Categorising tweets for the Pakistan earthquake

Crowdcrafting BalochistanMany disaster responders are very open to the idea of using social media for situational awareness, but don’t know how to manage the process. “How will we have the time to do this” is one of the questions I hear most frequently. Patrick Meier and the people from QCRI have been working on a system that crowdsources the task of categorizing tweets, so that responders only get the useful messages. Following this week’s earthquake in Balochistan the team decided to go public with “MicroMappers” ahead of schedule.

MicroMappers is built on the open source Crowdcrafting platform  that has already been used after the floods in the Philippines last year and aside from some new graphical (and in my opinion superfluous) elements, it works pretty much the same.

Reducing the information-burden for disaster responders

So why is this interesting? While the Crowdcrafting system was bootstrapped spontaneously for the Philippines emergency in 2012, MicroMappers now serves as a link, a service provider which humanitarian organizations can contact to crowdsource the task of mining twitter for actionable information.

This is interesting because many humanitarian organizations are overwhelmed by the thought that they have to set up a system, find and manage volunteers and even provide support to the volunteers, when all they want is the information. With the help of MicroMappers, which is a free service, the organizations can outsource all these tasks and simply wait for the output, i.e. data that is actionable for disaster responders. I don’t see MicroMappers as a technological innovation but as a simplification of workflows and processes.

Is this useful?

The screenshot shows how the crowdcrafting platform works. It’s very simple: users get shown a tweet or a photo and categorize the information, answering the basic question “is this useful for disaster responders?”. The few tweets that are tagged as useful can then be shared with disaster response organizations.

While the Balochistan earthquake is only a first and unscheduled test of the system, it also shows the limitations:

  • There are only very few original tweets from the area, because not many people in Balochistan have access to Twitter. This limits the amount of actionable information from the ground.
  •  The filters that decide which tweets are loaded into Crowdcrafting need to be refined. While I categorized tweets, I got a lot of messages from India where the quake was also felt. While that might be interesting, it is not relevant for disaster responders in Pakistan.
  • At the moment re-tweets are included into the dataset. As a result, I saw some tweets ten or more times, which is a waste of resources.
  • The system is not set up to handle messages in languages other than English.

Aside from the first and last point, these are all problems that are comparatively easy to fix and I think MicroMappers looks very promising.

You should give it a try:

Update (28.09.2013): Patrick Meier just posted “Results of MicroMappers Response to Pakistan Earthquake” which is a very frank and honest assessment of the lack of value that Twitter and crowd sourcing added in this instance. Worth reading. I’d like to add that I don’t think this means that it was a failure; after all the purpose of outsourcing the analysis of the tweets is to move this burden from the responders/decision makers onto many shoulders so that the decision makers only see the relevant information. In this case the amount of relevant information was zero, but coming to that conclusion didn’t tie up valuable resources.