UNIVERSITY PARK, Pa. -- During a 911 call, critical information is gathered that help answer the six Ws: Where, What, Weapons, When, Who and Why. The answers to these questions help to equip first responders with necessary details to approach an emergency scene.
But how can that same critical information be collected from online requests for help? A team of researchers, including several from Penn State, is working to refine a coding scheme to identify social media messages that could be useful to emergency responders. Through their work, they aim to create training datasets for machine learning models that can effectively filter useful information from the millions of social media posts created during a disaster.
“Currently, there’s no way for many public safety answering points (PSAPs) to collect information other than through phone calls,” said Jess Kropczynski, former visiting faculty member in the College of Information Sciences and Technology, co-principal investigator on the project, and current faculty member at the University of Cincinnati. “We’re trying to expand that in the digital age when we’re communicating in a lot of different ways, such as text-based communication, pictures and videos.”
In their study, the researchers tapped into a dataset of nearly 4,000 tweets collected during six disasters over the course of nine months, including the 2013 Boston Marathon bombing and Los Angeles International Airport shooting, among others, to see how the posts fit in to the 911 system’s current model.
“Although the application of the six W qualitative coding scheme demonstrated that many of the Ws are present in these tweets, examination of the content contained in these tweets showed that very few contained information to dispatch first responders,” the researchers wrote in their paper.
They then utilized a qualitative coding scheme that they previously developed in collaboration with public safety officials and applied it to the social media data. The intent was to look beyond the six Ws and incorporate subcategories that could uncover additional details, such as who needs assistance, and words that indicate a weapon is present.
Applying the coding scheme to social media data revealed that additional subcategories of the coding scheme are necessary to satisfy information requirements needed to dispatch first responders to an incident, according to the researchers’ paper.
“The coding scheme and associated datasets provide opportunities to develop tools that provide better information to emergency responders during a crisis,” said Rob Grace, doctoral student in the College of Information Sciences and Technology and member of the research team.
As their research progresses, the team aims to develop better algorithms for identifying actionable information and to build dashboards that can be utilized by 911 call centers to enhance filtering of actionable tweets.
Kropczynski and Grace have also worked with Shane Halse and Andrea Tapia, both from the Penn State College of IST; Nathan Elrod, University of Cincinnati; Doina Caragea, Kansas State University; and Cornelia Caragea, University of Illinois at Chicago, on the project. The team has had the findings of this research, as well as five other papers, accepted for publication at the 2019 International Systems for Crisis Response and Management (ISCRAM) Conference in Valencia, Spain, May 19-22.
“We look forward to presenting our work to academics and professionals similarly engaged in designing and analyzing the use of information systems for crisis response and management,” Grace concluded. “ISCRAM provides an opportunity to develop collaborative relationships with both academics and professionals active in the field.”