Abstract

This project aims to propose a Mobility-on-Demand (MOD) service with a destination recommender system that interacts with users and recommends activity destinations to them. The proposed system implements a learning algorithm to learn the users' destination preferences efficiently, dealing with a trade-off between balancing an option selection that could efficiently learn the uncertainty (exploration) and providing users with rewarding choices (exploitation). It will be especially helpful for seniors who have limited access to contemporary ICT because they can act as physical search engines for travelers. This project has a threefold objective: 1. Better understand the mobility needs of the elderly across different cities. This is accomplished via a joint survey conducted with collaborators from the University of Texas, El Paso (UTEP), on elderly living in El Paso, TX, and in New York, NY. 2. Implement a proof-of-concept of a recommender system that can be readily adapted to MOD services, one that considers routing constraints. 3. Conduct computational experiments with the proof of concept to demonstrate the existence of the effect that adding spatial constraints has on the performance of a recommender system. Based on these computational experiments, we draw new guidelines for expanding this research for MOD service providers using publicly available data. Codes used for the analysis are provided here: https://github.com/BUILTNYU/recommender-system Also, the expanded version of the recommender system is available: https://github.com/BUILTNYU/DestinationRecoMOD