HALIFAX, N.S. —
Dalhousie University researchers are using machine learning to analyze social media in a study of how people’s emotions progressed during the first wave of the coronavirus and subsequent lockdown.
Rita Orji, Canada Research Chair in persuasive technology and associate professor in the faculty of computer science, Stan Matwin, Canada Research Chair and director of the Institute for Big Data Analytics, and Swarna Weerasinghe, associate professor in the department of community health and epidemiology conducted the research.
When the pandemic broke out, there were a lot of uncertainties around the world, and how to handle it, Orji said in a recent interview.
“It’s something very new,” she said. “People of my age have never had a time in life where almost the majority of people around the world have been issued the mandate that people stay indoors, there has to be social isolation and everything.”
Orji has always been interested in designing technologies to promote health and wellness, so when this started happening, she saw that people took alternative approaches to try to understand what was happening and to connect to others using social media.
“However … we also saw that people are using social media to express different kinds of emotions, comment about how they are feeling about what is going on around them, to react to the events that (are) coming up and prompt a lot of uncertainties happening,” she said.
“So we thought that this is actually and interesting time to try to use this comment – this naturally flowing comment from people – to understand how the pandemic and all that is affecting people. However, beyond that, not just general pandemic, we’re measuring trusted and particular research on how the stay-at-home policy is affecting people.”
She said global leaders had the best interests of their people at heart, but the policies and precautionary measures put in place had no modern precedent.
“What we don’t understand is how are they actually working with people? Is there any way they are affecting people negatively or positively, especially when you think about emotional and mental health.
“So we gathered millions of tweets that people are commenting with the #stayathome hashtag, and we employed machine learning approaches to try to analyze these tweets and divide them into different emotions that are embedded in each of the tweets, using a different emotion classes.”
Orji said there are eight different emotions: trust, surprise, sadness, joy, fear, disgust, anticipation and anger. They tried to use deep learning to try to classify each of the tweets to get different emotions over time.
“And the results were really interesting,” she said.
Different reactions they saw included anger, anticipation and, initially, trust, although that waned over time. Anticipation indicated high anxiety around the proliferation of the coronavirus.
“The results show the capability of this type of framework or the system we developed for decrypting and monitoring people’s reactions or how different policies are affecting people’s emotions over time. And the most important thing about this is that we think that this framework is very important and can be used for public-health monitoring, too.”
The researchers thought this kind of system that can pick free-flowing, unstructured text from social media and be able to decipher emotions can be used to monitor the impact of events and policies.
“And what is interesting here, is that it can enable timely intervention,” Orji said.
“If we have this system to pick these tweets and pass it into our system, so monitor the impact, it is possible that we saw that today the emotion is very down, it’s one of sadness and fear, then that might actually suggest a need for some kind of intervention.”
That timely intervention might be actually able to sway the emotions to more of hope, trust and reassurance, Orji said.
It could also have applications beyond just coronavirus but also other events across many domains where people react and post responses to social media, she said.
The system could also lend itself to a new form of tool for measuring opinions beyond the traditional questionnaire, particularly in this type of situation when gatherings are limited but the potential sample size is on the large scale – potentially world-wide.
“This data was collected especially during the first wave of the coronavirus and the lockdown,” Orji said. “And the data is for the whole world. What we have analyzed right now is looking at the global trends.”
This is the first step in the research, with many avenues to go from here, she said. To be able to isolate data from different countries would potentially give more insight into how people from different regions are faring.
The data collection is ongoing and the first research paper is already submitted for publication.
Credit: Google News