On critical event observability using social networks: A disaster monitoring perspective

Abstract

The proliferation of social networks with large scale information dissemination capabilities, such as Twitter, significantly increases the degree of observability of critical events, such as natural or man-made disasters. This paper analyzes the extent to which critical physical events indeed are observable, thanks to social networks, as well as the extent to which the offered view into event state that affects the state's own evolution. As a case study, we investigate the gas shortage that ensued around New York City in the aftermath of Hurricane Sandy in November 2012. Both ground truth data regarding the shortage as well as Twitter data describing it are collected. Results suggest that the social network responds to the shortage in a manner that enables (noisy) reconstruction of actual damage evolution. Non-linear models of social response tend to fit the data better, suggesting that the response switches from an initial rational reaction to a subsequent panic reaction that is largely a function of its own history, as opposed to that of the physical event. Deriving a good model of this (over)reaction is therefore critical for correct reconstruction of the actual damage. Similarly, the paper presents models for actual damage and demonstrates that combining social response with actual damage can improve the damage modeling capability.