End-to-end generative model improves recommendation retrieval and ranking efficiency
GRP v0.1 Technical Report
Information Retrieval
Summary
Recommendation systems usually use separate steps to find and rank items for users, which can make upgrades tricky. The authors introduce GRP, a model that combines finding, scoring, and learning rewards for recommendations all in one system. They tested it offline and online, showing it can speed up item retrieval and improve user engagement slightly compared to existing methods. Their approach also supports gradually changing how recommendations are made without breaking the system.
What this means in practice
- •For software engineers: Build recommendation systems that integrate retrieval and ranking in one model to improve efficiency and ease incremental upgrades.
- •For advertising technology teams: Use a joint generative approach to optimize item selection and ranking while maintaining stable user engagement metrics in online platforms.
Authors
Wenfeng Zhuo, Vincent Xue, Charles Wei, Cong Ni, Ruiming Lu, Jiwen Ren, Mo Li, Peng Yang, Xufei Wang, Dongheng Li, Jiacong He, Yi Song, Yufei Fan, Mikhail Obukhov, Yiwen Chen, Yvette Liu, Yin Ye, Chengjie Wu, Mingtao Zhang, Jinchao Ye, Lili Zhang, Chunhui Zhu
Abstract
Industrial recommendation systems rely on multi-stage cascades whose retrieval, ranking, and serving components are difficult to replace jointly. We present GRP, a generative recommendation framework that combines retrieval, ranking, and reward modeling in a single encoder-decoder model, and evaluate a progressive path toward end-to-end recommendation. The model generates multimodal Semantic IDs and scores candidates with a jointly trained ranking module. The frozen ranking module then supplies rewards for reinforcement-learning post-training. We introduce mGRPO, which adds a reference-anchored margin to reward optimization to preserve the likelihood of logged targets. Offline experiments examine history encoding, model capacity allocation, event selection, tokenization, and reward discrimination. Serving optimizations reduce end-to-end retrieval latency by 69%. Online experiments evaluate the model as a retrieval source, with early-ranking bypass, and with replacement of weaker sources. In a retrieval-only comparison, view time increases by 0.46% and shares by 0.77% relative to production. A separate comparison combining bypass and source replacement yields increases of 0.82% in view time and 2.56% in shares, with neutral platform-level guardrails. These results support progressive deployment while identifying remaining gaps in ranking quality and performance across recommendation metrics.