About this document
Reinforced Preference Optimization for Recommendations by kerveros_99 is a document available to read on EtoBox.
The document introduces Reinforced Preference Optimization for Recommendation (ReRe), a novel approach that enhances generative recommender systems by addressing limitations in negative modeling and reward reliance. By employing constrained beam search and augmenting rule-based accuracy rewards with ranking rewards, ReRe improves sampling efficiency and diversifies negative samples, leading to superior ranking performance across various datasets. The study also explores the design space of reinforcement lea
- Author
- kerveros_99
- Language
- EN