Travel choice with information sharing via social networks

Abstract

Nowadays, wide usage of social media and mobile devices motivates more and more travelers to share traffic information with each other. This thesis focuses on traveler’s choices in commute, including departure time choice and route choice, to study the impact of information sharing among travelers on their choice behavior. For departure time choice, we first conduct an online experiment to examine whether travelers with shared information from others will behave differently from those without such information or with other kinds of information. The statistical analysis of the experimental data gives a positive answer to this question. Travelers with shared information from friends are significantly different from those who without such information in both departure time choice and corresponding travel cost. After the experiment, subjects were asked to rate the usefulness of shared information. Most of them consider it very helpful for departure time choice, with their average usefulness score is higher than that of subjects’ own previous days’ travel experience. A departure time choice model is constructed based on the ordinal probit model to investigate the relationship between shared information and departure time adjustments, whose coefficients are estimated from the experimental data. The results show that travelers will follow the best choice and avoid the worst choice among friends, and they are reluctant to choose a later departure time unless they have information indicating a later time might be a better choice. The coefficients of variables representing shared information are also estimated for different time periods with 10 days as an interval, and we find the impact of shared information changes over time. Friends’ information has a significant impact on traveler’s departure time choices when they are learning an unfamiliar traffic environment with fixed road capacity. It is even more useful for travelers when their familiar traffic environment changes to have higher variability. We further propose a Bayesian learning model to model the choice behavior of travelers with shared information from day to day. They are assumed to use shared information from friends to update their belief of the utility of each departure time interval dynamically. The properties of the model and the impact of social networks are analyzed. It is a general framework to study the travel choice with social interactions over time, and flexible for many extensions. The model is assessed by the experimental data. Based on the learning model, we propose an agent-based simulation method to study the impact of information sharing on more general traffic environment and social networks. Three typical social networks are selected: lattice, small-world, and scale-free. It is found that the node degree in the social network (i.e., the number of friends) is an important factor in affecting departure time choice. And in a social network having an imbalanced distribution of node degree, such as the scale-free network, travelers with more friends (higher degree) get lower travel cost on average. The impact of information sharing on day-to-day route choice is studied then. We incorporate the shared information to the day-to-day Markov traffic assignment model, in which travelers used shared travel time (cost) information from friends to make choice. It can be shown that this discrete time Markov chain will converge to a unique stationary probability distribution on the route choice state space, if the social connections among travelers do not change over time. The stationary probability distribution is determined by the social networks of travelers. On the other hand, the convergence rate of the process with information sharing among travelers can be slower than that of the process assuming all travelers have the same and perfect information of previous states. To minimize the expected total travel time of the transport system with travelers’ social interactions, we formulate it as a Markov Decision Process (MDP) and solve it to find the optimal pricing policy. This pricing policy is compared with existing pricing strategies without consideration of information sharing. Moreover, a group-based model is proposed for the case that travelers form communities in the social network, which can largely reduce the number of states in the model. Numerical examples are used to illustrate the findings.

Date of Award	2017
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology