Research on Crowds and Crises

Mass Convergence within Disaster and Humanitarian Response

The emCOMP Lab examines mass participation in the context of crisis (i.e. large-scale, natural, and man-made disasters). Disaster events have long been known to occasion convergence behavior—i.e. after disasters people converge physically onto the scene to— among other things—offer help. With the widespread adoption of social media and other online tools, we are now seeing digital convergence after disaster events, whereby people all over the world come together in virtual spaces, participating in a number of capacities, including volunteerism. Digital volunteerism has quickly become a recognized feature of the informational landscape after disaster events. The emCOMP Lab pursues empirical research and design opportunities in this space, seeking to find ways to support and leverage digital volunteerism and other forms of social media participation during crises, and to integrate the products of these activities into formal response efforts.

We believe that digital volunteerism offers an interesting perspective on crowdsourcing and crowdwork, one where participation is volunteer-based and motivations are largely altruistic, yet in many ways fluid (changing as an event progresses).

Examining the Flow of Misinformation after Disasters

A current focus of the emCOMP lab is the study of online rumoring during disaster events. We are approaching this problem from multiple perspectives. We have dual goals of 1) better understanding how and why misinformation spreads online after disaster events and 2) informing solutions for automatically detecting misinformation. Aligned with the lab focus on crowdwork, a significant component of this research looks specifically at how the crowd questions or challenges rumors.

In one study looking at Twitter data collected after the 2013 Boston Marathon Bombings, a collaborative effort with the Social Media (SoMe) Laboratory at UW, we draw parallels between qualitative and quantitative methods to identify multi-dimensional “signatures” or patterns of information propagation. These dimensions include a rumor’s origins, changes over time, relationships between rumor spreading and rumor challenging, and other salient features. Rumor signatures enhance our understanding of how different kinds of rumors propagate online during crisis events and, we propose, could help inform future techniques for automatically identifying and classifying rumors.

Network of Co-Occuring Hashtags: 2013 Boston Marathon Bombings

Crowdsourcery: When Emergent Collaborations Work

One feature of crises is that people naturally come together to respond. Such emergent responses typically happen quickly and (just as the disaster itself) unexpectedly. This makes it challenging for researchers to observe and consequently understand how such collaborations form and what makes them work. Now, social media data can give us a window into the micro-interactions that take place to form and sustain emergent collaborations during crises. When combined with other data such as contextual interviews, we can model how the interactions we observe in social media fit into the larger context of a particular event. 

Since 2011, we have been constructing an empirically-derived model of how information needs were met within one rural community that was impacted by Hurricane Irene. We follow the flow of specific information needs, the platforms through which each need was mediated, and the work involved to address each specific need, including reaching out  to specific audiences across gaps in the telecommunications infrastructure.

With this level of detail, we are able to 1) differentiate the many kinds of situated knowledge and expertise that came into play and observe the different kinds of micro-collaborations that formed in relation to meeting information needs.

Emergent collaboration that functioned as a "Crowd-Powdered Mesh Network"

Designing a Tool for Collaborative Crisis Curation

EmCOMP researchers are working to design and implement a tool that will help distributed teams of digital volunteers collaboratively monitor and process social media during disaster events. In the past few years, groups of remote volunteers have formed with the goal of collectively processing social media data to assist in response efforts. Currently, these groups rely on ad hoc infrastructures and work practices to coordinate their work. We have completed a first round of user studies examining their tools and practices with the goal of designing to support this work, and have implemented an initial prototype that allows collaborative tagging of tweets. We plan to do a first round of user testing in Fall 2014.

© Kate Starbird 2014