University of Washington
On October 29, 2012, Hurricane Sandy slammed into the US eastern seaboard, becoming one of the deadliest and costliest storms in US history. “Superstorm Sandy” caused 72 direct fatalities in the United States and tens of billions of dollars of damage, due in large part to a catastrophic storm surge that flooded hundreds of thousands of homes and businesses (Blake et al. 2013). The aftermath of the storm brought major disruptions to transportation systems, long-term power outages, and gas rationing in parts of New York and New Jersey.
THE SOCIAL MEDIA SURGE DURING HURRICANE SANDY
Like other disaster events in recent years, Sandy precipitated a huge surge in social media use. Twitter reported that it hosted more than 20 million tweets with search terms related to the event during a six-day window around the US impact. Instagram, a popular photo-sharing site, announced that users posted more than ten photos per second as Sandy came ashore.
Research suggested that a large portion of this content would have come from users outside affected areas and that much of it would have been “derivative"— i.e., reposted and remixed content (Starbird et al. 2010). But in this instance these platforms facilitated real-time information sharing that effectively informed response efforts. Residents of affected areas shared first-hand reports of actionable information—photos of flooded streets, videos of trees falling and houses catching fire, and tweets reporting stranded people. Emergency responders turned to social media to broadcast storm warnings and to quell rumors.
Problems with the propagation of misinformation drew widespread attention on social and mainstream media. One Twitter user reported, among other dubious
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 11
Crowds, Crisis, and Convergence: Crowdsourcing in the Context of Disasters Kate Starbird University of Washington On October 29, 2012, Hurricane Sandy slammed into the US eastern sea- board, becoming one of the deadliest and costliest storms in US history. “Super- storm Sandy” caused 72 direct fatalities in the United States and tens of billions of dollars of damage, due in large part to a catastrophic storm surge that flooded hundreds of thousands of homes and businesses (Blake et al. 2013). The aftermath of the storm brought major disruptions to transportation systems, long-term power outages, and gas rationing in parts of New York and New Jersey. The Social Media Surge during Hurricane Sandy Like other disaster events in recent years, Sandy precipitated a huge surge in social media use. Twitter reported that it hosted more than 20 million tweets with search terms related to the event during a six-day window around the US impact. Instagram, a popular photo-sharing site, announced that users posted more than ten photos per second as Sandy came ashore. Research suggested that a large portion of this content would have come from users outside affected areas and that much of it would have been “derivative”— i.e., reposted and remixed content (Starbird et al. 2010). But in this instance these platforms facilitated real-time information sharing that effectively informed response efforts. Residents of affected areas shared first-hand reports of actionable information—photos of flooded streets, videos of trees falling and houses catching fire, and tweets reporting stranded people. Emergency responders turned to social media to broadcast storm warnings and to quell rumors. Problems with the propagation of misinformation drew widespread attention on social and mainstream media. One Twitter user reported, among other dubious 11
OCR for page 11
12 FRONTIERS OF ENGINEERING claims, that the New York Stock Exchange had been flooded, and this misinfor- mation spread rapidly before being called out by other Twitter users in what one writer called a “savage self-correction” (Herrman 2012). Social media platforms and other online forums hosted self-organized com- munity response efforts and other forms of volunteerism. The latter included the establishment of a Twitter hashtag to share information about open gas stations and a related project by a group of high school students who created and maintained a live “gas map”—an online map that displayed in real time where gas was available. The role of social media during Sandy’s lead-up, impact, and response gener- ated considerable media attention. Some claimed the event marked a significant shift in the use of these services for emergency response, and at least one jour- nalist suggested (in personal communication) that Sandy was the first “social” disaster. But social media were already becoming an established feature of disaster events—after the 2010 Haiti earthquake, for example, and the 2011 Japan tsunami. And disasters have always been inherently social, since well before the emergence of social media. Sociology of Disaster Meets Web 2.0: Challenges and Opportunities Sociologists of disaster have long known that people “converge” on the scene of disaster events (Fritz and Mathewson 1957; Kendra and Wachtendorf 2003). Fritz and Mathewson (1957) explained that, though this convergence is often physical, it can also be informational as people use available channels to seek and share information. Palen and colleagues (Hughes et al. 2008; Palen et al. 2010) connected this phenomenon to what now occurs online, whereby disaster events act as catalysts for massive “digital” convergence—of the kind that can generate 20 million tweets in six days. This digital convergence carries considerable promise for improving isaster d response. First-hand observations of events from citizen reporters on the ground can increase situational awareness both for other affected people and for esponders. r Social media can also be used for formal crisis communications, and emergency responders are increasingly turning to these platforms for outgoing messaging during and between disaster events. Challenges As the examples from Hurricane Sandy suggest, there remain several signifi- cant challenges in using social media as a real-time information source. The first is volume. Clearly, it is difficult for an individual to make sense of tens of tweets and photos per second. Similarly, if the focus is on finding actionable informa- tion coming from the site of the events, a vast quantity of social media data can be considered noise—some portion is completely off-topic, and another large
OCR for page 11
CROWDS, CRISIS, AND CONVERGENCE 13 percentage contains repeated, retweeted, or otherwise “derivative” information (Starbird et al. 2010). Another particularly vexing issue is the problem of lost context, as informa- tion loses the connection to its original author, time, or place. For example, a tweet sent at 4:00 pm indicating a voluntary evacuation for a fire-affected neighborhood could become dangerous misinformation if reposted a few hours later, after the evacuation has become mandatory. Misinformation and intentional disinformation are also major concerns. And the unstructured nature of social media content represents a challenge for those trying to make sense of it in aggregate form. Automatically Filtering and Classifying the Flood of Data Purely computational solutions for filtering and otherwise processing social media streams show promise, but have some limitations. Although terms of ser- vice and protocols continually change, accessing social media data is often the easiest part of the problem, because many social media platforms provide applica- tion programming interfaces for collecting public data. Storing and searching these massive datasets presents a more complex chal- lenge, one addressed in broader conversations about dealing with “big data.” Moreover, because the textual content of social media streams is not quite the “natural language” for which traditional natural language processing techniques have been designed and tested, new approaches for computational content analysis are needed. Additionally, accuracy is extremely important in time- and safety- critical environments like those of a disaster, and currently even the best automatic classification techniques along relatively simpler data dimensions (e.g., identify- ing situational awareness information) achieve only about 80 percent accuracy (Verma et al. 2011). Harnessing the Power of the Crowd Another solution for filtering the flood of data during disasters involves human computation or crowdsourcing, using a large number of people, connected via the Internet, to manually process the data. In considering the use of these techniques, researchers are very much following the crowd. During recent disaster events, people have appropriated social media platforms and other available online tools such as Skype and shared Google Documents to improvise response efforts, often in the form of informational assistance (e.g., the New Jersey students and their gas map). The new digital volunteer behavior aligns with another long-recognized disaster phenomenon, spontaneous volunteerism, whereby people make them- selves available to help in various capacities, often by improvising to fill gaps in formal response efforts (Kendra and Wachtendorf 2003). During the 2009 Red
OCR for page 11
14 FRONTIERS OF ENGINEERING River floods in North Dakota and Minnesota, for example, volunteer program- mers created algorithms that automatically tweeted river heights at various loca- tions (Starbird et al. 2010). After the Haiti earthquake, a group of self-named “ voluntweeters” used the Twitter platform to help coordinate aid efforts, eventu- ally connecting with each other to form a new organization (Starbird and Palen 2011). In another highly publicized effort during that event, students at Tufts University created and maintained a public map of humanitarian needs, translating and geolocating thousands of reports arriving from Haitian people via an SMS short code (Meier and Munro 2010). During the impact of Hurricane Irene in the Catskills in September 2011, a group of journalists served as “crowdsourcerors,” organizing a community information-sharing and response effort through a combi- nation of a Liveblog, Facebook, Twitter, and even radio broadcasts and phone calls from landline phones in more remote areas (Dailey and Starbird, forthcoming). Although each event spawns new crowd-powered solutions to newly rec- ognized needs, a number of ongoing virtual volunteer organizations have been established (e.g., the Standby Task Force, Humanity Road, Crisis Commons, and several Virtual Operation Support Teams connected to emergency responders). These groups use available online tools to respond to disaster events all over the world. However, questions remain about how they will sustain committed par- ticipation and how they can connect both the products of their work and this new information-processing capacity more broadly to the established work practices of formal responders. One research opportunity lies in understanding the work of digital volun- teers and designing tools and platforms to support their efforts—for example, by developing crowdsourcing solutions that align with the motivations of disaster volunteers, initial altruism that soon becomes augmented by social and reputa- tion “capital.” Using the Noise to Find the Signal The collective behavior of the crowd can be leveraged to address information- processing challenges. Social media users, intentionally and not, shape the infor- mation space through their behavior within it. Instead of viewing crowd activity as simply noise, it is possible to consider every repost, “like,” “follow,” and user mention as productive crowd work and to use this “noise” to find the signal. For example, algorithms can be designed to identify misinformation through features of crowd behavior—i.e., sensing the “savage self-correction” of dozens of voices publicly questioning false information. Alternatively, the crowd itself could serve as a “sensor” for other (e.g., actionable) kinds of information. It has been demon- strated that retweet and follow patterns on Twitter can be used to home in on users tweeting from the site of an event, but there is still work to be done in designing solutions that function in real time, and questions remain about how best to com- municate these solutions to decision makers during an event.
OCR for page 11
CROWDS, CRISIS, AND CONVERGENCE 15 Integrating Machine and Human Computation The most powerful solutions in the social media space may depend on the integration of machine- and human-powered approaches. These would involve machine learning algorithms that learn from volunteers’ and other crowd members’ online actions and then feed processed data back to volunteers who verify and synthesize the output before forwarding it to responders and affected citizens. Along these lines, it will be important to design solutions that both align with the values and motivations of digital volunteers and fit into formal emergency response processes. Conclusion: The Need for Human-Centered Design in the Context of Disaster Events Massive online convergence is now an established feature of crisis events and carries with it great potential for improving outcomes during response efforts—if the right information can be transmitted to the right people at the right time and in the right form. The challenges at this intersection of crowds and crises are both technical and social. Solutions will likely benefit from a human-centered approach to understand and support the informational needs and goals of the people affected, responders, volunteers, and the broader public during disaster events. The most effective solutions will probably integrate the social media– based work of the crowd with computational algorithms that can scale up with the ever increasing size and complexity of the information-processing needs. References Blake ES, Kimberlain TB, Berg RJ, Cangialosi JP, Beven JL II. 2013. Tropical Cyclone Report, National Hurricane Center (AL182012). Silver Spring MD: US National Oceanic and Atmo- spheric Administration’s National Weather Service. Available at www.nhc.noaa.gov/data/tcr/ AL182012_Sandy.pdf. Dailey D, Starbird K. Forthcoming. Journalists as crowdsourcerers: Responding to crisis by reporting with a crowd. Fritz CE, Mathewson JH. 1957. Convergence Behavior in Disasters: A Problem in Social Control. Washington DC: National Academy of Sciences. Herrman J. 2012. Twitter is a truth machine. Blog on Buzzfeed, October 30. Available at http://gofwd. tumblr.com/post/34623466723/twitter-is-a-truth-machine. Hughes A, Palen L, Sutton J, Liu S, Vieweg S. 2008. “Site-seeing” in disaster: An examination of on-line social convergence. Proceedings of the Information Systems for Crisis Response and Management Conference (ISCRAM), Washington DC, August. Kendra JM, Wachtendorf T. 2003. Reconsidering convergence and converger: Legitimacy in response to the World Trade Center disaster. Terrorism and Disaster—New Threats, New Ideas: Research in Social Problems and Public Policy, vol 11, ed. Bingley CL. UK: Emerald Group Publishing. pp 97–122. Meier P, Munro R. 2010. The unprecedented role of SMS in disaster response: Learning from Haiti. SAIS Review 30(2):91–103.
OCR for page 11
16 FRONTIERS OF ENGINEERING Palen L, Anderson KM, Mark G, Martin J, Sicker D, Palmer M, Grunwald D. 2010. A vision for technology-mediated support for public participation and assistance in mass emergencies and disasters. Proceedings of the 2010 ACM-BCS Visions of Computer Science Conference, E dinburgh, April 14–16. Swinton UK: ACM-BCS Visions of Computer Science, British Com- puter Society. pp 1–12. Starbird K, Palen L. 2011. “Voluntweeters”: Self-organizing by digital volunteers in times of crisis. Proceedings of the 2011 ACM Conference on Human Factors in Computing Systems, Vancouver, May. New York: ACM. pp 1071–1080. Starbird K, Palen L, Hughes A, Vieweg S. 2010. Chatter on the Red: What hazards threat reveals about the social life of microblogged information. Proceedings of the 2010 ACM Conference on Computer Supported Cooperative Work, Savannah, February. New York: ACM. pp 241–250. Verma S, Vieweg S, Corvey W, Palen L, Martin J, Palmer M, Schram A, Anderson K. 2011. Natural language processing to the rescue? Extracting “situational awareness” tweets during mass emergency. Proceedings of the Fifth International AAAI Conference on Weblogs and Social Media, Barcelona, July. pp 17–21.