Monthly Archives: July 2012

Crisis Tweets: Natural Language Processing to the Rescue?

My colleagues at the University of Colorado, Boulder, have been doing some very interesting applied research on automatically extracting “situational awareness” from tweets generated during crises. As is increasingly recognized by many in the humanitarian space, Twitter can at times be an important source of relevant information. The challenge is to make sense of a potentially massive number of crisis tweets in near real-time to turn this information into situational awareness.

Using Natural Language Processing (NLP) and Machine Learning (ML), Colorado colleagues have developed a “suite of classifiers to differentiate tweets across several dimensions: subjectivity, personal or impersonal style, and linguistic register (formal or informal style).” They suggest that tweets contributing to situational awareness are likely to be “written in a style that is objective, impersonal, and formal; therefore, the identification of subjectivity, personal style and formal register could provide useful features for extracting tweets that contain tactical information.” To explore this hypothesis, they studied the follow four crisis events: the North American Red River floods of 2009 and 2010, the 2009 Oklahoma grassfires, and the 2010 Haiti earthquake.

The findings of this study were presented at the Association for the Advancement of Artificial Intelligence. The team from Colorado demonstrated that their system, which automatically classifies Tweets that contribute to situational awareness, works particularly well when analyzing “low-level linguistic features,” i.e., word-frequencies and key-word search. Their analysis also showed that “linguistically-motivated features including subjectivity, personal/impersonal style, and register substantially improve system performance.” In sum, “these results suggest that identifying key features of user behavior can aid in predicting whether an individual tweet will contain tactical information. In demonstrating a link between situational awareness and other markable characteristics of Twitter communication, we not only enrich our classification model, we also enhance our perspective of the space of information disseminated during mass emergency.”

The paper, entitled: “Natural Language Processing to the Rescue? Extracting ‘Situational Awareness’ Tweets During Mass Emergency,” details the findings above and is available here. The study was authored by Sudha Verma, Sarah Vieweg, William J. Corvey, Leysia Palen, James H. Martin, Martha Palmer, Aaron Schram and Kenneth M. Anderson.

Situational Awareness in Mass Emergency: Behavioral & Linguistic Analysis of Disaster Tweets

Sarah Vieweg‘s doctoral dissertation from the University of Colorado is a must-read for anyone interested in the use of twitter during crises. I read the entire 300-page study because it provides important insights on how automated natural language processing (NLP) can be applied to the Twittersphere to provide situational awareness following a sudden-onset emergency. Big thanks to Sarah for sharing her dissertation with QCRI. I include some excerpts below to highlight the most important findings from her excellent research.


“In their research on human behavior in disaster, Fritz and Marks (1954) state: ‘[T]he immediate problem in a disaster situation is neither un-controlled behavior nor intense emotional reaction, but deficiencies of coordination and organization, complicated by people acting upon individual…definitions of the situation.'”

“Fritz and Marks’ assertion that people define disasters individually, which can lead to problematic outcomes, speaks to the need for common situational awareness among affected populations. Complete information is not attained during mass emergency, else it would not be a mass emergency. However, the more information people have and the better their situational awareness, and the better equipped they are to make tactical, strategic decisions.”

“[D]uring crises, people seek information from multiple sources in an attempt to make locally optimal decisions within given time constraints. The first objective, then, is to identify what tweets that contribute to situational awareness ‘look like’—i.e. what specific information do they contain? This leads to the next objective, which is to identify how information is communicated at a linguistic level. This process provides the foundation for tools that can automatically extract pertinent, valuable information—training machines to correctly ‘understand’ human language involves the identification of the words people use to communicate via Twitter when faced with a disaster situation.”

Research Design & Results

Just how much situational awareness can be extracted from twitter during a crisis? What constitutes situational awareness in the first place vis-a-vis emergency response? And can the answer to these questions yield a dedicated ontology that can be fed into automated natural language processing platforms to generate real-time, shared awareness? To answer these questions, Sarah analyzed four emergency events: Oklahoma Fires (2009), Red River Floods (2009 & 2010) and the Haiti Earthquake (2010).

She collected tweets generated during each of these emergencies and developed a three-step qualitative coding process to analyze what kinds of information on Twitter contribute to situational awareness during a major emergency. As a first step, each tweet was categorized as either:

O: Off-topic
“Tweets do not contain any information that mentions or relates to the emergency event.”

R: On-topic and Relevant to Situational Awareness
“Tweets contain information that provides tactical, actionable information that can aid people in making decisions, advise others on how to obtain specific information from various sources, or offer immediate post- impact help to those affected by the mass emergency.”

N: On-topic and Not Relevant to Situational Awareness
“Tweets are on-topic because they mention the emergency by including offers of prayer and support in relation to the emergency, solicitations for donations to charities, or casual reference to the emergency event. But these tweets do not meet the above criteria for situational relevance.”

The O, R, and N coding of the crisis datasets resulted in the following statistics for each of the four datasets:

For the second coding step, on-topic relevant tweets were annotated with more specific information based on the following coding rule:

S: Social Environment
“These tweets include information about how people and/or animals are affected by a hazard, questions asked in relation to the hazard, responses to the hazard and actions to take that directly relate to the hazard and the emergency situation it causes. These tweets all include description of a human element in that they explain or display human behavior.”

B: Built Environment
“Tweets that include information about the effect of the hazard on the built environment, including updates on the state of infrastructure, such as road closures or bridge outages, damage to property, lack of damage to property and the overall state or condition of structures.”

P: Physical Environment
“Tweets that contain specific information about the hazard including particular locations of the hazard agent or where the hazard agent is expected or predicted to travel or predicted states of the hazard agent going forward, notes about past hazards that compare to the current hazard, and how weather may affect hazard conditions. These tweets additionally include information about the type of hazard in general [...]. This category also subsumes any general information about the area under threat or in the midst of an emergency [...].”

The result of this coding for Haiti is depicted in the figures below.

According to the results, the social environment (‘S’) category is most common in each of the datasets. “Disasters are social events; in each disaster studied in this dissertation, the disaster occurred because a natural hazard impacted a large number of people.”

For the third coding step, Sarah created a comprehensive list of several dozen  “Information Types” for each “Environment” using inductive, data-driven analysis of twitter communications, which she combined with findings from the disaster literature and official government procedures for disaster response. In total, Sarah identified 32 specific types of information that contribute to situational awareness. The table below compares the Twitter Information Types for all three environments as related to government procedures, for example.

“Based on the discourse analysis of Twitter communications broadcast during four mass emergency events,” Sarah identified 32 specific types of information that “contribute to situational awareness. Subsequent analysis of the sociology of disaster literature, government documents and additional research on the use of Twitter in mass emergency uncovered three additional types of information.”

In sum, “[t]he comparison of the information types [she] uncovered in [her] analysis of Twitter communications to sociological research on disaster situations, and to governmental procedures, serves as a way to gauge the validity of [her] ground-up, inductive analysis.” Indeed, this enabled Sarah to identify areas of overlap as well as gaps that needed to be filled. The final Information Type framework is listed below:

And here are the results of this coding framework when applied to the Haiti data:

“Across all four datasets, the top three types of information Twitter users communicated comprise between 36.7-52.8% of the entire dataset. This is an indication that though Twitter users communicate about a variety of informa-tion, a large portion of their attention is focused on only a few types of in-formation, which differ across each emergency event. The maximum number of information types communicated during an event is twenty-nine, which was during the Haiti earthquake.”

Natural Language Processing & Findings

The coding described above was all done manually by Sarah and research colleagues. But could the ontology she has developed (Information Types) be used to automatically identify tweets that are both on-topic and relevant for situational awareness? To find out, she carried out a study using VerbNet.

“The goal of identifying verbs used in tweets that convey information relevant to situational awareness is to provide a resource that demonstrates which VerbNet classes indicate information relevant to situational awareness. The VerbNet class information can serve as a linguistic feature that provides a classifier with information to identify tweets that contain situational awareness information. VerbNet classes are useful because the classes provide a list of verbs that may not be present in any of the Twitter data I examined, but which may be used to describe similar information in unseen data. In other words, if a particular VerbNet class is relevant to situational awareness, and a classifier identifies a verb in that class that is used in a previously unseen tweet, then that tweet is more likely to be identified as containing situational awareness information.”

Sarah identified 195 verbs that mapped to her Information Types described earlier. The results of using this verb-based ontology are mixed, however. “A majority of tweets do not contain one of the verbs in the identified VerbNet classes, which indicates that additional features are necessary to classify tweets according to the social, built or physical environment.”

However, when applying the 195 verbs to identify on-topic tweets relevant to situational awareness to previously unused Haiti data, Sarah found that using her customized VerbNet ontology resulted in finding 9% more tweets than when using her “Information Types” ontology. In sum, the results show that “using VerbNet classes as a feature is encouraging, but other features are needed to identify tweets that contain situational awareness information, as not all tweets that contain situational awareness information use one of the verb members in the […] identified VerbNet classes. In addition, more research in this area will involve using the semantic and syntactic information contained in each VerbNet class to identify event participants, which can lead to more fine-grained categorization of tweets.”


“Many tweets that communicate situational awareness information do not contain one of the verbs in the identified VerbNet classes, [but] the information provided with named entities and semantic roles can serve as features that classifiers can use to identify situational awareness information in the absence of such a verb. In addition, for tweets correctly identified as containing information relevant to situational awareness, named entities and semantic roles can provide classifiers with additional information to classify these tweets into the social, built and physical environment categories, and into specific information type categories.”

“Finding the best approach toward the automatic identification of situational awareness information communicated in tweets is a task that will involve further training and testing of classifiers.”

Crowdsourcing for Human Rights Monitoring: Challenges and Opportunities for Information Collection & Verification

This new book, Human Rights and Information Communication Technologies: Trends and Consequences of Use, promises to be a valuable resource to both practitioners and academics interested in leveraging new information & communication technologies (ICTs) in the context of human rights work. I had the distinct pleasure of co-authoring a chapter for this book with my good colleague and friend Jessica Heinzelman. We focused specifically on the use of crowdsourcing and ICTs for information collection and verification. Below is the Abstract & Introduction for our chapter.


Accurate information is a foundational element of human rights work. Collecting and presenting factual evidence of violations is critical to the success of advocacy activities and the reputation of organizations reporting on abuses. To ensure credibility, human rights monitoring has historically been conducted through highly controlled organizational structures that face mounting challenges in terms of capacity, cost and access. The proliferation of Information and Communication Technologies (ICTs) provide new opportunities to overcome some of these challenges through crowdsourcing. At the same time, however, crowdsourcing raises new challenges of verification and information overload that have made human rights professionals skeptical of their utility. This chapter explores whether the efficiencies gained through an open call for monitoring and reporting abuses provides a net gain for human rights monitoring and analyzes the opportunities and challenges that new and traditional methods pose for verifying crowdsourced human rights reporting.


Accurate information is a foundational element of human rights work. Collecting and presenting factual evidence of violations is critical to the success of advocacy activities and the reputation of organizations reporting on abuses. To ensure credibility, human rights monitoring has historically been conducted through highly controlled organizational structures that face mounting challenges in terms of capacity, cost and access.

The proliferation of Information and Communication Technologies (ICTs) may provide new opportunities to overcome some of these challenges. For example, ICTs make it easier to engage large networks of unofficial volunteer monitors to crowdsource the monitoring of human rights abuses. Jeff Howe coined the term “crowdsourcing” in 2006, defining it as “the act of taking a job traditionally performed by a designated agent and outsourcing it to an undefined, generally large group of people in the form of an open call” (Howe, 2009). Applying this concept to human rights monitoring, Molly Land (2009) asserts that, “given the limited resources available to fund human rights advocacy…amateur involvement in human rights activities has the potential to have a significant impact on the field” (p. 2). That said, she warns that professionalization in human rights monitoring “has arisen not because of an inherent desire to control the process, but rather as a practical response to the demands of reporting – namely, the need to ensure the accuracy of the information contained in the report” (Land, 2009, p. 3).

Because “accuracy is the human rights monitor’s ultimate weapon” and the advocate’s “ability to influence governments and public opinion is based on the accuracy of their information,” the risk of inaccurate information may trump any advantages gained through crowdsourcing (Codesria & Amnesty International, 2000, p. 32). To this end, the question facing human rights organizations that wish to leverage the power of the crowd is “whether [crowdsourced reports] can accomplish the same [accurate] result without a centralized hierarchy” (Land, 2009). The answer to this question depends on whether reliable verification techniques exist so organizations can use crowdsourced information in a way that does not jeopardize their credibility or compromise established standards. While many human rights practitioners (and indeed humanitarians) still seem to be allergic to the term crowdsourcing, further investigation reveals that established human rights organizations already use crowdsourcing and verification techniques to validate crowdsourced information and that there is great potential in the field for new methods of information collection and verification.

This chapter analyzes the opportunities and challenges that new and traditional methods pose for verifying crowdsourced human rights reporting. The first section reviews current methods for verification in human rights monitoring. The second section outlines existing methods used to collect and validate crowdsourced human rights information. Section three explores the practical opportunities that crowdsourcing offers relative to traditional methods. The fourth section outlines critiques and solutions for crowdsourcing reliable information. The final section proposes areas for future research.

The book is available for purchase here. Warning: you won’t like the price but at least they’re taking an iTunes approach, allowing readers to purchase single chapters if they prefer. Either way, Jess and I were not paid for our contribution.

For more information on how to verify crowdsourced information, please visit the following links:

  • Information Forensics: Five Case Studies on How to Verify Crowdsourced Information from Social Media (Link)
  • How to Verify and Counter Rumors in Social Media (Link)
  • Social Media and Life Cycle of Rumors during Crises (Link)
  • Truthiness as Probability: Moving Beyond the True or False Dichotomy when Verifying Social Media (Link)
  • Crowdsourcing Versus Putin (Link)


PeopleBrowsr: Next-Generation Social Media Analysis for Humanitarian Response?

As noted in this blog post on “Data Philanthropy for Humanitarian Response,” members of the Digital Humanitarian Network (DHNetwork) are still using manual methods for media monitoring. When the United Nations Office for the Coordination of Humanitarian Affairs (OCHA) activated the Standby Volunteer Task Force (SBTF) to crisis map Libya last year, for example, SBTF volunteers manually monitored hundreds of Twitter handles, news sites for several weeks.

SBTF volunteers (Mapsters) do not have access to a smart microtasking platform that could have distributed the task in more efficient ways. Nor do they have access to even semi-automated tools for content monitoring and information retrieval. Instead, they used a Google Spreadsheet to list the sources they were manually monitoring and turned this spreadsheet into a sign-up sheet where each Mapster could sign on for 3-hour shifts every day. The SBTF is basically doing “crowd computing” using the equivalent of a typewriter.

Meanwhile, companies like Crimson Hexagon, NetBase, RecordedFuture and several others have each developed sophisticated ways to monitor social and/or mainstream media for various private sector applications such as monitoring brand perception. So my colleague Nazila kindly introduced me to her colleagues at PeopleBrowsr after reading my post on Data Philanthropy. Last week, Marc from PeopleBrowsr gave me a thorough tour of the platform. I was definitely impressed and am excited that Marc wants us to pilot the platform in support of the Digital Humanitarian Network. So what’s the big deal about PeopleBrowsr? To begin with, the platform has access to 1,000 days of social media data and over 3 terabytes of social data per month.

To put this in terms of information velocity, PeopleBrowsr receives 10,000 social media posts per second from a variety of sources including Twitter, Facebook, fora and blogs. On the latter, they monitor posts from over 40 million blogs including all of Tumblr, Posterious, Blogspot and every WordPress-hosted site. They also pull in content from YouTube and Flickr. (Click on the screenshots below to magnify them).

Lets search for the term “tsunami” on Twitter. (One could enter a complex query, e.g., and/or, not, etc., and also search using twitter handles, word or hashtag clouds, top URLs, videos, pictures, etc). PeopleBrowsr summarizes the result by Location and Community. Location simply refers to where those generating content referring to a tsunami are located. Of course, many Twitter users may tweet about an event without actually being eye-witness accounts (think of Diaspora groups, for example). While PeopleBrowsr doesn’t geo-tag the location of reports events, you can very easily and quickly identify which twitter users are tweeting the most about a given event and where they are located.

As for Community, PeopleBrowsr has  indexed millions of social media users and clustered them into different communities based on their profile/bio information. Given our interest in humanitarian response, we could create our own community of social media users from the humanitarian sector and limit our search to those users only. Communities can also be created based on hashtags. The result of the “tsunami” search is displayed below.

This result can be filtered further by gender, sentiment, number of twitter followers, urgent words (e.g., alert, help, asap), time period and location, for example. The platform can monitor and view posts in any language that is posted. In addition, PeopleBrowsr have their very own Kred score which quantifies the “credibility” of social media users. The scoring metrics for Kred scores is completely transparent and also community driven. “Kred is a transparent way to measure influence and outreach in social media. Kred generates unique scores for every domain of expertise. Regardless of follower count, a person is influential if their community is actively listening and engaging with their content.”

Using Kred, PeopleBrows can do influence analysis using Twitter across all languages. They’ve also added Facebook to Kred, but only as an opt in option.  PeopleBrowsr also has some great built-in and interactive data analytics tools. In addition, one can download a situation report in PDF and print that off if there’s a need to go offline.

What appeals to me the most is perhaps the full “drill-down” functionality of PeopleBrowsr’s data analytics tools. For example, I can drill down to the number of tweets per month that reference the word “tsunami” and drill down further per week and per day.

Moreover, I can sort through the individual tweets themselves based on specific filters and even access the underlying tweets complete with twitter handles, time-stamps, Kred scores, etc.

This latter feature would make it possible for the SBTF to copy & paste and map individual tweets on a live crisis map. In fact, the underlying data can be downloaded into a CSV file and added to a Google Spreadsheet for Mapsters to curate. Hopefully the Ushahidi team will also provide an option to upload CSVs to SwiftRiver so users can curate/filter pre-existing datasets as well as content generated live. What if you don’t have time to get on PeopleBrowsr and filter, download, etc? As part of their customer support, PeopleBrowsr will simply provide the data to you directly.

So what’s next? Marc and I are taking the following steps: Schedule online demo of PeopleBrowsr of the SBTF Core Team (they are for now the only members of the Digital Humanitarian Network with a dedicated and experienced Media Monitoring Team); SBTF pilots PeopleBrowsr for preparedness purposes; SBTF deploys  PeopleBrowsr during 2-3 official activations of the Digital Humanitarian Network; SBTF analyzes the added value of PeopleBrowsr for humanitarian response and provides expert feedback to PeopleBrowsr on how to improve the tool for humanitarian response.

From Crowdsourcing Crisis Information to Crowdseeding Conflict Zones (Updated)

Friends Peter van der Windt and Gregory Asmolov are two of the sharpest minds I know when it comes to crowdsourcing crisis information and crisis response. So it was a real treat to catch up with them in Berlin this past weekend during the “ICTs in Limited Statehood” workshop. An edited book of the same title is due out next year and promises to be an absolute must-read for all interested in the impact of Information and Communication Technologies (ICTs) on politics, crises and development.

I blogged about Gregory’s presentation following last year’s workshop, so this year I’ll relay Peter’s talk on research design and methodology vis-a-vis the collection of security incidents in conflict environments using SMS. Peter and mentor Macartan Humphreys completed their Voix des Kivus project in the DRC last year, which ran for just over 16 months. During this time, they received 4,783 text messages on security incidents using the FrontlineSMS platform. These messages were triaged and rerouted to several NGOs in the Kivus as well as the UN Mission there, MONUSCO.

How did they collect this information in the first place? Well, they considered crowdsourcing but quickly realized this was the wrong methodology for their project, which was to assess the impact of a major conflict mitigation program in the region. (Relaying text messages to various actors on the ground was not initially part of the plan). They needed high-quality, reliable, timely, regular and representative conflict event-data for their monitoring and evaluation project. Crowdsourcing is obviously not always the most appropriate methodology for the collection of information—as explained in this blog post.

Peter explained the pro’s and con’s of using crowdsourcing by sharing the framework above. “Knowledge” refers to the fact that only those who have knowledge of a given crowdsourcing project will know that participating is even an option. “Means” denotes whether or not an individual has the ability to participate. One would typically need access to a mobile phone and enough credit to send text messages to Voix des Kivus. In the case of the DRC, the size of subset “D” (no knowledge / no means) would easily dwarf the number of individuals comprising subset “A” (knowledge / means). In Peter’s own words:

“Crowdseeding brings the population (the crowd) from only A (what you get with crowdsourcing) to A+B+C+D: because you give phones&credit and you go to and inform the phoneholds about the project. So the crowd increases from A to A+B+C+D. And then from A+B+C+D one takes a representative sample. So two important benefits. And then a third: the relationship with the phone holder: stronger incentive to tell the truth, and no bad people hacking into the system.”

In sum, Peter and Macartan devised the concept of “crowdseeding” to increase the crowd and render that subset a representative sample of the overall population. In addition, the crowdseeding methodology they developed genera-ted more reliable information than crowdsourcing would have and did so in a way that was safer and more sustainable.

Peter traveled to 18 villages across the Kivus and in each identified three representatives to serve as the eyes and years of the village. These representatives were selected in collaboration with the elders and always included a female representative. They were each given a mobile phone and received extensive training. A code book was also shared which codified different types of security incidents. That way, the reps simply had to type the number corresponding to a given incident (or several numbers if more than one incident had taken place). Anyone in the village could approach these reps with relevant information which would then be texted to Peter and Macartan.

The table above is the first page of the codebook. Note that the numerous security risks of doing this SMS reporting were discussed at length with each community before embarking on the selection of 3 village reps. Each community decided to voted to participate despite the risks. Interestingly, not a single village voted against launching the project. However, Peter and Macartan chose not to scale the project beyond 18 villages for fear that it would get the attention of the militias operating in the region.

A local field representative would travel to the villages every two weeks or so to individually review the text messages sent out by each representative and to verify whether these incidents had actually taken place by asking others in the village for confirmation. The fact that there were 3 representatives per village also made the triangulation of some text messages possible. Because the 18 villages were randomly selected as part the randomized control trial (RCT) for the monitoring and evaluation project, the text messages were relaying a representative sample of information.

But what was the incentive? Why did a total of 54 village representatives from 18 villages send thousands of text messages to Voix des Kivus over a year and a half? On the financial side, Peter and Macartan devised an automated way to reimburse the cost of each text message sent on a monthly basis and in addition provided an additional $1.5/month. The only ask they made of the reps was that each had to send at least one text message per week, even if that message had the code 00 which referred to “no security incident”.

The figure above depicts the number of text messages received throughout the project, which formally ended in January 2011. In Peter’s own words:

“We gave $20 at the end to say thanks but also to learn a particular thing. During the project we heard often: ‘How important is that weekly $1.5?’ ‘Would people still send messages if you only reimburse them for their sent messages (and stop giving them the weekly $1.5)?’ So at the end of the project [...] we gave the phone holder $20 and told them: the project continues exactly the same, the only difference is we can no longer send you the $1.5. We will still reimburse you for the sent messages, we will still share the bulletins, etc. While some phone holders kept on sending textmessages, most stopped. In other words, the financial incentive of $1.5 (in the form of phonecredit) was important.”

Peter and Macartan have learned a lot during this project, and I urge colleagues interested in applying their project to get in touch with them–I’m happy to provide an email introduction. I wish Swisspeace’s Early Warning System (FAST) had adopted this methodology before running out of funding several years ago. But the leadership at the time was perhaps not forward thinking enough. I’m not sure whether the Conflict Early Warning and Response Network (CEWARN) in the Horn has fared any better vis-a-vis demonstrated impact or lack thereof.

To learn more about crowdsourcing as a methodology for information collection, I recommend the following three articles:

Surprising Findings: Using Mobile Phones to Predict Population Displacement After Major Disasters

Rising concerns over the consequences of mass refugee flows during several crises in the late 1970’s is what prompted the United Nations (UN) to call for the establishment of early warning systems for the first time. “In 1978-79 for example, the United Nations and UNHCR were clearly overwhelmed by and unprepared for the mass influx of Indochinese refugees in South East Asia. The number of boat people washed onto the beaches there seriously challenged UNHCR’s capability to cope. One of the issues was the lack of advance information. The result was much human suffering, including many deaths. It took too long for emergency assistance by intergovernmental and non-governmental organizations to reach the sites” (Druke 2012 PDF).

Forty years later, my colleagues at Flowminder are using location data from mobile phones to nowcast and predict population displacement after major disasters. Focusing on the devastating 2010 Haiti earthquake, the team analyzed the movement of 1.9 million mobile users before and after the earthquake. Naturally, the Flowminder team expected that the mass exodus from Port-au-Prince would be rather challenging to predict. Surprisingly, however, the predictability of people’s movements remained high and even increased during the three-month period following the earthquake.

The team just released their findings in a peer-reviewed study entitled: “Predictability of population displacement after the 2010 Haiti earthquake” (PNAS 2012). As the analysis reveals, “the destinations of people who left the capital during the first three weeks after the earthquake was highly correlated with their mobility patterns during normal times, and specifically with the locations in which people had significant social bonds, as measured by where they spent Christmas and New Year holidays” (PNAS 2012).

For the people who left Port-au-Prince, the duration of their stay outside the city, as well as the time for their return, all followed a skewed, fat-tailed distribution. The findings suggest that population movements during disasters may be significantly more predictable than previously thought” (PNAS 2012). Intriguingly, the analysis also revealed the period of time that people in Port-au-Prince waited to leave the city (and then return) was “power-law distributed, both during normal days and after the earthquake, albeit with different exponents (PNAS 2012).” Clearly then, “[p]eople’s movements are highly influenced by their historic behavior and their social bonds, and this fact remained even after one of the most severe disasters in history” (PNAS 2012).


I wonder how this approach could be used in combination with crowdsourced satellite imagery analysis on the one hand and with Agent Based Models on the other. In terms of crowdsourcing, I have in mind the work carried out by the Standby Volunteer Task Force (SBTF) in partnership with UNHCR and Tomnod in Somalia last year. SBTF volunteers (“Mapsters”) tagged over a quarter million features that looked liked IDP shelters in under 120 hours, yielding a triangulated country of approximately 47,500 shelters.

In terms of Agent Based Models (ABMs), some colleagues and I  worked on “simulating population displacements following a crisis”  back in 2006 while at the Santa Fe Institute (SFI). We decided to use an Agent Based Model because the data on population movement was simply not within our reach. Moreover, we were particularly interested in modeling movements of ethnic populations after a political crisis and thus within the context of a politically charged environment.

So we included a preference for “safety in numbers” within the model. This parameter can easily be tweaked to reflect a preference for moving to locations that allow for the maintenance of social bonds as identified in the Flowminder study. The figure above lists all the parameters we used in our simple decision theoretic model.

The output below depicts the Agent Based Model in action. The multi-colored panels on the left depict the geographical location of ethnic groups at a certain period of time after the crisis escalates. The red panels on the right depict the underlying social networks and bonds that correspond to the geographic distribution just described. The main variable we played with was the size or magnitude of the sudden onset crisis to determine whether and how people might move differently around various ethnic enclaves. The study long with the results are available in this PDF.

In sum, it would be interesting to carry out Flowminder’s analysis in combination with crowdsourced satellite imagery analysis and live sensor data feeding into an Agent Base Model. Dissertation, anyone?

DeadUshahidi: Neither Dead Right Nor Dead Wrong

There’s a new Crowdmap in town called DeadUshahidi. The site argues that “Mapping doesn’t equal change. Using crowdsourcing tech like Ushahidi maps without laying the strategic and programmatic ground work is likely not going to work. And while we think great work has been done with crowdsourced reporting, there is an increasing number of maps that are set up with little thought as to why, who should care, and how the map leads to any changes.”

In some ways this project is stating the obvious, but the obvious sometimes needs repeating. As Ushahidi’s former Executive Director Ory Okolloh warned over two years ago: “Don’t get too jazzed up! Ushahidi is only 10% of solution.” My own doctoral research, which included a comparative analysis of Ushahidi’s use in Egypt and the Sudan, demonstrated that training, preparedness, outreach and strategic partnerships were instrumental. So I do appreciate DeadUshahidi’s constructive (and entertaining!) efforts to call attention to this issue and explain what makes a good crowd-sourced map.

At the same time, I think some of the assumptions behind this initiative need questioning. According to the project, maps with at least one of the following characteristics is added to the cemetery:

  • No one has submitted a report to your map in the last 12 months.
  • For time-bound events, like elections and disasters, the number of reports are so infinitesimally small (in relation to the number of the community the map is targeting) that the map never reached a point anywhere near relevance. (Our measure for elections is, for instance, # of submissions / # of registered voters > .0001).
  • The map was never actually started (no category descriptions, fewer than 10 reports). We call that a stillbirth.

Mapping doesn’t equal change, but why assume that every single digital map is launched to create change? Is every blog post written to create change? Is every Wikipedia article edit made to effect change? Every tweet? What was the impact of the last hard copy map you saw? Intention matters and impact cannot be measured without knowing the initial motivations behind a digital map, the intended theory of change and some kind of baseline to measure said change. Also, many digital maps are event-based and thus used for a limited period of time only. They may no longer receive new reports a year after the launch, but this doesn’t make it a “dead” map, simply a completed project. A few may even deserve to go to map heaven—how about a UshahidiHeaven crowdmap?

I’m also not entirely convinced by the argument that the number of reports per map has to cross a certain threshold for the crowdsourced map to be successful. A digital map of a neighborhood in Sydney with fewer than one hundred reports could very well have achieved the intended goal of the project. So again, without knowing or being able to reliably discern the motivations behind a digital map, it is rather farfetched to believe that one can assess whether a project was success-ful or not. Maybe most of the maps in the DeadUshahidi cemetery were never meant to live beyond a few days, weeks or months in the first place.

That said, I do think that one of the main challenges with Ushahidi/Crowdmap use is that the average number of reports per map is very, very low. Indeed, the vast majority of Crowdmaps are stillborn as a forthcoming study from Internews shows. Perhaps this long-tail effect shouldn’t be a surprise though. The costs of experimenting are zero and the easier the technology gets, the more flowers will bloom—or rather the more seeds become available. Whether these free and open source seeds actually get planted and grow into flowers (let alone lush eco-systems) is another issue and one dependent on a myriad of factors such as the experience of the “gardener”, the quality of the seeds, the timing and season, the conditions of the soil and climate, and the availability of other tools used for planting and cultivation.

Or perhaps a better analogy is photography. Thanks to Digital Cameras, we take zillions more pictures than we did just 5 years ago because each click is virtually free. We’re no longer limited to 24 or 36 pictures per roll of film, which first required one to buy said roll and later to pay for it again to be developed. As a result of digital cameras, one could argue that there are now a lot more bad quality (dead) pictures being uploaded everywhere. So what? Big deal. There is also more excellent amateur photography out there as well. What about other technologies and media? There are countless of “dead” Twitter accounts, WordPress blogs, Ning platforms, customized Google Maps, etc. Again, so what?

Neogeography is about democratizing map-making and user-generated maps. Naturally, there’s going to be learning and experimentation involved. So my blog post is not written in defense of Ushahidi/Crowdmap but rather in defense of all amateur digital mappers out there who are curious and just want to map whatever the heck they well please. In sum, and to return to the gardening analogy if I may, the more important question here is why the majority of (Usha)seeds aren’t planted or don’t grow, and what can be done about this in a pro-active manner. Is there something wrong with the seed? Do would-be gardeners simply need more gardening manuals? Or do they need more agile micro-tasking and data-mining tools? The upcoming Internews report goes a long way to explaining the why & what and TechChange’s course on Ushahidi may be one way to save some future maps from ending up in the DeadUshahidi cemetery prematurely.