Sensing Social Media Context of Natural Disasters for Effective Response with Time-Frequency Analysis

Abstract

In the modern era, social media serves as a plethora of information with its own mixture of good bad and ugly. Any physical event that has an impact on living beings, leaves an unbiased digital footprint on social media. The constantly evolving nature of communication between people has shaped it and granted it its own characteristics and features like any other media. The singular most important characteristics of social media are the spontaneous connectivity which maximizes collectivism by provoking individualism. The collectivism provides validation to information and the connectivity makes it accessible to all in real time. This phenomenon has motivated us to investigate the platform for a domain that needs support now more than ever. With the increase of natural disasters all over the world, we are in crucial need of innovative solutions with inexpensive implementations to assist the emergency response systems. Information collected through conventional sources (e.g., incident reports, physical volunteers, etc.) are insufficient. Responsible organizations are now leaning towards research grounds that explore digital human connectivity and freely available sources of information. In the wake of an emergency, even the responsible organizations turn to social media, the most inexpensive and readily available source of enormous information, in order to sense it to assist in disaster faster and effective response. In the past couple of years, in the United States, during some of the deadly hurricanes (e.g., Harvey, IRMA, Michael, Florence, etc.), people took on different social media platforms like never seen before, in search of help for rescue, shelter, and relief. Their posts reflect crisis updates and their real-time observations on the devastation that they witness. The information on social media is time sensitive and validated by many which make it an unwavering sensor that consistently broadcasts information like a non-stationary signal. This dissertations proposes a novel end-to-end pipeline to gather extract, track and connect meaningful information from the social media using time-frequency features obtained from Continuous Wavelet Transform. Continuous Wavelet Transform is comprehensive, it bridges the gap in information and can be analyzed with the finest granularity. We propose and implement methodologies to build a novel end-to-end pipeline to characterize and analyze social media events with time-frequency features and implement unsupervised, pragmatic applications to assist in effective and efficient decision support for disaster response.