Near real-time detection of crisis situations

When disaster strikes, be it natural or man-made, the immediacy of notifying emergency professionals is critical to be able to best initiate a helping response. As has social media become ubiquitous in the recent years, so have affected citizens become fast reporters of an incident. However, wanting to exploit such `citizen sensors' for identifying a crisis situation comes at a price of having to sort, in near real-time, through vast amounts of mostly unrelated, and highly unstructured information exchanged among individuals around the world. Identifying bursts in conversations can, however, decrease the burden by pinpointing an event of potential interest. Still, the vastness of information keeps the computational requirements for such procedures, even if optimized, too high for a non-distributed approach. This is where currently emerging, real-time focused distributed processing systems may excel. This paper elaborates on the possible practices, caveats, and recommendations for engineering a cloud-centric application on one such system. We used the distributed real-time computation system Apache Storm and its Trident API in conjunction with detecting crisis situations by identifying bursts in a streamed Twitter communication. We contribute a system architecture for the suggested application, and a high level description of its components' implementation.