By Rob Mitchum // December 17, 2013
[The Eric & Wendy Schmidt Data Science for Social Good Summer Fellowship is now officially accepting applications for 2014 at at dssg.uchicago.edu. The deadline for fellows and mentors is February 1st, while project partners are encouraged to apply before January 10th. During the application period, we will cross-post excerpts from articles about last year’s DSSG projects.]
While some people still consider social media to be a time-wasting fad, many scientists have started using these services as massive databases, growing every day by millions of new entries from people around the world. These projects have used the platform for predicting stock markets, box office receipts, and reality show results, but DSSG fellows Zahra Ashktorab, Christopher Brown, Manojit Nandi and mentor Aron Culotta worked with the Qatar Computational Research Institute to enlist tweets for a more noble goal: assistance during disasters and emergencies.
Drawing from the flood of tweets sent out about 2012’s Hurricane/Superstorm Sandy as it caused damage along the East Coast, the team used machine learning and statistics to classify tweets, remove redundancy, and extract information that can be used by emergency responders. The methods will be incorporated into an app that QCRI is developing to share with the Red Cross, the United Nations, and other disaster relief organizations.
Early on in the fellowship, we met with representatives from the U.N. who said they wanted a tool that would not only detect when human casualties or infrastructure damage occurred, but also what kind of disaster event caused them. That’s because different crisis scenarios demand different responses – a collapsed building might require a search and rescue squad, while a flooded hospital might need immediate medical aid and electrical engineers.
So DSSG fellows Christopher Brown, Manojit Nandi and I, along with our mentor Aron Culotta, set out to build a system that could quickly extract where disaster events are taking place, what category of event they are, and any documented human victims or physical damage.
To read the rest of Zahra Ashktorab’s post on the team’s project, visit the Data Science for Social Good blog.