. Skunkan, Public perception of the COVID-19 pandemic on Twitter: sentiment analysis and topic
modeling study, JMIR Public Health and Surveillance, 2020, 6, e21978, doi: 10.2196/21978.
[2] J. Spencer, G. Uchyigit, Sentimentor: Sentiment analysis of twitter data, SDAD@ European Conference on
Machine Learning and Principles and Practice of Knowledge Discovery in Database, 2012, 56–66.
[3] S. Elbagir, J. Yang, Twitter sentiment analysis using natural language toolkit and VADER sentiment, Proceedings
of the International MultiConference of Engineers and Computer Scientists, 2019, 122, 16.
[4] A. Agarwal, B. Xie, I. Vovsha, O. Rambow, R. Passonneau, Sentiment analysis of twitter data, Proceedings of the
Workshop on Language in Social Media (LSM 2011), Portland, Oregon, June 2011, 30–38, Accessed: May 12, 2021.
[5] L. W. Heyerdahl, M. Vray, B. Lana, N. Tvardik, N. Gobat, M. Wanat, S. Tonkin-Crine, S. Anthierens, H. Goossens,
T. Giles-Vernick, Conditionality of COVID-19 vaccine acceptance in European countries, Vaccine, 2022, 40, 1191-
1197, doi: 10.1016/j.vaccine.2022.01.054.
[6] E. D. Liddy, Natural language processing, In Encyclopedia of Library and Information Science, 2nd Ed. NY. Marcel
Decker, Inc., 2001.
[7] R. B. Cleveland, W. S. Cleveland, J. E. McRae, I. Terpenning, STL: A seasonal-trend decomposition, Journal of
Official Statistics, 1990, 6, 3–73.
[8] M. Alhajji, A. Al Khalifah, M. Aljubran, M. Alkhalifah, Sentiment analysis of tweets in Saudi Arabia regarding
governmental preventive measures to contain COVID-19, 2020, doi: 10.20944/preprints202004.0031.v1.
[9] C. Kaur, A. Sharma, Twitter sentiment analysis on coronavirus using Textblob, EasyChair Preprint 2974, 2020.
[10] J. Ling, Coronavirus public sentiment analysis with BERT deep learning, 2020.
[11] A. J. Nair, Veena G, A. Vinayak, Comparative study of Twitter sentiment On COVID-19 Tweets, 2021 5th
International Conference on Computing Methodologies and Communication (ICCMC), April 2021, 1773–1778, doi:
10.1109/ICCMC51019.2021.9418320.
[12] C. Hutto, E. Gilbert, VADER: A parsimonious rule-based model for sentiment analysis of social media text,
Proceedings of the International AAAI Conference on Web and Social Media, 2014, 8, 1, doi:
10.1609/icwsm.v8i1.14550.
[13] N.A Sharma, A.B.M.S Ali, M.A Kabir, A review of sentiment analysis: tasks, applications, and deep learning
techniques, International Journal of Data Science and Analytics, 2025, 19, 351–388, doi: 10.1007/s41060-024-00594-
x.
[14] R. J. Medford, S. N. Saleh, A. Sumarsono, T. M. Perl, C. U. Lehmann, An ‘Infodemic: leveraging high-volume
twitter data to understand early public sentiment for the COVID-19 Outbreak, Open Forum Infectious Diseases, 2020,
7, ofaa258, doi: 10.1093/ofid/ofaa258
[15] C. K. Pastor, Sentiment analysis of Filipinos and effects of extreme community quarantine due to coronavirus
(COVID-19) pandemic, Available at SSRN 3574385, 2020.
[16] A. Chopra, A. Prashar, C. Sain, Natural language processing, International Journal of Technology Enhancements
and Emerging Engineering Research, 2013, 1, 131–134.
[17] A. D. Dubey, Twitter sentiment analysis during COVID19 outbreak, Available at SSRN 3572023, 2020.
[18] K. Khan et al., A study on development of PKL power, Computational Intelligence and Machine Learning,
Proceedings of the 7th International Conference on Advanced Computing, Networking, and Informatics, 2020, 151–
171, doi: 10.1007/978- 981-15-8610-1_17.
[19] N. S. Yadav, V. Goar, Role of Metaverse in Pioneering Healthcare 4.0. In: Chowdhary, C.L. (eds), The metaverse
for the healthcare industry, Springer, Cham, 2024, doi: 10.1007/978-3-031-60073-9_10.
[20] V. K. Goar, N. S. Yadav, C. L. Chowdhary, P. Kumaresan, M. Mittal, An IoT and artificial intelligence-based
patient care system focused on COVID-19 pandemic, International Journal of Networking and Virtual Organisations,
25, 232-251, doi: 10.1504/IJNVO.2021.120169.
[21] I. Roman, A. Mendiburu, R. Santana, J. A. Lozano, Sentiment analysis with genetically evolved Gaussian kernels,
Proceedings of the Genetic and Evolutionary Computation Conference, 2019, 1328–1337.
[22] K. H. Manguri, R. N. Ramadhan, P. R. M. Amin, Twitter sentiment analysis on worldwide COVID-19 outbreaks,
Kurdistan Journal of Applied Research, 2020, 54–65.
[23] K. Jahanbin, V. Rahmanian, Using twitter and web news mining to predict COVID-19 outbreak, Asian Pacific
Journal of Tropical Medicine, 2020, 13, 378, doi: 10.4103/1995-7645.279651.
[24] T. Singh, M. Kumari, Role of text pre-processing in twitter sentiment analysis, Procedia Computer Science, 2016,
89, 549–554, doi: 10.1016/j.procs.2016.06.095.
[25] A. Krouska, C. Troussas, M. Virvou, The effect of pre-processing techniques on Twitter sentiment analysis, 2016
7th International Conference on Information, Intelligence, Systems & Applications, 2016, 1–5.
[26] C. Gallagher, E. Furey, K. Curran, The application of sentiment analysis and text analytics to customer experience
reviews to understand what customers are really saying, International Journal of Data Warehousing and Mining, 2019,
15, 21–47.
[27] E. M. Younis, Sentiment analysis and text mining for social media microblogs using open-source tools: an
empirical study, International Journal of Computer Applications, 2015, 112, doi: 10.5120/19665-1366.
[28] W. McKinney, J. Perktold, S. Seabold, Time series analysis in Python with statsmodels, Python in Science
Conference, Jarrodmillman Company, 2011, 96–102, doi: 10.25080/Majora-ebaa42b7-012.
[29] J. R. Bence, Analysis of short time series: Correcting for autocorrelation, Ecology, 1995, 76, 628–639,
[30] A. Pal, P. K. S. Prakash, Practical time series analysis: master time-series data processing, visualization, and
modeling using python, Packt Publishing Ltd, 2017.
[31] M. Gustafsson, M. Davidsson, Sentiment analysis for tweets in Swedish, Bachelor Degree Project, 2020, 42.
[32] R. K. Botchway, A. B. Jibril, M. A. Kwarteng, M. Chovancova, Z. K. Oplatkov, A review of social media posts
from UniCredit bank in Europe: a sentiment analysis approach, Proceedings of the 3rd International Conference on
Business and Information Management - ICBIM ’19, Paris, France, 2019, 74–79. doi: 10.1145/3361785.3361814.
[33] V. Bonta, N. K. N. Janardhan, A comprehensive study on lexicon-based approaches for sentiment analysis, Asian
Journal of Computer Science and Technology, 2019, 8, 1–6.
[34] V. D. Chaithra, Hybrid approach: Naive Bayes and sentiment VADER for analyzing sentiment of mobile unboxing
video comments, International Journal of Electrical and Computer Engineering, 2019, 9, 4452, doi:
10.11591/ijece.v9i5.pp4452-4459.
[35] A. Borg, M. Boldt, Using VADER sentiment and SVM for predicting customer response sentiment, Expert
Systems with Applications, 2020, 162, 113746, doi: 10.1016/j.eswa.2020.113746.
Publisher Note: The views, statements, and data in all publications solely belong to the authors and contributors. GR
Scholastic is not responsible for any injury resulting from the ideas, methods, or products mentioned. GR Scholastic
remains neutral regarding jurisdictional claims in published maps and institutional affiliations.
Open Access
This article is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License, which
permits the non-commercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long
as appropriate credit to the original author(s) and the source is given by providing a link to the Creative Commons
License and changes need to be indicated if there are any. The images or other third-party material in this article are
included in the article's Creative Commons License, unless indicated otherwise in a credit line to the material. If
material is not included in the article's Creative Commons License and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view
a copy of this License, visit: https://creativecommons.org/licenses/by-nc/4.0/
© The Author(s) 2025
