Hemant Purohit regularly uses social media to find behaviors about social issues, and he is now applying his knowledge to find insights on behaviors about COVID-19.
Twitter may hold hidden clues to personal health behaviors during the coronavirus pandemic.
Assistant Professor of Information Sciences and Technology Hemant Purohit and his research team are sifting through millions of tweets to gain insight into people’s response to COVID-19 in the Washington D.C., Maryland, and Virginia (DMV) area by using an online tool.
“We are extracting intelligence from social media posts on risks, symptoms, compliance with social distancing, and more relevant information on COVID-19 for the DMV region using a human-artificial intelligence teaming approach,” says Purohit.
The tool, called CitizenHelper, sorts through tweets to identify behaviors that could assist emergency agencies and give them an understanding of the population’s attitudes. It uses artificial intelligence (AI) techniques to filter the posts and then determine the relevance and information level of each tweet. It was developed as part of Purohit’s larger research project, which is funded by the National Science Foundation (NSF). Purohit has also received a new grant from the NSF for this coronavirus effort.
Purohit has teamed up with Steve Peterson from Montgomery County, Maryland’s Community Emergency Response Team (CERT), which is spearheading this initiative and bringing it to other areas of the DMV region, including Fairfax County’s CERT.
The tool helps these CERTs to scale work that would be difficult for humans to do alone. “Humans are good at contextual understanding to filter content but they can’t scale. Machines, on the other hand, are good at scaling, but they don’t deeply understand the context very well, and hence, a human-AI teaming approach is invaluable,” says Purohit.
On their most recent run of the tool, 6,500 tweets were sampled for volunteers to look through to help identify sentiments on preventative measures for COVID-19 and risks people were taking. “About 20 percent of the tweets were relevant, and of that, 60 percent of the communications associated with risk had a negative sentiment,” says Peterson.
There is a huge benefit to looking at social media, says Purohit. “Social media is so real-time, and it gives a real reflection of people’s behavior as opposed to expectations of what people’s behaviors are.”
Purohit and his graduate students Yasas Senarath and Rahul Pandey are also working to improve the tool with Peterson. “This is uncharted work, rarely do you see AI systems being trained in real-time during a pandemic or disaster, usually it is post-disaster. This differentiates CitizenHelper from all other tools,” says Peterson.
Peterson says he hopes this can expand throughout the DMV and beyond. “We are very grateful for this partnership and want it to keep going,” he says.