Leveraging Twitter Data for Real-Time Public Health Responses to Coronavirus

July 23, 2020

To gain a better understanding of the impacts of COVID-19, Professor Gloria Mark of the Donald Bren School of Information and Computer Sciences (ICS) is turning to Twitter. In particular, she and her colleague, Computer Science Professor Chen Li, have teamed up with Suellen Hopfer from UCI’s School of Public Health on a new NSF-funded grant, “Leveraging Twitter Data for Real-time Public Health Responses to Coronavirus: Identifying Affective Desensitization, Loneliness and Depression, and Trust.” Working with ICS students Qiushi Bai, Ted Grover and Yicong Huang, along with public health students Emilia Fields and Magdalen Ramirez, the researchers aim to leverage the immediacy of Twitter data to provide signals in real time of citizen responses to the pandemic.

The team is exploring three social aspects related to the escalation of COVID-19 — desensitization, loneliness and trust — using two datasets of tweets. One includes approximately 2.5 million geotagged tweets with related keywords, mainly collected since January 2020. The other includes about 47 million tweets using relevant keywords such as “coronavirus,” “covid19” and “wuhan,” collected from Jan. 28 through July 16, 2020 (with more tweets being added daily).

“First, we will investigate whether we can detect affective desensitization to COVID-19, through detecting a numbing of emotional reactivity to the dangers of the virus over time and space (geographic location),” says Mark. Affective desensitization refers to a decrease in emotional reactivity to violence or other events that typically would elicit a strong emotional response. “Experiencing affective desensitization can lead people to downplay and ignore public health recommendations,” she says.

Second, as COVID-19 leads to an increase in social distancing, the team plans to look for areas with an association of increased expression of loneliness and increased social isolation and quarantines. The team will explore whether associations exist in certain geographic locations, in rural versus urban areas, or in places with higher incidence of disease.

Third, they will examine the role of trust and erosion of trust in the COVID-19 experience. “We will investigate to what extent public health agencies, the U.S. federal government, crisis management institutions, and institutions for higher education are communicating trust and mistrust during the COVID-19 pandemic outbreak initially and over time and space,” explains Mark. Reviewing citizen narratives trending on Twitter, such as those related to social distancing, school closures, the shortage of test kits, flattening the curve and shutting down group gatherings, the team will address questions such as “what areas of confusion and concern about coronavirus can we identify over time and space, and can these be tied to message content from public agencies?”

This interdisciplinary team, combining expertise in public health communication, human-computer interaction, and data management and analysis, plans to answer these research questions using a mixed-methods approach, incorporating keywords, machine learning, topic modeling, sentiment analysis and qualitative analysis. The goal is to inform public health communication about prevention. As outlined in the grant, “given the rapid spread of the disease, it is critical to identify the social repercussions immediately so that public health agencies and organizations can adapt and respond quickly, dynamically and more effectively to build trust.”

Shani Murray