A Deployment Study of a Data-Driven Volunteer Engagement System for Food Security

The challenges of food waste and insecurity arise in wealthy and developing nations alike, impacting millions of livelihoods. A major force to combat food waste and insecurity, food rescue organizations match food donations to the community organizations that serve the low-resource population. However, they rely on external volunteers to pick up and deliver the food, and this brings significant uncertainty to the food rescue operation. We work with a large food rescue organization, 412 Food Rescue (412FR), to address this uncertainty. We make the following contributions: (i) We develop a data-driven optimization algorithm to compute the optimal generic intervention and notification scheme applicable to all rescues. The recommended actions have been adopted by 412FR and are shown to have improved the rescue efficiency. (ii) We develop a rescue-specific recommender system to send push notifications to the most likely volunteers for each given rescue. We leverage a mathematical programming-based approach to diversify our recommendations and propose an online algorithm to dynamically select the volunteers to notify without the knowledge of future rescues. Our recommendation system improves the hit ratio from $44\%$ achieved by the previous method to $73\%$ on historical data. (iii) We designed and ran a randomized controlled trial for the recommender system. The trial showed that the algorithm significantly improved the claim rate and hit rate.

History: This paper was refereed.

Funding: This work was supported in part by NSF grant IIS-2046640 (CAREER) and a Google Academic Research Award.