Abstract
Complex multistage recommender systems that utilize multiple stages of ranking models and non-model parameters at different stages are used to identify a set of items as output, e.g., advertisements to be delivered by an advertising network. This disclosure describes a framework to optimize the non-model parameters to improve recall, defined based on items delivered as output in comparison to groundtruth items. The optimization can be performed offline, using a simulation that takes as input candidate items and labels of items that are known positives. The optimization can improve the quality of recommendations and can reduce computational cost.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Anonymous, "End-to-end Optimization of Multistage Recommender Systems", Technical Disclosure Commons, (May 29, 2020)
https://www.tdcommons.org/dpubs_series/3281