Hurricanes often threaten to have catastrophic impacts on the lives of residents in coastal areas of the U.S. Timely evacuation limits this impact, but people may choose to evacuate or not during extreme weather conditions owing to differing personal environmental constraints that have little do with the direct risk. For example, during Hurricane Sandy a significant portion of New York and New Jersey residents facing potentially life-threatening storm-surge risk elected not to evacuate. In some cases, evacuation decisions are based solely on personal obligations and needs, yet they can often be influenced by the people with whom an individual is in contact. Previous sociological studies suggest that social networks serve the purpose of transmitting warning messages by disseminating information about an impending threat, and individuals having more social connections can be expected to receive more warning information. However, the empirical literature is inconclusive about how warnings received from social connections factor into evacuation-related decision making. This study uses novel social network data obtained from interviewing people from high-storm-surge-risk areas to understand evacuation responses during Hurricane Sandy. Individuals’ egocentric social network information was obtained using the personal network research design approach. A mixed (random parameters) logit model of individual-level evacuation decision making is developed to explain the combined effects of individual, household, and social network characteristics, along with the reliability of different information sources within a unified modeling framework. This model will enable emergency managers and planners to better predict evacuation demand: the number of individuals evacuating to a safe destination during a major hurricane threat. Researchers exploring different dimensions of evacuation logistics (e.g., departure time, destination, modal split, route choice) and simulations may also find this study informative.
30 May 2018 in Transportation Research Record: Journal of the Transportation Research Board
Hurricane Sandy was one of the deadliest and costliest of hurricanes of the past few decades. Many states experienced significant power outage; however, many people used social media to communicate while having limited or no access to traditional information sources. Using machine learning techniques, this study explored the evolution of various communication patterns and determined user concerns that emerged over the course of Hurricane Sandy. The original data included ∼52M tweets coming from ∼13M users between October 14, 2012 and November 12, 2012. A topic model was run on ∼763K tweets from the top 4,029 most frequent users who tweeted about Sandy at least 100 times. Some 250 well-defined communication patterns based on perplexity were identified. Conversations of the most frequent and relevant users indicate the evolution of numerous storm-phase (warning, response, and recovery) specific topics. People were also concerned about storm location and time, media coverage, and activities of political leaders and celebrities. Also presented is each relevant keyword that contributed to one particular pattern of user concerns. Such keywords would be particularly meaningful in targeted information-spreading and effective crisis communication in similar major disasters. Each of these words can also be helpful for efficient hash-tagging to reach the target audience as needed via social media. The pattern recognition approach of this study can be used in identifying real-time user needs in future crises.
The complex topology of real networks allows its actors to change their functional behavior. Network models provide better understanding of the evolutionary mechanisms being accountable for the growth of such networks by capturing the dynamics in the ways network agents interact and change their behavior. Considerable amount of research efforts is required for developing novel network modeling techniques to understand the structural properties such networks, reproducing similar properties based on empirical evidence, and designing such networks efficiently. First, we demonstrate how to construct social interaction networks using social media data and then present the key findings obtained from the network analytics. We analyze the characteristics and growth of such interaction networks, examine the network properties and derive important insights based on the theories of network science literature. We also discuss the application of such networks as a useful tool to effectively disseminate targeted information during planned special events. We observed that the degree-distributions of such networks follow power-law that is indicative of the existence of fewer nodes in the network with higher levels of interactions, and many other nodes with less interactions. While the network elements and average user degree grow linearly each day, densities of such networks tend to become zero. Largest connected components exhibit higher connectivity (density) when compared with the whole graph. Network radius and diameter become stable over time evidencing the small-world property. We also observe increased transitivity and higher stability of the power-law exponents as the networks grow. Data is specific to the Purdue University community and two large events, namely Purdue Day of Giving and Senator Bernie Sanders' visit to Purdue University as part of Indiana Primary Election 2016.
Hurricane Sandy was one of the deadliest and costliest of hurricanes over the past few decades. Many states experienced significant power outage, however many people used social media to communicate while having limited or no access to traditional information sources. In this study, we explored the evolution of various communication patterns using machine learning techniques and determined user concerns that emerged over the course of Hurricane Sandy. The original data included ~52M tweets coming from ~13M users between October 14, 2012 and November 12, 2012. We run topic model on ~763K tweets from top 4,029 most frequent users who tweeted about Sandy at least 100 times. We identified 250 well-defined communication patterns based on perplexity. Conversations of most frequent and relevant users indicate the evolution of numerous storm-phase (warning, response, and recovery) specific topics. People were also concerned about storm location and time, media coverage, and activities of political leaders and celebrities. We also present each relevant keyword that contributed to one particular pattern of user concerns. Such keywords would be particularly meaningful in targeted information spreading and effective crisis communication in similar major disasters. Each of these words can also be helpful for efficient hash-tagging to reach target audience as needed via social media. The pattern recognition approach of this study can be used in identifying real time user needs in future crises.
Online social media have become an integral part of our social beings. Analyzing conversations in social media platforms can lead to complex probabilistic models to understand social interaction networks. In this paper, we present a modeling approach for characterizing social interaction networks by jointly inferring user communities and interests based on social media interactions. We present several pattern inference models: i) Interest pattern model (IPM) captures population level interaction topics, ii) User interest pattern model (UIPM) captures user specific interaction topics, and iii) Community interest pattern model (CIPM) captures both community structures and user interests. We test our methods on Twitter data collected from Purdue University community. From our model results, we observe the interaction topics and communities related to two big events within Purdue University community, namely Purdue Day of Giving and Senator Bernie Sanders' visit to Purdue University as part of Indiana Primary Election 2016. Constructing social interaction networks based on user interactions accounts for the similarity of users' interactions on various topics of interest and indicates their community belonging further beyond connectivity. We observed that the degree-distributions of such networks follow power-law that is indicative of the existence of fewer nodes in the network with higher levels of interactions, and many other nodes with less interactions. We also discuss the application of such networks as a useful tool to effectively disseminate specific information to the target audience towards planning any large-scale events and demonstrate how to single out specific nodes in a given community by running network algorithms.