Lin analyzes Boston bombing using geocoded Twitter data
According to The Boston Globe, on April 15, 2013, three people were killed and over 260 others injured during the Boston Marathon bombing. In relation to the tragedy, Yu-Ru Lin, an assistant professor of the School of Information Sciences at the University of Pittsburgh, as well as a Bostonian herself, presented a Human-Computer Interaction seminar lecture last Wednesday, titled “The Diffusion of Fear, Comfort, and Solidarity during the Boston bombing.” In the lecture, she used Twitter data to study how communities are affected by tragedies.
“How do we examine fear, its diffusion, and its relationship to other community dynamics?” posed Lin at the beginning of her lecture. She explained that past studies on how catastrophes affect communities rely mainly on surveys, which are unreliable because of time lag and memory bias.
Instead, Lin’s approach uses data obtained from geocoded tweets: “Social media, like Twitter, allows me to sample large populations and study how people respond to real world events. This immediate response would avoid biases of memories.” Lin also noted that the data from these tweets allows her to study larger geographical areas.
She used over 180 million tweets which came from 95 of the largest cities in the world — 60 from the United States and 35 from other countries.
One focus of Lin’s study was fear. In order to determine which tweets express fear and which do not, Lin used the fear lexicon, a collection of words that linguists have determined to be associated with fear.
As an example, Lin explained that in the tweet, “What the hell just happened in Boston?” the word “hell” would contribute to the tweet’s fear index. To demonstrate the validity of this method of indexing, Lin showed that on a graph of fear indices versus time, tweets from the city of Boston showed peaks on the day of the bombing, as well as a few days later, when a manhunt occurred.
To determine how the fear felt by Bostonians corresponded to the fear felt by people living outside Boston, Lin studied shared fear. “To study how people’s fear response is similar to that of the people from Boston, we computed a shared fear as the temporal correlation of fear indices after the bombing,” Lin explained.
Lin also focused on two types of social support: comfort and solidarity. Comfort was measured by the frequency the hashtag #prayforboston was used, and solidarity was measured by the frequency the hashtag #bostonstrong was used.
In Lin’s study, she endeavored to use tweets from the two weeks prior to the Boston Marathon bombing to predict the amount of fear, comfort, and solidarity that was observed in cities outside of Boston after the bombing. The factors which Lin used to try to make these predictions were geoproximity (how close someone lived to Boston), social connections (the strength of social connections, based on the number of replies sent between cities), and visits (if a person visited Boston recently). She looked at these factors both individually and in combinations.
Using these models, Lin discovered that the personal visit factor had the greatest impact on whether a person from outside Boston felt fear during the same periods that Bostonians were feeling fear. She also found that the geo-social combination model was the best predictor for solidarity, although when she looked at just cities which were far from Boston, the best predictor was the visit factor. The best predictor for comfort was, again, the visit factor.
But how does shared fear correspond to comfort and solidarity? Lin found that fear is significantly associated with solidarity, but not with comfort. “So what this means is that, holding other factors constant, cities with more shared fear tend to show more solidarity,” Lin said. This result is interesting, because it suggests that fear could have a productive role in social support.
Although Lin’s study offers a comprehensive picture of community dynamics following the Boston Marathon bombing, she ended her talk with an invitation for collaboration with Carnegie Mellon researchers to further improve the study.