With the growing concerns about pollution, fuel costs and congestion, many governments and private companies are developing ride-sharing systems that offer incentives to users. Ride-sharing also is gaining popularity as smartphone apps make it easier for commuters to connect. Participants can download a ride-share app to their phones, log in, choose to be a driver or a passenger, and select from among many potential partners.
But what motivates users to participate in ride-sharing programs? Two Haslam College of Business researchers at the University of Tennessee, Knoxville, found that social opportunities and saving money were important to users of such services.
Wenjun Zhou, an associate professor at Haslam, and Wangcheng Yan, a doctoral candidate and graduate research assistant with the college, won the Best Paper Award at the 2018 INFORMS Workshop on Data Science. Their paper, “Employee Ride-sharing: User Choices and Reinforcement Learning,” delves into data from one company’s ride-share program with a focus on developing a model to understand and adapt to ride-share users’ motivations. Learning what incentivizes them will help in designing better driver-passenger matching systems, which improves user experience and retention.
Monetary considerations might seem the obvious reason to participate in the company’s ride-sharing program, as it offered a cash allowance to drivers in the program, and passengers could save on public transit fares when ride sharing.
Yan and Zhou, however, revealed that among other findings in their analysis, drivers’ strongest incentive was not the cash allocation, but good social experiences with colleagues. Passengers’ strongest incentive for using the program was saving money, but interacting with colleagues was important as well.
Currently, the passengers in the pick-up list are ranked only based on destination distances compared with the driver’s destination, without considering factors like social relations and departure time differences. Yan and Zhou’s algorithm can account for such factors, better predicting and ranking possible matches for ride-sharers.
“The algorithm assesses the utility for every driver and passenger on their own devices, so the ones that our model shows will be better will be displayed on top,” Zhou said.
In addition, the data may help companies improve their incentive programs. Yan said that within the company they studied, for instance, the financial allowance remained the same regardless of the length of the drive.
To better motivate participants, Zhou said, “They might need more dynamic pricing to decide the amount to give drivers.”
Yan was given the company data to analyze in an internship, and one of the goals of such internships in Haslam is “to work on real-world applications that apply very advanced models to real-world problems.”
Yan and Zhou said their modeling foundation, i.e., the reinforcement-learning framework, is commonly used in strategic games and is applicable to business-oriented analytics. Reinforcement learning is a form of machine learning in which an algorithm learns to maximize its objective, such as scoring as many points as possible in a game.
Yan and Zhou added a modification that allowed different transportation mode choices to have different risks, such as unsuccessful pairings, missed rides, and more, whereas traditional methods would have assumed them to have the same risk. This refinement enabled them to estimate model parameters more accurately and more realistically mimic real-world situations. Both noted that the model might be useful for broader ride-share applications, including commercial systems such as Uber and Lyft.
