Research highlight

COVID-19: Tracking changes in sentiment on social media during the pandemic

Nature Human Behaviour

March 18, 2022

The first wave of the COVID-19 pandemic had marked negative effects on the sentiments expressed by people on Twitter and Weibo worldwide, whereas lockdowns had a small, positive effect on average, according to an analysis of over 650 million posts, published in Nature Human Behaviour.

The COVID-19 pandemic and associated government policies have affected global health and economies, and this can be tracked using medical and financial data. However, measuring the effect of the pandemic on the psychological well-being of individuals is not so straightforward, as surveys are often expensive, time-consuming to perform and have limited coverage.

Siqi Zheng and colleagues used 654 million posts published on Twitter and Weibo by 10.56 million individuals in over 100 countries between January and May 2020 to develop a daily sentiment index using machine learning. The authors used this index to track how people expressed emotions on social media during the first wave of the COVID-19 pandemic in 2020. Outbreaks of COVID-19 caused a rapid decline in sentiment in all countries studied, with the largest declines occurring in Australia, Spain, United Kingdom and Colombia. On average, there was an observed slower recovery of positive expressed emotions — defined as the number of days taken for the sentiment of a country to recover to half its’ stationary state — ranging from 1.2 days in Israel to 29.0 days in Turkey. By contrast, lockdown policies had a small, positive effect on sentiment expressed in most countries. In countries badly hit by the pandemic, the authors suggest that this result could reflect how comparable or even greater levels of psychological distress would occur by allowing the virus to propagate without such restrictions.

This study demonstrates how data from social media can help us to understand changes in emotion on a global scale, with the authors suggesting that this may be a useful tool for policymakers. However, they also caution that social media users are not representative of the wider population, and therefore their method should be used alongside — and not in place of — traditional surveys.

doi: 10.1038/s41562-022-01312-y

Return to research highlights

PrivacyMark System