Researchers hope to help prevent violent events
In the aftermath of the August 2017 “Unite the Right” rally in Charlottesville, nearly everyone in the city was wondering what they could have done to change the outcome, and what they might be able to do to prevent a similar one in the future.
The rally resulted in three deaths and many injuries. And it gained much of its traction through social media posts, although city officials and law enforcement workers had no idea how destructive the weekend would become.
Using their advanced technical skills as data scientists to address the issue, a team of Master of Science in Data Science (MSDS) students at the University of Virginia’s Data Science Institute chose to work on a capstone project in the 2017-18 academic year to increase emergency preparedness by analyzing social media posts.
“I wanted to work on this project because I'm interested in the intersection of social issues and data science,” said MSDS ‘18 student Duncan Rule. “The project provided an opportunity to actually make a difference in the Charlottesville community.”
Rule and fellow MSDS 2018 Cohort students Kia Duan, Nitesh Prakash, and Ni Zhuang worked with professors Don Brown, Phil Bourne and Abigail Flower to create a program that could more accurately assess threat levels from these types of events in the future. The researchers believe that if law enforcement had access to more information, emergency response to the Unite the Right rally and its aftermath would have been more efficient and would have resulted in fewer injuries.
The team analyzed posts from social media sites such as Twitter and Discord to find common phrases and sentiments regarding the rally. Using text mining, they chose topic words such as ‘white supremacist’ and ‘neonazi event’ to identify organizer and event attendee communications.
“Our group focused mainly on text mining techniques because of the wealth of social media data available,” Rule said. “Specifically, we used topic models and sentiment analysis to automatically detect events before they happen, and provide a user with analysis on the nature/severity of those events.”
With this information, the team is worked to create a predictive model that can be applied to other potentially threatening rallies and events. Specifically, they hope that this model will help law enforcers determine if social media posts about events are reliable or just smoke screens designed to cover up an actual event.
This research is being sponsored by the Virginia Governor’s Office and has the potential to be life saving in preventing future violence.
“I think data science is particularly useful here,” said Rule, “because of the way we use social media today – social media use is so prevalent that almost any real-world movements will be reflected or echoed in online activity.
“With dangerous groups using social media to plan events, it would be crazy not to use that information to improve our emergency response systems.”