Preference learning application in peer-to-peer logistics platforms

Official URL: https://risc01.sabanciuniv.edu/record=b2579237_(Table of contents)

Peer-to-peer logistics platforms iteratively match free agents willing to offer services to the demand requests. In some settings, the platform may not have full information about the agents’ request preferences. To reduce this uncertainty, the platform can offer a menu of requests to each agent. By collecting the agents’ preference information in each decision period, the platform has access to the historical data of agents. The gathered information helps the platform learn the agents’ utility functions over time. We simulate a ride-sharing setting, iteratively generating menus of requests for each driver and using the drivers’ preference information in each period to learn the drivers’ utility functions and predict their future choices. We show that as we learn the drivers’ utility functions, offering too many options might cause the driver to deviate from the platform’s optimal request assignment therefore, it is beneficial to decrease the menu size as we learn.