How Can an Algorithm Make Ride‑Sharing Passengers Happier?

In smart cities, ride‑sharing has become a real alternative to private transport. Yet sharing a car isn’t enough: passengers want to arrive on time, travel with people who have similar habits, and avoid unnecessary detours. A research team in Athens created an algorithm that considers three key factors: where users are (spatial), when they want to leave (temporal), and who they have traveled with or share interests with (social).

The system analyzes real data – GPS traces, work schedules, and travel profiles – to compute a “compatibility distance” between passengers. If two users have overlapping routes, close departure times, and a shared social context, the algorithm pairs them, producing a more direct and less crowded trip.

Tests showed that, compared to a simple location‑only grouping, the new method cuts waiting times by 15 % and boosts traveler satisfaction by 20 %. It also optimizes seat usage, lowering CO₂ emissions.

This research proves that integrating spatial, temporal and social information can make ride‑sharing more comfortable, efficient and sustainable, reshaping how city dwellers move every day.

The proposed algorithm relies on a multi‑objective formulation that combines three distance functions: d_spatial, d_temporal and d_social. The spatial distance is computed using the Haversine metric between origin and destination points, weighted by a tolerance factor that depends on Athens' urban density. The temporal component uses a time window Δt defined in minutes; two requests are compatible if the difference between their departure times is less than Δt, with a higher weight for closer times. The social distance is more complex: a graph of interactions is built based on previous shared rides, declared interests (e.g., sports, work) and demographic profiles. The social weight derives from a Jaccard similarity measure, normalized to avoid bias toward more active users.

The matching model solves a bipartite assignment problem where each vehicle is a node with limited capacity and each passenger is a node to be linked. The goal is to minimize the weighted sum of the three distances, subject to capacity constraints (number of seats) and a maximum travel time (t_max). The algorithm uses a variant of the Hungarian method modified to handle capacity constraints, guaranteeing an optimal solution in polynomial time.

For evaluation, metrics such as average waiting time (T_wait), trip completion rate (R_completion), load efficiency (η_load) and passenger satisfaction score (S_passenger) derived from post‑trip surveys were employed. Experimental results, obtained on a real dataset of 12,000 ride‑sharing requests over one month, show a 15 % reduction in T_wait, a 10 % increase in η_load and a 20 % improvement in S_passenger compared to a purely geographic distance algorithm. Moreover, CO₂ emission analysis indicates a decrease of 0.35 kg per passenger‑km, confirming the positive environmental impact of the contextual approach.

Implementation was carried out in Python, leveraging libraries such as NetworkX for graph handling and SciPy for linear optimization. The code has been released under an MIT license, allowing developers and public transport operators to adapt and integrate the solution into their urban mobility systems.