Tag Archives: Crises

Update: Twitter Dashboard for Disaster Response

Project name: Artificial Intelligence for Disaster Response (AIDR). For a more recent update, please click here.

My Crisis Computing Team and I at QCRI have been working hard on the Twitter Dashboard for Disaster Response. We first announced the project on iRevolution last year. The experimental research we’ve carried out since has been particularly insightful vis-a-vis the opportunities and challenges of building such a Dashboard. We’re now using the findings from our empirical research to inform the next phase of the project—namely building the prototype for our humanitarian colleagues to experiment with so we can iterate and improve the platform as we move forward.


Manually processing disaster tweets is becoming increasingly difficult and unrealistic. Over 20 million tweets were posted during Hurricane Sandy, for example. This is the main problem that our Twitter Dashboard aims to solve. There are two ways to manage this challenge of Big (Crisis) Data: Advanced Computing and Human Computation. The former entails the use of machine learning algorithms to automatically tag tweets while the latter involves the use of microtasking, which I often refer to as Smart Crowdsourcing. Our Twitter Dashboard seeks to combine the best of both methodologies.

On the Advanced Computing side, we’ve developed a number of classifiers that automatically identify tweets that:

  • Contain informative content (in contrast to personal messages or information unhelpful for disaster response);
  • Are posted by eye-witnesses (as opposed to 2nd-hand reporting);
  • Include pictures, video footage, mentions from TV/radio
  • Report casualties and infrastructure damage;
  • Relate to people missing, seen and/or found;
  • Communicate caution and advice;
  • Call for help and important needs;
  • Offer help and support.

These classifiers are developed using state-of-the-art machine learning tech-niques. This simply means that we take a Twitter dataset of a disaster, say Hurricane Sandy, and develop clear definitions for “Informative Content,” “Eye-witness accounts,” etc. We use this classification system to tag a random sample of tweets from the dataset (usually 100+ tweets). We then “teach” algorithms to find these different topics in the rest of the dataset. We tweak said algorithms to make them as accurate as possible; much like training a dog new tricks like go-fetch (wink).


We’ve found from this research that the classifiers are quite accurate but sensitive to the type of disaster being analyzed and also the country in which said disaster occurs. For example, a set of classifiers developed from tweets posted during Hurricane Sandy tend to be less accurate when applied to tweets posted for New Zealand’s earthquake. Each classifier is developed based on tweets posted during a specific disaster. In other words, while the classifiers can be highly accurate (i.e., tweets are correctly tagged as being damage-related, for example), they only tend to be accurate for the type of disaster they’ve been trained for, e.g., weather-related disasters (tornadoes), earth-related (earth-quakes) and water-related (floods).

So we’ve been busy trying to collect as many Twitter datasets of different disasters as possible, which has been particularly challenging and seriously time-consuming given Twitter’s highly restrictive Terms of Service, which prevents the direct sharing of Twitter datasets—even for humanitarian purposes. This means we’ve had to spend a considerable amount of time re-creating Twitter datasets for past disasters; datasets that other research groups and academics have already crawled and collected. Thank you, Twitter. Clearly, we can’t collect every single tweet for every disaster that has occurred over the past five years or we’ll never get to actually developing the Dashboard.

That said, some of the most interesting Twitter disaster datasets are of recent (and indeed future) disasters. Truth be told, tweets were still largely US-centric before 2010. But the international coverage has since increased, along with the number of new Twitter users, which almost doubled in 2012 alone (more neat stats here). This in part explains why more and more Twitter users actively tweet during disasters. There is also a demonstration effect. That is, the international media coverage of social media use during Hurricane Sandy, for example, is likely to prompt citizens in other countries to replicate this kind of pro-active social media use when disaster knocks on their doors.

So where does this leave us vis-a-vis the Twitter Dashboard for Disaster Response? Simply that a hybrid approach is necessary (see TEDx talk above). That is, the Dashboard we’re developing will have a number of pre-developed classifiers based on as many datasets as we can get our hands on (categorized by disaster type). In addition to that, the dashboard will also allow users to create their own classifiers on the fly by leveraging human computation. They’ll also be able to microtask the creation of new classifiers.

In other words, what they’ll do is this:

  • Enter a search query on the dashboard, e.g., #Sandy.
  • Click on “Create Classifier” for #Sandy.
  • Create a label for the new classifier, e.g., “Animal Rescue”.
  • Tag 50+ #Sandy tweets that convey content about animal rescue.
  • Click “Run Animal Rescue Classifier” on new incoming tweets.

The new classifier will then automatically tag incoming tweets. Of course, the classifier won’t get it completely right. But the beauty here is that the user can “teach” the classifier not to make the same mistakes, which means the classifier continues to learn and improve over time. On the geo-location side of things, it is indeed true that only ~3% of all tweets are geotagged by users. But this figure can be boosted to 30% using full-text geo-coding (as was done the TwitterBeat project). Some believe this figure can be doubled (towards 75%) by applying Google Translate to the full-text geo-coding. The remaining users can be queried via Twitter for their location and that of the events they are reporting.

So that’s where we’re at with the project. Ultimately, we envision these classifiers to be like individual apps that can be used/created, dragged and dropped on an intuitive widget-like dashboard with various data visualization options. As noted in my previous post, everything we’re building will be freely accessible and open source. And of course we hope to include classifiers for other languages beyond English, such as Arabic, Spanish and French. Again, however, this is purely experimental research for the time being; we want to be crystal clear about this in order to manage expectations. There is still much work to be done.

In the meantime, please feel free to get in touch if you have disaster datasets you can contribute to these efforts (we promise not to tell Twitter). If you’ve developed classifiers that you think could be used for disaster response and you’re willing to share them, please also get in touch. If you’d like to join this project and have the required skill sets, then get in touch, we may be able to hire you! Finally, if you’re an interested end-user or want to share some thoughts and suggestions as we embark on this next phase of the project, please do also get in touch. Thank you!


Social Network Analysis for Digital Humanitarian Response

Monitoring social media for digital humanitarian response can be a massive undertaking. The sheer volume and velocity of tweets generated during a disaster makes real-time social media monitoring particularly challenging if not near impossible. However, two new studies argue that there is “a better way to track the spread of information on Twitter that is much more powerful.”


Manuel Garcia-Herranz and his team at the Autonomous University of Madrid in Spain use small groups of “highly connected Twitter users as ‘sensors’ to detect the emergence of new ideas. They point out that this works because highly co-nnected individuals are more likely to receive new ideas before ordinary users.” The test their hypothesis, the team studied 40 million Twitters users who “together totted up 1.5 billion follows’ and sent nearly half a billion tweets, including 67 million containing hashtags.”

They found that small groups of highly connected Twitter users detect “new hashtags about seven days earlier than the control group.  In fact, the lead time varied between nothing at all and as much as 20 days.” Manuel and his team thus argue that “there’s no point in crunching these huge data sets. You’re far better off picking a decent sensor group and watching them instead.” In other words, “your friends could act as an early warning system, not just for gossip, but for civil unrest and even outbreaks of disease.”

The second study, “Identifying and Characterizing User Communities on Twitter during Crisis Events,” (PDF) is authored by Aditi Gupta et al. Aditi and her co-lleagues analyzed three major crisis events (Hurricane Irene, Riots in England and Earthquake in Virginia) to “to identify the different user communities, and characterize them by the top central users.” Their findings are in line with those shared by the team in Madrid. “[T]he top users represent the topics and opinions of all the users in the community with 81% accuracy on an average.” In sum, “to understand a community, we need to monitor and analyze only these top users rather than all the users in a community.”

How could these findings be used to prioritize the monitoring of social media during disasters? See this blog post for more on the use of social network analysis (SNA) for humanitarian response.

Why the Public Does (and Doesn’t) Use Social Media During Disasters

The University of Maryland has just published an important report on “Social Media Use During Disasters: A Review of the Knowledge Base and Gaps” (PDF). The report summarizes what is empirically known and yet to be determined about social media use pertaining to disasters. The research found that members of the public use social media for many different reasons during disasters:

  • Because of convenience
  • Based on social norms
  • Based on personal recommendations
  • For humor & levity
  • For information seeking
  • For timely information
  • For unfiltered information
  • To determine disaster magnitude
  • To check in with family & friends
  • To self-mobilize
  • To maintain a sense of community
  • To seek emotional support & healing

Conversely, the research also identified reasons why some hesitate to use social media during disasters: (1) privacy and security fears, (2) accuracy concerns, (3) access issues, and (4) knowledge deficiencies. By the latter they mean the lack of knowledge on how to use social media prior to disasters. While these hurdles present important challenges they are far from being insurmountable. Educa-tion, awareness-raising, improving technology access, etc., are all policies that can address the stated constraints. In terms of accuracy, a number of advanced computing research centers such as QCRI are developing methodologies and pro-cesses to quantify credibility on social media. Seasoned journalists have also been developing strategies to verify crowdsourced information on social media.

Perhaps the biggest challenge is privacy, security and ethics. Perhaps the new mathematical technique, “differential privacy,” may provide the necessary break-through to tackle the privacy/security challenge. Scientific American writes that differential privacy “allows for the release of data while meeting a high standard for privacy protection. A differentially private data release algorithm allows researchers to ask practically any question about a database of sensitive informa-tion and provides answers that have been ‘blurred’ so that they reveal virtually nothing about any individual’s data—not even whether the individual was in the database in the first place.”

The approach has already been used in a real-world applications: a Census Bureau project called OnTheMap, “which gives researchers access to agency data. Also, differential privacy researchers have fielded preliminary inquiries from Facebook and the federally funded iDASH center at the University of California, San Diego, whose mandate in large part is to find ways for researchers to share biomedical data without compromising privacy.” So potential solutions are al-ready on the horizon and more research is on the way. This doesn’t mean there are no challenges left. There will absolutely be more. But the point I want to drive home is that we are not completely helpless in the face of these challenges.

The Report concludes with the following questions, which are yet to be answered:

  • What, if any, unique roles do various social media play for commu-nication during disasters?
  • Are some functions that social media perform during disasters more important than others?
  • To what extent can the current body of research be generalized to the U.S. population?
  • To what extent can the research on social media use during a specific disaster type, such as hurricanes, be generalized to another disaster type, such as terrorism?

Have any thoughts on what the answers might be and why? If so, feel free to add them in the comments section below. Incidentally, some of these questions could make for strong graduate theses and doctoral dissertations. To learn more about what people actually tweet during this disasters, see these findings here.

What Percentage of Tweets Generated During a Crisis Are Relevant for Humanitarian Response?

More than half-a-million tweets were generated during the first three days of Hurricane Sandy and well over 400,000 pictures were shared via Instagram. Last year, over one million tweets were generated every five minutes on the day that Japan was struck by a devastating earthquake and tsunami. Humanitarian organi-zations are ill-equipped to manage this volume and velocity of information. In fact, the lack of analysis of this “Big Data” has spawned all kinds of suppositions about the perceived value—or lack thereof—that social media holds for emer-gency response operations. So just what percentage of tweets are relevant for humanitarian response?

One of the very few rigorous and data-driven studies that addresses this question is Dr. Sarah Vieweg‘s 2012 doctoral dissertation on “Situational Awareness in Mass Emergency: Behavioral and Linguistic Analysis of Disaster Tweets.” After manually analyzing four distinct disaster datasets, Vieweg finds that only 8% to 20% of tweets generated during a crisis provide situational awareness. This implies that the vast majority of tweets generated during a crisis have zero added value vis-à-vis humanitarian response. So critics have good reason to be skeptical about the value of social media for disaster response.

At the same time, however, even if we take Vieweg’s lower bound estimate, 8%, this means that over 40,000 tweets generated during the first 72 hours of Hurricane Sandy may very well have provided increased situational awareness. In the case of Japan, more than 100,000 tweets generated every 5 minutes may have provided additional situational awareness. This volume of relevant infor-mation is much higher and more real-time than the information available to humanitarian responders via traditional channels.

Furthermore, preliminary research by QCRI’s Crisis Computing Team show that 55.8% of 206,764 tweets generated during a major disaster last year were “Informative,” versus 22% that were “Personal” in nature. In addition, 19% of all tweets represented “Eye-Witness” accounts, 17.4% related to information about “Casualty/Damage,” 37.3% related to “Caution/Advice,” while 16.6% related to “Donations/Other Offers.” Incidentally, the tweets were automatically classified using algorithms developed by QCRI. The accuracy rate of these ranged from 75%-81% for the “Informative Classifier,” for example. A hybrid platform could then push those tweets that are inaccurately classified to a micro-tasking platform for manual classification, if need be.

This research at QCRI constitutes the first phase of our work to develop a Twitter Dashboard for the Humanitarian Cluster System, which you can read more about in this blog post. We are in the process of analyzing several other twitter datasets in order to refine our automatic classifiers. I’ll be sure to share our preliminary observations and final analysis via this blog.