ReCast: Recasting Learning Signals for Reinforcement Learning in Generative Recommendation

TL;DR

ReCast introduces a novel framework that improves reinforcement learning in generative recommendation by repairing and contrasting learning signals, leading to significant performance and system efficiency gains.

Contribution

It proposes a repair-then-contrast learning framework that enhances learnability and efficiency in sparse-hit generative recommendation RL tasks.

Findings

ReCast achieves up to 36.6% relative improvement in Pass@1.

ReCast reduces actor-side update time by 16.60x.

ReCast lowers peak memory usage by 16.5%.

Abstract

Generic group-based RL assumes that sampled rollout groups are already usable learning signals. We show that this assumption breaks down in sparse-hit generative recommendation, where many sampled groups never become learnable at all. We propose ReCast, a repair-then-contrast learning-signal framework that first restores minimal learnability for all-zero groups and then replaces full-group reward normalization with a boundary-focused contrastive update on the strongest positive and the hardest negative. ReCast leaves the outer RL framework unchanged, modifies only within-group signal construction, and partially decouples rollout search width from actor-side update width. Across multiple generative recommendation tasks, ReCast consistently outperforms OpenOneRec-RL, achieving up to 36.6% relative improvement in Pass@1. Its matched-budget advantage is substantially larger: ReCast reaches the baseline's target performance with only 4.1% of the rollout budget, and this advantage widens with model scale. The same design also yields direct system-level gains, reducing actor-side update time by 16.60x, lowering peak allocated memory by 16.5%, and improving actor MFU by 14.2%. Mechanism analysis shows that ReCast mitigates the persistent all-zero / single-hit regime, restores learnability when natural positives are scarce, and converts otherwise wasted rollout budget into more stable policy updates. These results suggest that, for generative recommendation, the decisive RL problem is not only how to assign rewards, but how to construct learnable optimization events from sparse, structured supervision.