Multi-Perspectives Systematic Review on the Applications of Sentiment Analysis for Vaccine Hesitancy

"[S]tudying people's perceptions on social media to understand their sentiment presents a powerful medium for researchers to identify the causes of vaccine hesitancy and therefore develop appropriate public health messages and interventions."

Viral propagation of misinformation about vaccines has contributed to shaping the landscape of doubt regarding vaccination - that is, to vaccine hesitancy. Mining the underlying emotions of comments, attitudes, and opinions through artificial intelligence (AI) is one tool for vaccine hesitancy assessment. Sentiment analysis enables the categorisation of public opinion with regard to polarity (e.g., positive, negative, or neutral), emotions (e.g., anger and joy), and/or degree of agreement. This study provides an overview of the use of sentiment analysis to analyse vaccine hesitancy from technological, social, and medical perspectives.

The introduction of the review is structured in a question-and-answer setup, featuring questions such as: What is vaccine hesitancy? What are the factors influencing vaccine hesitancy? What are the consequences of vaccine hesitancy? What is the current state of literature in relation to vaccine hesitancy? What is the current state of computer science literature in relation to vaccine hesitancy? What is sentiment analysis?

The systematic literature review (SLR) involved: a mapping of the peer-reviewed literature, published from 2010 to July 31 2021, indexed in six databases; a summary of the studies' main findings and methods; and an evaluation of various aspects of the studies, including their challenges, motivations, and recommendations, in addition to taxonomy analysis.

Thirty-three articles satisfied the inclusion criteria. The researchers summarise the articles according to:

• Sentiment analysis and vaccine hesitancy from a disease perspective, i.e., vaccine hesitancy linked with (i) general disease and (ii) specific disease - For example, one article cited here discussed systematic delineation of media polarity on COVID-19 vaccines in Africa. The authors of that study collected 637 Twitter posts and 569 Google News headlines or descriptions during the COVID-19 pandemic. Data were analysed using three standard computational linguistic models. Their results revealed that contrary to general perceptions, Google News headlines or snippets and Twitter posts within the stated period were generally passive or positive towards COVID-19 vaccines in Africa, and understanding these patterns in light of increasingly sustained efforts by various media and health actors was possible to ensure the availability of factual information about the pandemic.
• Vaccine hesitancy captured in studies by assessing online discussions and debates - Combating vaccine misinformation requires an understanding of the prevalence and types of arguments being made and the ability to track how these arguments change over time. Eleven studies fell under this subcategory. For example, one systematic scoping review cited here examined the methods 86 articles used to monitor and analyse vaccination-related topics on different social media platforms, along with their effectiveness and limitations. The final results showed that most studies focused on negative and positive sentiments towards vaccination and that they may have failed to capture the nuances and complexity of emotions around vaccination. Eleven studies that assessed online debates in vaccine hesitancy could be further divided into (1) vaccine hesitancy cases and (2) opinion. For example, one article cited here collected 40,359 social media posts on childhood vaccination, classified sentiments, and analysed posts using frequency, trend, logistic regression, and association rules. The authors developed a childhood vaccination ontology; results suggested that childhood vaccination trends in sentiments were affected by news about vaccinations. Posts indicating parents' health belief, vaccination availability, and vaccination policy were associated with positive sentiments, whereas posts of experience of vaccine adverse events were associated with negative sentiments.

Sentiment analysis for vaccine hesitancy contends with three key challenges:

• Technological challenges relate to those faced by computer and data scientists, and they are discussed here in terms of data nature, collection, annotation, and processing.
• Social challenges are faced by members of the public and include: understanding, emotions, beliefs, behaviour, and strategies. For example, there are issues linked to the vagueness of the current promotion strategies and their lack of information and persuasive power.
• Medical challenges relate to health and medicine and fall into the subcategories of understanding, support, and treatment.

The researchers pursue similar analyses of the literature on sentiment analysis for vaccine hesitancy in terms of technological, social, and medical aspects of: (i) motivation (the main benefits of pursuing this particular area of research and exploring its potential benefits) - for example, motivations of the government are discussed in terms of pursuing surveillance, assessment, communications, and information sharing; and (ii) recommendations - for example, in the "social" category, recommendations stress the importance of investigating users who spread misinformation, studying specific causes underlying vaccine hesitancy, using social media to inform the public, and communicating with users.

Implications of the SLR are presented for:

• The technology and computer science discipline: For example, not all social media outlets provide the options to crawl their data, thus hindering future research in the topic. Therefore, more social media platforms should integrate information-crawling application programming interface (APIs) for their public user data to be used for research purposes.
• Social Science: Sample findings from the review:
  • A consensus from the studies reviewed is that the spread misinformation online is one of the main factors contributing to vaccine hesitancy. Therefore, stakeholders are encouraged to explore and assess public sentiment, especially among influential users. Moreover, users who spread misinformation should be investigated.
  • People's confidence in healthcare workers (HCWs) has a significant impact on their vaccination decisions. Consequently, trust must be built by creating a direct hotline between patients and HCWs to address questions and concerns about vaccines.
  • While both lay and expert sources can encourage vaccine uptake, ordinary people may be comparatively more effective in communicating with other people in terms of vaccine uptake.
  • Social media is an outlet to receive and send influential messages to the public. Therefore, encouraging celebrities, influencers, politicians, and even country presidents to be agents is necessary to raise awareness for vaccine uptake via social media.
  • People follow religious leaders for life matters, so these leaders can play a role in promoting people to receive vaccines using social media.
• Medical/public health: The study recommends that sentiment analysis be adopted by governments and institutions that are concerned with public health issues, allowing for real-time examination of people's thoughts, trust levels, and apprehensions about vaccines to devise more effective policies and communication methods. However, policymakers and public health personnel should keep in mind that this tool is still developing and that it has limitations that need to be addressed as follows:
  • A major limitation is the inefficiency of sentiment analysis to catch the distinctions and intricacy of emotions and opinions, with sarcasm as an example. In addition, the misuse of bots to convey anti-vaccine messages or retweets at a high rate may potentially skew conclusions to overestimate the phenomenon or underestimate the effectiveness of the implemented strategies.
  • The current expansion of social media is not evenly spread across countries; this may result in overrepresentation in terms of vaccine hesitancy detection of particular regions - specifically in analyses that do not take geography into account.
  • Policymakers and public health entities may be unable to respond quickly based on tracked trends via sentiment analysis to public opinions that are shifting at a very fast pace. Therefore, collective efforts by all parties and agencies are needed to initiate timely and effective measures.

In conclusion: "Sentiment analysis could be used as a part of wider strategies and in conjunction with surveys and other traditional approaches of gauging community perspectives, with the hopes that sentiment analysis approach could advance in the near future..., taking into account the continuous evolution of the field."