Monash universties Machine Learning Researcher Caitlin Doogan tells DaveWhile Australians tweeted about panic buying, New Zealanders tweeted their support of strict COVID-19 lockdowns. A Monash University study which analysed close to three million tweets across six countries, has found key trends associated with public attitudes towards the pandemic

The interdisciplinary study was conducted by researchers from the Faculty of Information Technology (IT), together with a frontline medical professional. The researchers analysed Twitter-based discussions of public health measures, such as proper hand-hygiene, social distancing, travel bans and working from home, during the first four months of the COVID-19 pandemic. 


The data was obtained across six countries – Australia, Canada, New Zealand, Ireland, the United Kingdom, and the United States, to determine which public health measures received support from the community and whether public commentary offered insights into potential enablers and barriers towards community acceptance. 


ultra106.5fm is proudly supported by:

In the absence of a vaccine, public health measures have been implemented by governments around the world to slow the spread of COVID-19. In order for governments to maximise adherence towards future public health measures and curb unnecessary behaviours, such as panic buying and spreading misinformation, researchers say clear, consistent and timely communication is paramount.


Key findings of the study included:

  • New Zealand displayed the greatest acceptance of public health measures, while the US showed the lowest. This is in spite of these countries having the most and least restrictive of public health measures respectively.
  • Australian and Irish users perceived the issuing of frivolous fines based on ambiguous rules to be a revenue raising activity. This eroded the public’s trust in the regimes and police.
  • Racially charged language and use of anti-China hashtags to signal tweets about COVID-19 were almost exclusive to the US. 
  • Australia tweeted about panic buying more than any other country, especially about toilet paper and limits on alcohol purchases.
  • The UK were most concerned with at-risk individuals such as immunocompromised and the elderly, while New Zealand was most concerned with vulnerable community groups such as Māori, and those without stable accommodation. 
  • Australians made repeated reference to the decisions made over non-essential service closures, specifically about the restrictions on mourners at funerals and the allowance of hair salons to remain open. 
  • Ireland’s delay in shutting pubs and clubs for St Patrick’s day frustrated users who were juggling working from home with homeschooling children.
  • Anxiety and sadness were expressed in UK tweets. The availability and expense of mental health services was perceived poorly. 

Led by Machine Learning Researcher Caitlin Doogan, together with Professor of Data Science and AI Wray Buntine, Associate Professor of Information Systems Henry Linger, and Dr Samantha Brunt, the team used Machine Learning technology to analyse nearly three million public tweets related to the implementation of public health measures in response to COVID-19.


“In order to analyse this huge amount of data, we employed an unsupervised machine learning technique called topic modelling. Topic models are algorithms that identify co-occuring words in texts, then group those texts together into collections that represent a ‘topic’,” Caitlin Doogan said.


Across the six countries, 22 different public health measures were identified across the millions of tweets. They were then categorised into seven different groups which included: lockdowns; workplace closures; testing and tracing; personal protection; gathering restrictions; travel restrictions and social distancing.


“We used an advanced model called MetaLDA, which was created here at the Faculty of IT, that specialises in accurately modeling topics on content found specifically in tweets. This hybrid methodology is a new way to integrate both computational and qualitative analysis to optimise the accuracy using all the contextual data which we would typically leave out,” added Caitlin Doogan. 


Dr Brunt, who was working in the Emergency Department of Royal Perth Hospital when COVID-19 first appeared in Australia, explains how an analysis of Twitter data can offer timely feedback and insights about public responses to COVID-19 public health measures. 


“Understanding public attitudes towards these NPIs and the factors that promote public understanding and support can provide valuable insights to governments seeking to encourage acceptance of and adherence to these interventions,” said Dr Brunt. 


“These insights can also support government decision-making, implementation, and communication strategies around future NPIs in real-time as well as encourage further discussion about the management of NPI programs for future global health events. Ongoing monitoring of social media would help us ward off some of the side effects of messaging, like people not presenting at ED when they really should be.