Abstract
This study examines social media content to identify adverse effects of COVID-19 vaccination as perceived by the public. Existing studies did not categorize tweets on vaccine adverse effects as personal experience, informative, or advice-seeking. Authors manually classified tweets into categories and used the data to train four machine learning models. LSTM algorithm yielded the highest accuracy of 90.13%. The LSTM model with GloVe embedding was determined to be most suitable. This research aims to fill a research gap and increase public awareness of COVID-19 vaccine side effects. The study highlights the importance of analyzing social media content to better understand public perception of vaccines.
TopIntroduction
A new coronavirus, severe acute respiratory syndrome coronavirus 2 (SARSCoV-2), was discovered during a recent pneumonia outbreak in Wuhan, Hubei Province, China (Ciotti et al., 2020) in December 2019 (Ciotti et al., 2019). COVID-19, the twenty-first century's most dangerous pandemic, has infected over 304 million people and killed 5.4 million people worldwide (Ferawati, Liew, Aramaki, & Wakamiya, 2022). Fever, cold, headache, bone pain, and breathing issues are the most common symptoms of this viral infection, which can develop into pneumonia (Haleem, Javaid, & Vaishya, 2020). COVID-19 appears to impact our daily lives, enterprises, international trade, and transportation (Donthu & Gustafsson, 2020; Haleem et al., 2020).
The World Health Organization (WHO) continues investigating more than 250 possible SARS-CoV-2 vaccines. More than 80 vaccinations are being tested in clinical studies, with another 180 being tested in preclinical trials (Lentzen et al., 2022). Pfizer vaccines were initially approved by the US Food and Drug Administration (FDA), then by the European Medicines Agency for the European Union and then by several other countries worldwide (Rogliani, Chetta, Cazzola, & Calzetta, 2021). However, some people are still unwilling to have the COVID-19 vaccine, according to a study conducted by the WHO. People believe there are more hazards than benefits because of a relatively new and still little-known pandemic, rapid vaccine development, short drug trial follow-up and vaccines’ side effects (Aloweidi et al., 2021). Even though millions of people have become vaccinated, some people are still sceptical because of vaccination's adverse effects. Side effects include headache, muscle aches, redness and swelling at the injection site, cold, fever, weariness and sleepiness (Lian, Du, & Tang, 2022).
The Internet has played a significant role in people's lives over the last few decades. It currently serves as an essential source of information for about two-thirds of the world's population. People share, discuss, bookmark and network remarkably on media platforms, which has grown as a type of online discourse. Social media like WhatsApp, Twitter, Instagram, YouTube, Snapchat and Facebook are primary public sources for spreading information and news. Due to its ease of reach, speed and use, social media rapidly changes public discussions, defining trends in areas ranging from politics to the environment and the entertainment sector to technology (Asur & Huberman, 2010). Researchers can analyse people's feelings with the help of social media. Most of the population today utilises social media to express their opinions and do it often (Hussain, 2020).
Twitter has become a significant communication platform in recent years in pandemic situations. This social media site has 1.3 billion members and 336 million active users, with 500 million tweets sent per day. Researchers gathered Twitter data to examine a variety of subjects (O'Leary, 2015). To make data collection quicker, Twitter provides Application Programming Interfaces (APIs) (Lian et al., 2022). Twitter mining analyses Twitter message data to predict, detect, or examine contributing factors. Additional information linked to tweets, such as names, hashtags and other attributes, can be analysed using Twitter mining.
Machine learning (ML) enables software to increase prediction accuracy without being explicitly programmed. ML is needed to make systems intelligent enough to accomplish tasks without human involvement, based on learning and continuously increasing experiences to realise the problem complexity and requirement for agility (Alzubi, Nayyar, & Kumar, 2018). ML algorithms take historical data as input to anticipate new output values. ML is important because it lets businesses see patterns in customer behaviour and business operating patterns while also contributing to developing new products. Most of the world's most famous organisations, including Facebook, Google and Uber, use ML to predict their business operations and other patterns (Zhang, 2020).
Key Terms in this Chapter
SVM: An ML model uses a hyper-plane best to partition the datasets in n-dimensional space into categories. SVM is utilised for both regression and classification. Support Vector Regression (SVR), a variant of SVC, is one example of a specific sort of SVM that can be used for specific ML problems ( Gunn, 1998 ).
Machine Learning: With the help of ML, which uses learning and ever expanding the experience to understand issue complexity and the requirement for adaptation, computers can accomplish difficult tasks without human interaction ( Zhang, 2020 ). ML models can be categorised into ten categories based on how an algorithm is trained and the availability of the result during training. These include reinforcement, evolutionary, ensemble, artificial neural networks, instance-based, semi-supervised, unsupervised, supervised, and hybrid learning ( Alzubi et al., 2018 ).
LSTM: The LSTM model is a recurrent neural network, memory-extending model. Recurrent neural networks typically have short-term memory in that they use relevant prior information in the present neural network. The preceding knowledge is utilised in the current task ( Mansoor et al., 2020 ).
Supervised Machine Learning: In order to apply what they have learned in the past to new data and predict future events, supervised ML algorithms utilise labelled examples. Based on the analysis of a known training dataset, the learning technique creates an inferred function to forecast the output values. The system might provide goals for any new input after receiving enough training. In order to identify errors and enable the model to be corrected as necessary, the learning algorithm may also compare its output to the correct, intended result ( Alzubi et al., 2018 ).
Logistic Regression: LR is a classification problem-solving supervised ML technique. With the exception of how they are applied, LR and linear regression are very similar. While LR is used to solve classification problems, linear regression is used to solve regression problems ( Ferawati et al., 2022 ).
Deep Learning: ML can be considered a subset of deep learning. The field relies on studying computer algorithms to learn and advance independently. Deep learning uses artificial neural networks created to simulate how humans think and learn, whereas machine learning uses simpler principles ( Nyawa, Tchuente, & Fosso-Wamba, 2022 ).