A natural language processing approach for analyzing COVID-19 vaccination response in multi-language and geo-localized tweets

Abstract

Social media platforms, such as Twitter, have been paramount in the COVID-19 context due to their ability to collect public concerns about the COVID-19 vaccination campaign, which has been underway to end the COVID-19 pandemic. This worldwide campaign has heavily relied on the actual willingness of individuals to get vaccinated independently of the language they speak or the country they reside. This study analyzes Twitter posts about Pfizer/BioNTech, Moderna, AstraZeneca/Vaxzevria, and Johnson & Johnson vaccines by considering the most spoken western languages. Tweets were sampled between April 15 and September 15, 2022, after the injections of at least three doses, collecting 9,513,063 posts that contained vaccine-related keywords. To determine the success of vaccination, temporal and sentiment analysis have been conducted, reporting opinion changes over time and their corresponding events whenever possible concerning each vaccine. Furthermore, we have extracted the main topics over languages providing potential bias due to the language-specific dictionary, such as Moderna in Spanish, and grouped them per country. Once performed the pre-processed procedure we worked with 8,343,490 tweets. Our findings show that Pfizer has been the most debated vaccine worldwide, and the main concerns have been the side effects on pregnant women and children and heart diseases.

Keywords

• Natural language processing
• Time series
• COVID-19 vaccines
• Sentiment analysis
• Vaccine hesitancy
• Topic modeling