Social Network Influence on Ridesharing, Disaster Communications, and Community Interactions
The complex topology of real networks allows network agents to change their functional behavior. Conceptual and methodological developments in network analysis have furthered our understanding of the effects of interpersonal environment on normative social influence and social engagement. Social influence occurs when network agents change behavior being influenced by others in the social network and this takes place in a multitude of varying disciplines. The overarching goal of this thesis is to provide a holistic understanding and develop novel techniques to explore how individuals are socially influenced, both on-line and off-line, while making shared-trips, communicating risk during extreme weather, and interacting in respective communities. The notion of influence is captured by quantifying the network effects on such decision-making and characterizing how information is exchanged between network agents. The methodologies and findings presented in this thesis will benefit different stakeholders and practitioners to determine and implement targeted policies for various user groups in regular, special, and extreme events based on their social network characteristics, properties, activities, and interactions. Chapter 2 synthesizes studies relevant to ridesharing, behavior modeling of activity-based travel and evacuation decision-making, social media research in transportation and disaster management. It also provides a comprehensive summary of the network science literature and a well-established approach to quantify social influence, known as ego-centric network design. Part I is about social influence on ridesharing which is getting more popular and people are more likely to carpool with friends as compared to traditional modes of travel. Chapter 3 presents a zero-inflated Poisson model to predict the frequency of joint trips, using ego-centric social network data, for regular activity travel decisions. Chapter 4 presents a multinomial logit model of travel mode choice and carpooling during special events such as game-day. Although ridesharing can yield in effective matching of trips, it does not necessarily provide desirable end results. The knowledge of a better understanding in terms of how people share rides in a network setting would help policy makers and city planners in modifying existing urban transportation systems by introducing more ridesharing incentives to commuters and building more efficient ridesharing platforms that can result in a more sustainable transportation system. Part II is about social influence on the way individuals communicate risk during crisis. Disaster communication networks play a salient role during emergencies since people may exchange relevant information in social media while having limited access to traditional sources of information. Previous sociological studies suggest that social networks serve the purpose of transmitting warning message by disseminating information about an impending threat, however, the empirical literature is inconclusive about how warnings received from social connections weigh into evacuation decision-making. Chapter 5 presents a mixed-logit model to capture how social networks influence individual-level evacuation decision-making based on ego-centric network data obtained from Hurricane Sandy. Chapter 6 develops a multilevel model of joint evacuation decision outcome at the dyadic (ego-alter social tie) level by using a hierarchical generalized linear modeling approach. Chapter 7 analyzes large-scale Twitter data (~52 M tweets, ~13 M users, Oct 14-Nov 12, 2012) to identify disaster communication network that was active on Twitter before, during, and after Hurricane Sandy’s landfall at different scales of user activity. Important network properties (both local and global) were obtained to examine the relationship between network structure and the information spreading capacity of network agents. It also explores the crisis communication patterns that appeared at different phases of Hurricane Sandy using advanced machine learning techniques. The findings of this chapter are useful in identifying influential nodes in such a network and disseminating targeted information more efficiently in similar crisis events. Part III is about how individuals interact in their respective communities based on common interests. Online social sharing platforms have become an integral part of daily life and people are more connected now than ever before. The objective of this part of thesis is to take advantage of the enriched evidence of social influence in social media and identify a group of like-minded users. Chapter 8 demonstrates how to construct social interaction networks from such a social sharing platform and analyzes the properties and growth of such networks. Analysis reveals that social interactions in such networks follow a power law which is indicative of fewer nodes in the network with higher level of social interactions. Chapter 9 presents a modeling framework to jointly infer user communities and interests in social interaction networks. Several pattern inference models are developed: 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 by accounting for user interests and interactions jointly. The network properties revealed and methodologies proposed in this part of the thesis will be useful to efficiently detect user communities based on their social interactions and applicable for any social sharing platform having such user interactions.
Ukkusuri, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our