Chopra, H., Vashishtha, A., Pal, R., Tyagi, A., & Sethi, T. (2021). Mining Trends of COVID-19 Vaccine Beliefs on Twitter with Lexical Embeddings. arXiv preprint arXiv:2104.01131. https://arxiv.org/abs/2104.01131

University of Michigan - Ann Arbor

Researchers were interested in using temporal and demographic analysis of lexical categories of Twitter discussions to show that COVID-19 vaccine opposition is correlated with misinformation. They separated tweets into categories of emotions and influencing factors and compared the results across 5 countries – India, US, Brazil, UK, Australia. 

Mining Trends of COVID-19 Vaccine Beliefs on Twitter with Lexical Embeddings

- Used Snscrape library to collect more than 1 million tweets from June 2020 to February 2021 related to vaccines through keywords and hashtags
- Created 10 lexical categories for psychometric evaluation of tweets on COVID-19 vaccine – 6 emotions (hesitation, anxiety, sorrow, faith, contentment, urgency) and 4 influencing factors (misinformation, vaccine rollout, inequities, health effects) and made a list of seed words for each category 
- Using cosine similarity, researcher measured how close the tweet contents were to seed words in each of the categories
- To determine correlation, pairwise Pearson correlation, weighted networks of categories and edge-strengths, and community detection were created and implemented 

Learn Before

Related