Individual car transport significantly contributes to both emissions and traffic congestion. Ride-sharing is a promising alternative that improves road utilization— without requiring costly infrastructure investments such as new metro lines. However, ride-sharing services still lack user popularity. In this thesis, we identify and address three key opportunities to enhance user satisfaction in ride-sharing. First, we show that in ride-sharing systems— particularly in those powered by deep reinforcement learning—complex user preferences beyond cost and waiting time have largely been overlooked. Through a survey, we provide a comprehensive ranking of 41 user preferences, including analyses of demographic groups (e.g., for younger people), and reveal a mismatch between preferences human users consider important and those explored in prior literature. Second, we argue that neural networks are prevalent in ride-sharing applications and identify optimization potential in both the neural network types and input feature spaces used. Our experiments on two core tasks—demand prediction and time of arrival estimation—demonstrate that suboptimal design choices in current approaches can degrade the quality of ride opportunities provided to users, and thus impact their experience. Third, we highlight the opacity introduced by neural networks in ride-sharing. To address this, we introduce two novel methods—ADESSE and EVINAM—that contribute to explain ride opportunities to users. Both approaches outperform baselines—shown via a user study and several computational experiments—and their applicability extends beyond ride-sharing to other neural network applications. Although we addressed the three research gaps identified, further work is needed to integrate our contributions into a unified, user-facing ride-sharing framework. Additionally, future research could explore variants of explanation techniques using large language models to replace human participants in pre-studies within this relatively low-stakes domain of ride-sharing. By advancing user-centered ridesharing solutions, our work contributes to making ride-sharing more appealing, thereby indirectly supporting the reduction of emissions and traffic in the transportation sector.
Sören Schleibaum (Wed,) studied this question.