Analyzing Crowd-Sourced Information and Social Media for Crisis Management

Abstract

The analysis of potentially large volumes of crowd-sourced and social media data is central to meeting the requirements of the ATHENA project. Here, we discuss the various stages of the pipeline process we have developed, including acquisition of the data, analysis, aggregation, filtering, and structuring. We highlight the challenges involved when working with unstructured, noisy data from sources such as Twitter, and describe the crisis taxonomies that have been developed to support the tasks and enable concept extraction. State-of-the-art techniques such as formal concept analysis and machine learning are used to create a range of capabilities including concept drill down, sentiment analysis, credibility assessment, and assignment of priority. We ground many of these techniques using results obtained from a set of tweets which emerged from the Colorado wildfires of 2012 in order to demonstrate the applicability of our work to real crisis scenarios.