Social Influence and Big Social Media Data Mining: Exploration, Modeling, and Application in Transportation.

摘要

This thesis investigates the influence of social networks on traveler behavior and the application of mining big social media data in transportation. It consists of four distinct contributions. Each examines a different dimension of traveler choices or phenomenon in transportation situations.;The first study explores information diffusion through social networks and its impact on the formation of user attitudes that influence their behavior, especially towards sustainable transportation. The objective of this research is to present a model of social network-based attitude diffusion in the context of activity and travel choice behavior. The principal mechanisms that contribute to attitude formation are first identified and then mathematical models are developed to capture these processes. The primary contributions of this research are (1) modeling attitude diffusion according to social and learning mechanisms and (2) the evolution of these attitudes over time in a lattice neighborhood social network. The agent-based framework presented is sufficiently general and flexible to allow the building of a more complete representation of information diffusion and attitude formation within activity and travel behavior choice dimensions (e.g., mode choice or departure time choice). The framework allows extending the presented approach with additional social network structures, information sources, and social interaction mechanisms in the physical and virtual realms, or extending and modifying the presented approach to simulate the impact of information-based management strategies. This research develops an application to adopt a new Park-and-Ride (P&R) alternative in four types of social networks.;The second study investigates the impact of social networks on drivers' route choice behavior. Route choice is a daily question that drivers face under varying traffic conditions. Although a variety of studies have focused on route choice behavior, actual route choice behavior on real-world networks and the mechanisms that govern it have eluded complete characterization. Part of the difficulty has been the growing availability of multiple information sources that may influence drivers' route choices, precluding the collection of adequate observational data. In addition, with the continuing emergence of new forms and sources of information, such as social media with varying degrees of interactivity, engagement, and immediacy, existing frameworks and models for representing these choices and the influence of new information sources have been lacking. This study uses an agent-based modeling approach to investigate the effects of social influence on drivers' route choice. The drivers communicate with each other in the same social network and this information is combined with their previous experiences. The first aim of this study is to investigate the impact of social network connectivity and information exchange on route choice. The second is to characterize when all drivers in the social network are satisfied with their route choices. The simulation results show that a boundedly rational equilibrium is achieved when all drivers in the given social network are satisfied with their current route choices.;The third study presents a data exploration of social network-based opinion dynamics in choice set generation in the context of activity and travel choice behavior, especially in the context of location choice. Using data from an online location-based social network, the aim is to explore the spatiality of destinations in the context of social networks and the social network influence on travelers' destination choice. Analysis results show that social relationships play a role in travelers' destination choice and that distance between friends plays a strong role in social networks as in location choice. Based on an observed possible correlation between a user's travel behavior and influence from their friends, two models were developed, the N-check-ins (number of check-ins) model and the N-locations (number of destinations) model. The estimation results show that the number of friends significantly influences a traveler's behavior. Finally, I examine the dynamic change of choice set for each user and identify the common destination choices of all users in a social network.;Based on the exploration of social networks, a meeting location recommendation model is presented to capture users' possible meeting places. The model is based on an analysis of users' check-in data from social media, which can offer valuable insights into users' travel patterns and preferences towards a particular destination. Unlike static approaches like survey data, it is possible to incorporate updated data according to users' activities in the utility model of each user. The utility maximization model for potential meeting places is implemented and tested through utilizing check-in data from 42 users collected from July 2008 to October 2010 in Chicago. By clustering all users' utilities, the potential meeting destinations of all similar users are investigated.;The last contribution consists of evaluating the feasibility of using social media data to detect traffic incidents. The content posted by users on social media sites has generated a large number of data. In this study, I evaluate the feasibility of using social media data, specifically from Twitter, for detecting traffic incidents. For the purpose of incident management, a framework is developed to extract and search real-time traffic-related Twitter data by two methods, keywords search and specific users search. The presented framework consists of three main components, Twitter data mining, location extraction, and traffic management. The approach is implemented in the Chicago area, and the online simulation models, DYNASMART-X, is used to develop and evaluate the management strategies used to reduce the impact of incidents. The main focus is on the ability of such media, e.g., tweets to improve upon incident detection methods in terms of timeliness, accuracy, and richness of information.