Abstract: Dockless bike-sharing systems often suffer from indiscriminate parking, creating urban clutter. This paper proposes Weight Sensitive Local Search (WSLS), an incentive-based optimization approach to crowdsource bike repositioning. Unlike previous methods, WSLS incorporates an "ethical constraint" to limit user query fatigue. Simulations using real-world Singapore parking data demonstrate that WSLS improves repositioning rates by 34% over state-of-the-art baselines while maintaining a sustainable budget and protecting user wellbeing.
Introduction: The Challenge of Indiscriminate Parking
Dockless bike-sharing offers unparalleled convenience but leads to significant negative externalities when bikes are left in non-designated areas, blocking sidewalks and entrances. While nudging users during their active trips helps, it does not address the backlog of already misplaced bikes.
The authors address this by mobilizing the community through monetary incentives. However, a critical and often overlooked ethical dimension is "algorithmic spam." Constant low-value pings can degrade user experience and lead to platform abandonment. The goal is to maximize repositioning while respecting user attention as a finite resource.
Methodology: Weight Sensitive Local Search (WSLS)
The problem is modeled as a Revised Multi-dimensional Multiple Choice Knapsack Problem (MMKP). Each misplaced bike is a task with associated decisions: price ($p$), time slot ($h$), and expected number of queries ($pt$). The optimization is subject to two primary constraints: a total monetary budget ($B$) and a total query limit ($PT$) to prevent over-querying.
The proposed WSLS algorithm operates in four stages:
- Ranking: Tasks are prioritized by distance (shortest moves first).
- Initial Selection: The system selects price/query pairs that minimize the consumption of both budget and user attention.
- Feasibility Adjustment: Inefficient tasks are dropped if constraints are violated.
- Local Search: An iterative swapping process eliminates dominated choices to refine the solution.
Experimental Evaluation
The study utilized real-world spatial data for designated parking zones in Singapore, combined with a simulated environment of 10,000 bikes and up to 15,000 users. User behavior was modeled using a cumulative probability distribution, $F(d_l, h, p)$, representing the likelihood of a user accepting a price $p$ for a specific distance and time. This acknowledges that human behavior follows learnable patterns rather than being purely deterministic.
Key Results
WSLS was compared against four baselines: Random, Greedy, UCB, and KUBE. The results highlight the effectiveness of treating ethics as a formal constraint.
As shown in Figure 5, WSLS achieves the highest repositioning rate, outperforming the best baseline (KUBE) by 34%. Notably, it achieves these results with significantly fewer queries than the Greedy approach, demonstrating a superior trade-off between operational success and user experience. While Greedy might find lower prices for individual bikes, WSLS optimizes for the global fleet outcome.
Conclusion and Future Directions
By integrating user wellbeing directly into the optimization math, WSLS provides a sustainable model for community-driven urban maintenance. Future work aims to refine the learning of user acceptance distributions in volatile markets and explore social impacts, such as engaging senior citizens in repositioning tasks to promote active aging and community participation.
Summary updated