Investigation of Senders' and Couriers' Preferences in a Two-sided Crowdshipping Market

Crowdshipping (CS), as an innovative logistics service, is still in its infancy and many aspects of this system have not been fully investigated and understood. To bridge this gap, this thesis comprehensively investigates the service mainly from the CS senders’ preferences on the demand side, the CS couriers’ preferences on the supply side, and the core matching problem between the senders and the couriers. Specifically, what are the characteristics of senders in using the CS service and how to accurately predict senders’ delivery requests (demannds) in an area and a short time period, what are couriers’ delivery preferences and behaviors in bidding senders’ requests, and how to match senders’ requests and couriers based on their preferences equitably and satisfactorily. The used data comes from a real CS service in U.S. between April 2015 and August 2018. A series of data descriptive analyses are conducted firstly, and find that the CS has a price advantage over FedEx in the same-day or express service, extra-large, and huge size package delivery, as well as medium size packages of short to medium delivery distance. Critically, there are discrepancies in preferences between the senders and couriers on the package size, delivery time window, delivery distance, and delivery fee. Two classes of popular deep learning (DL) methods are then applied to predict senders’ delivery requests. One class only captures the temporal feature of the data, and the other captures both spatial and temporal features. The results find that a DL method capturing both spatial and temporal features correlations inherently in the CS dataset achieves the best performance. For couriers’ bidding behavior in the CS service, this study uses popular machine learning (ML) methods to explore how features of CS delivery requests influence couriers’ bidding preferences as well as the delivery status of a requests (delivered versus undelivered). As a result, most characteristics of requests and the created discrepancy related features significantly influence the prediction targets. Importantly, this study also demonstrates the feature impacts in ML models are consistent with the results of traditional logit models. For the sender-courier matching problem, this study proposes a practical matching mechanism that equitably considers both senders’ and couriers’ preferences, and ensures all agents are satisfied with their matching results and no pair of agents prefer each other to their current match. The proposed equitable and stable two-sided matching (ESTM) algorithm is also suitable to other two-sided markets. The results show that ESTM achieves good matching rate, equity metrics (e.g., egalitarian cost, side equality cost, and pair equality cost), senders’ benefits, and social welfare in our designed CS market. As goods transportation becomes more popular with outbreaking the COVID-19 pandemic. Finally, this research also analyzes people’s online grocery shopping (OGS) choice before, during, and after COVID-19 by descriptive analysis and logit models. The used data is collected by an online survey. The results show a significant shift from physical grocery shopping (PGS) to OGS due to the pandemic. Many socio-economic characteristics, health constraints, concern of the pandemic, and grocery shopping choice behavior before COVID-19 would influence people to choose OGS during and after COVID-19.