Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Recommendation Strategies Under Stochastic Supply in Surplus Food Platforms
Digital surplus-food platforms operate under opaque selling schemes and uncertainty in both supply and demand, making real-time ranking decisions crucial to platform performance. In these settings, recommendation algorithms do not only shape consumer choice but also deter- mine how demand is allocat...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
AUC Knowledge Fountain
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| Summary: | Digital surplus-food platforms operate under opaque selling schemes and uncertainty in both supply and demand, making real-time ranking decisions crucial to platform performance. In these settings, recommendation algorithms do not only shape consumer choice but also deter- mine how demand is allocated across stores with heterogeneous inventory reliability. While most recommender systems are designed to maximize engagement or revenue, these objectives alone can lead to highly concentrated demand, increased cancellations, and inefficient surplus allocation.
In this work, we study ranking strategies for surplus-food marketplaces under stochastic supply realization and irreversible customer decisions. We develop a discrete-event simulation framework grounded in real-world platform dynamics and calibrated using operational data extracted from Too Good To Go (TGTG), which follows our platform dynamic structure. Within this framework, we evaluate low-tech, inventory-aware ranking algorithms that dynamically regulate demand allocation, including penalty-based, mixed, and two-stage hybrid strategies. These algorithms are compared against greedy-based baselines and a point-wise learning-to- rank model to assess trade-offs between system-level, computational cost, and customer-level outcomes.
Our results show that simple, inventory-aware-based ranking strategies can effectively re- distribute demand across stores, reducing food waste and order cancellations, while maintaining competitive revenue levels. This is mainly without relying on complex predictive models or heavy learning pipelines, highlighting the effectiveness of light-lightweight rank- ing mechanisms. The findings highlight that, under opaque selling and supply uncertainty, low-tech, interpretable ranking mechanisms provide a robust and deployable alternative to high-complexity recommender systems, aligning platform-level performance with long-term operational efficiency. |
|---|