Using Social Media to Enhance the Search for Gasoline During a Hurricane Evacuation Event

Abstract

Panic-buying and shortages of essential commodities is common during early phases of a disaster or an epidemic. This paper has two goals. The first goal is to develop and analyze an optimization model for an efficient search of an essential commodity; this model needs as input data processed from social media posts. The second goal is to establish that the use of social media posts can significantly improve the search efficiency, based on the analysis from a recent hurricane event. Specific contributions in the data processing of social media posts include the development of a classifier that detects shortages and an event localizer that probabilistically infers the location and time of shortage. Specific contributions in the mathematical model development include an integer programming formulation of the resultant search problem on a graph, with the two objective different objective functions: (a) Maximizing probability of finding the commodity (b) Minimizing expected time to find the commodity given the commodity is found. The first model is solved optimally using CPLEX on a linearization of the non-linear model. For the second model, an approximate solution is found using CPLEX on the linearization and a modified branch and bound method is developed. Encouragingly, the modified branch and bound method reduced computational effort by a factor of ten. The methodology is validated using a case study on gasoline search during the Hurricane Irma evacuations. We found that social media posts can predict shortage at gas station for four major cities of Florida accurately with a MAPE of 12 %. We also found that addition of social media information to the search process improved the average search time by 41.74 %.