As urban transportation evolves, shared e-mobility is increasingly recognized as a socio-technical system shaping urban equity, social inclusion, and mobility behaviour. However, existing platforms often lack multimodal integration and user-centric adaptability, limiting their ability to address diverse and behaviourally heterogeneous travel preferences. This study proposes a cloud-based shared e-mobility platform integrating docking electric cars, e-bikes, and e-scooters with large language model assistance for natural-language preference interpretation. The system enables human-centred decision-making through lexicographic multi-objective route optimization. The platform is evaluated using 500 expert-annotated queries, assessing both preference alignment and optimization performance. Results show that gpt-4.1 achieves the highest semantic alignment score (0.928) and best route quality measured by a Mean Optimality Gap of 18.62%. To jointly capture alignment and optimization performance, we introduce the Combined Alignment–Performance Metric, under which gpt-4o achieves the highest score (1.0594). Optimization experiments demonstrate high system robustness under varying traffic conditions, as key metrics such as travel time ( p = 0 . 771 ), risk ( p = 0 . 341 ), and walking distance ( p = 0 . 153 ) show no statistically significant differences. Furthermore, the platform exhibits significant scalability, where increasing e-hub density from 20 to 100 stations reduced the mean travel time from 1771 ± 497 s to 955 ± 343 s ( p < 0 . 001 ). These findings contribute to interdisciplinary transportation research by linking optimization with human-centred mobility analysis, offering actionable insights for equity-aware urban planning, inclusive mobility system design, and policy development supporting adaptive and sustainable transport systems. • Cloud platform enables multimodal routing across e-bikes, e-scooters and e-cars. • LLMs translate user travel preferences into lexicographic optimization priorities. • Eight LLMs benchmarked for preference alignment and routing performance. • Compact models achieve comparable route quality with faster response time. • Denser e-hub networks significantly reduce average multimodal travel time.
Ding et al. (Mon,) studied this question.