Search papers, labs, and topics across Lattice.
This paper introduces Reward Guided Decoding (RGD), a novel framework that enhances generative recommendation systems by integrating business value into the decoding process. By formulating value-guided decoding as a KL-regularized reward maximization problem, RGD effectively combines generation probability with reward signals, allowing for real-time adjustments without the need for costly retraining. Extensive experiments show that RGD significantly improves alignment between personalization and business objectives, leading to successful deployment on the Kuaishou platform with measurable performance gains.
RGD reshapes the decoding process in generative recommendation, ensuring high-value candidates are prioritized without retraining the model.
Generative recommendation formulates recommendation task into an SID sequence autoregressive generation paradigm, but the decoding process is often dominated by generation likelihood. This may conflict with real-world business objectives, where high-value candidates can receive low generation probability and be pruned early during beam search. Existing reranking or training-time alignment methods either intervene too late or require costly model retraining when business preferences change. To this end, we propose \textbf{R}eward \textbf{G}uided \textbf{D}ecoding, named \textbf{RGD}, a controllable decoding framework for industrial value-oriented generative recommendation. We formulate value-guided decoding as a KL-regularized reward maximization problem, deriving a closed-form reward guided decoding distribution that principledly combines generation probability with reward signals. RGD treats the base generator as a reference policy and introduces a reward model as a test-time controller, injecting reward into each decoding step to reshape the search trajectory without retraining the generator. Extensive offline and online experiments demonstrate the effectiveness of our approach for aligning personalization and business value. RGD has been deployed on the Kuaishou platform, bringing consistent improvements in real-world recommendation scenarios.
