Harnessing the Crowdsourcing Power of Social Media for Disaster ...

Figure 2. Food requests after the Haiti earthquake. The Ushahisi-Haiti crisis map helps organizations intuitively ascertain wheresupplies are most needed.geo-location point. Someone mighttext the crisis map’s phone numberto report something they saw earlier,possibly texting from a shelter abouta bridge that collapsed 10 miles downthe road. Furthermore, fraud reportsfrom malicious persons might appearas normal requests on crisis map. Inaddition, there are often duplicatereports, and information essentialfor relief coordination is not readilyavailable or easily accessible, such aslists of relief resources or communicationprocedures and relief organizationcontact information.Lastly, current crowdsourcing applicationsdo not have adequate securityfeatures for relief organizationsand relief operations. For example,crowdsourcing applications that arepublicly available for reporting arealso publicly available for viewing.Although it is important to provideinformation to the public, this cancreate conflicts when decisions mustbe made about where and when reliefresources are needed. In somecircumstances, relief workers themselvesmight be targeted by nefariousgroups, so publicizing details aboutrelief efforts could endanger them.Challenges and ResearchIssuesSocial media as a crowdsourcingmechanism provides aggregate situationalawareness, important andnew communications pathways, andsome opportunities for assistance onan individual level. However, to makecrowdsourcing a useful tool, we mustaddress several challenges to leverageboth the data and communicationscapabilities, including sensemaking,security, and coordination. To tacklethese problems, we can employ textmining and social computing technologiesby managing social knowledgeand modeling social behavior duringthe disaster relief. 11Geo-tag DeterminationAs we mentioned earlier, crowdsourceddata might include inaccurategeo-tags, or might not includethem at all. Thus, using social miningto consider both a reporter’s socialand physical networks couldhelp locate a report’s actual spatialfeatures.In addition, research has shown theneed to leverage the information availablefrom the response group. For example,to supplement crowdsourcinginformation, groupsourcing 10 collectsdisaster information such as a reliefsituation summary, validated geotaginformation, and transportationconditions provided by a sanctionedgroup consisting of individuals withdisparate resources, goals, and capabilitiesworking at the disaster scene.Report VerificationTo some degree, crowdsourcing canautomatically filter reports throughphotos, videos, and comments fromother reports. Ushahidi, for example,lets users verify a report by clickinga verification button. It leaves theverification problem to the crowdsto get collective feedback. However,this strategy—an open self-adjustingmethod, in technical terms—dependson a large number of people validatingthe reports. For Ushahidi maps,however, generally only a few people12 www.computer.org/intelligent Ieee INTELLIGenT SYSTemS

Figure 3. Ushahidi’s Japanese tsunami map. By visualizing the most affected locations, the map helps relief organizationscoordinate and prioritize their responses.serve as verifiers, across the spectrumof Ushahidi map cases. A fraudulentmessage could conceivably propagateitself quickly with good camouflage,especially in a chaotic disaster environment,unless better incentives forverification are present.As crisis maps and similar crowdsourcingapplications are developed,it will be necessary to develop trustmanagement systems that can recognizeand report questionable postingsand flag them for more scrutiny andverification.Automated Report SummarizationPolice departments and other first responderswill need summarized requestreports and the ability to drilldown to individual tweets and texts.Because the volume and tempo ofmessages in a massive disaster will beso high, automated summaries andanalysis to make sense of hundreds, ifnot thousands, of unstructured messageswill be another greatly neededcapability.Spatial-Temporal Mining for SocialBehavior PredictionCurrently, volunteers and victimson the ground report incidents andrequests—that is, people who have aphone or other communication device.There are no applications forforecasting future requests or for fillingin the likely requests that wouldcome from blacked-out areas if suchpeople experience limited communicationability.Scientific data on earthquakes,floods, and other phenomena couldaugment user data to populate modelsto predict the time and locationof future requests and needs. Crowdsourceddata plus model results couldhelp first responders, governmentalagencies, and NGOs get ahead of thedemand curve and become more proactivein deploying aid and rescue capabilities.One possibility is to designspatiotemporal classifiers for modelsthat will assist in such demandforecasts.Scalability and SafetyScalability is a problem for many intelligentdistributed systems because theincreasing number of requests, agencies,and agent behaviors significantlymay/june 2011 www.computer.org/intelligent 13

enlarges the dimension of the searchspace. 3 Large-scale data-managementtechnology will be necessary to maintaina system’s stability.Safety is another problem thatmust be considered in a disaster reliefmanagement plan, especially in thedesign of a crowdsourcing system.While making data publically available,such systems must protect theNGOs’ privacy and ensure the safetyof their workers.Crowdsourcing integrated with crisismaps has been a powerful tool inhumanitarian assistance and disasterrelief. Future crowdsourcing applicationsmust provide capabilities to bettermanage unstructured messagesand enhance streaming data. Enhancementswill also include methodsof fusing information with data fromother sources, such as geological dataabout fault lines and reports of variousearthquake aftershocks. Researchersare just beginning to consider trustmanagement systems to help evaluatemessages with respect to the hugetroves of messages being traded incyberspace and via cell phones.Furthermore, metrics that gauge thesuccess of crowdsourcing and coordinationsystems for disaster relief willbe designed and leveraged for systemevaluation and improvement. Creatingan automatic recommendation systemthat helps improve resource distributioncoordination and transportationroutesuggestions is another researchdirection that can utilize real-timedata collection and relief history.There is no doubt that new coordinationplatform and crowdsourcingapplications such as crowdsourcedtranslation, filtering, and categorizationwill continue to become availableand will significantly contributeto future disaster relief efforts. Weare confident that great numbersof lives will be saved as governmentsand NGOs find better ways toemploy the integration of socialmedia, crowdsourcing, and collaborativetools into disaster relief systems.References1. J. Morgan, “Twitter and FacebookUsers Respond to Haiti Crisis,” BBCNews, 15 Jan. 2010; http://news.bbc.co.uk/2/hi/americas/8460791.stm.2. Digital Blogger, “Japan Tsunamiand Earthquake Use of Twitter,Facebook, Skype and Other SocialNetworks,” 12 Mar. 2011; http://bubblecube.wordpress.com/2011/03/12/japan-tsunami-earthquake.3. A. Bedrouni et al., Distributed IntelligentSystems: A Coordination Perspective,Springer, 2009.4. R. Goolsby, “Social Media as CrisisPlatform: The Future of CommunityMaps/Crisis Maps,” ACM Trans. IntelligentSystems Technology, vol. 1, no.1, 2010, article no. 7.5. J. Howe, “The Rise of Crowdsourcing,”Wired, vol. 14, no. 6, 2006, pp. 1–4.6. J. Surowiecki, “The Wisdom ofCrowds,” Am. J. Physics, vol. 75, no. 2,2007, p. 190.7. P. Meier and K. Brodock, “CrisisMapping Kenya’s Election Violence,”iRevolution blog, 23 Oct. 2008; http://irevolution.net/2008/10/23/mappingkenyas-election-violence.8. J. Heinzelman and C. Waters,Crowdsourcing Crisis Information inDisaster-Affected Haiti, US Inst. ofPeace, 2010.9. CrowdFlower, “Mission 4636,” 2010;www.mission4636.org.10. H. Gao et al., “Promoting Coordinationfor Disaster Relief: FromCrowdsourcing to Coordination,”Proc. Int’l Conf. Social Computing,Behavioral-Cultural Modeling, andPrediction, Springer-Verlag, 2011,pp. 197–204.11. F.-Y. Wang et al., “Social Computing:From Social Informatics to Social Intelligence,”IEEE Intelligent Systems,vol. 22, no. 2, 2007, pp. 79–83.Our experts.Your future.Huiji Gao is a PhD student in the DataMining and Machine Learning (DMML)lab at Arizona State University. Contact himat Huiji.Gao@asu.edu.Geoffrey Barbier is a doctoral candidatein computer science and a student inthe Data Mining and Machine Learning(DMML) lab at Arizona State University.Contact him at gbarbier@asu.edu.www.computer.org/bycRebecca Goolsby is an anthropologist forthe US Office of Naval Research. Contacther at rebecca.goolsby@navy.mil.14 www.computer.org/intelligent Ieee INTELLIGenT SYSTemS