Analyzing data from social media can identify potential disease outbreaks
A new method to analyze social media data could help predict future outbreaks of diseases and viruses like COVID-19 and the measles.
In a new study, researchers from the University of Waterloo examined computer simulations to develop a new method of analyzing interactions on social media that can predict when a disease outbreak is likely.
The method predicts the tipping point beyond which a series of small incidents, like people espousing anti-vaccine views, become significant enough to cause a larger more important change – like an outbreak of measles.
“We believe our method could be very useful for central decision-makers,” said Chris Bauch, a professor in Waterloo’s Faculty of Mathematics and lead researcher on the study. “Provinces, states, and countries monitoring vaccine sentiment in their jurisdictions could use this method to identify patterns in social media data to help determine areas most likely to have a disease outbreak.
“Once they’ve identified populations that are exhibiting these signals, they can try to build trust and boost vaccine coverage in vaccine-hesitant members of those populations,” said Bauch.
Using a simulated social media network, the team found that there are two reliable early warning signals that precede a disease outbreak: dissimilar joint counts and mutual information.
“A dissimilar joint count is the number of instances of communication between, for example, pro-vaxxers and anti-vaxxers, which we found tends to increase prior to an outbreak,” said Brendon Phillips, a PhD candidate in Waterloo’s Department of Applied Mathematics and co-author of the study describing the new mothod. “Mutual informationmeasures the relationship between someone’s opinion and whether they’re sick or not,” said Phillips.
“We used computer simulations to examine how likely it is that someone who is an anti-vaxxer was infected in the past or that they’re susceptible to infection now.”
In developing the new method, the researchers created a simulated network of individuals with features common to many global childhood diseases, such as measles and chickenpox. They then used computer simulations to assess if a likely outbreak manifests in how often clusters of people who think in similar ways connect.
The researchers found that at this stage in its development, the method can show that a disease outbreak is likely but not exactly when it will come as that depends partly on chance events.