Mining social media data to see shifting public reactions to COVID-19

From the moment news broke of an unknown, potentially deadly virus, the topic lit up social media channels. Assistant Professor Jia Xue, who uses computational approaches to study social justice issues, immediately began mining rich social media data — including Twitter, Weibo and YouTube users’ discussions and sentiments — to help policymakers and clinical practitioners better understand the public response to the COVID-19, and the psychological consequences of the pandemic.

“By analyzing social media content with a machine learning approach, we can get a rapid, real-time sense of how people are reacting to the pandemic, and how those reactions change as the pandemic evolves,” says Xue, who is cross appointed at the Factor-Inwentash Faculty of Social Work and the Faculty of Information. “This knowledge can then be used for better decision-making by public health authorities today and in the future.”

Assistant Professor Jia Xue says Twitter and Weibo, a Chinese social media platform, are “gold mines” for data that’s constantly renewing.

At U of T’s Artificial Intelligence for Social Justice Lab (AIJ lab), where Xue is the founder and director, the researchers leverage leading-edge technologies to examine intimate partner violence, sexual violence, child maltreatment, school bullying and, now, health information communication around COVID-19 on social media. In addition to identifying dominant themes in Twitter messages, they use sentiment analysis to gain insight into people’s thoughts and emotions.

Xue’s team has been continuously collecting millions of random pandemic-related tweets since early January 2020. She and her collaborators in China have been doing the same for the Chinese equivalent of Twitter, called Weibo. “Twitter and Weibo are gold mines for data that’s constantly renewing,” says Xue, noting that the collected tweets have no personally identifiable information to ensure that users’ privacy is maintained.

According to Xue’s analysis of Weibo messages, negative emotions such as anxiety and depression increased after the declaration of COVID-19 in China, while positive emotions and life satisfaction decreased. On Twitter, the two most predominant emotions her research uncovered in pandemic tweets were fear – related to reports of new cases and deaths – and anticipation – in connection with the anticipation of public health measures such as restrictions and closures.

Now that several months have passed since the pandemic began, Xue’s research has identified a shift in the most discussed COVID-19 themes on Twitter. In the 10 million English tweets her team collected from January 20 to March 7, the spotlight was on subjects such as preventive and protective measures and the economic impact, rather than treatment and symptoms. In the 23 million English Twitter messages collected between March 7 and April 21, the most discussed topics included the need for a vaccine, quarantine orders and social stigma about the virus.

“Our AI-powered method overcomes the limitations of time-consuming, small-scale, retrospective surveys,” says Xue. “We can almost immediately provide a rich source of information for policymakers as well as social workers, psychologists and psychiatrists who’ll be providing therapy for affected individuals and groups.”

Xue has already published four pandemic-focused research papers since March 2020, and several more are in the works. Most recently, she received a Canadian Institutes of Health Research Operating Grant: COVID-19 Rapid Research to explore the increased risk of family violence during COVID-19 quarantine in Canada.

“There will always be debates about the responsible use of big data,” she says. “But we’ve shown it can be used to inform strategies aimed at positive social change. This work on the pandemic is just another example.”

> Click here for links to recently published papers by the Artificial Intelligence for Social Justice Lab

By Megan Easton