Search papers, labs, and topics across Lattice.
This paper introduces GenRec, a generative recommendation framework designed for large-scale industrial deployment that tackles challenges related to inconsistent outputs, long user behavior sequences, and preference alignment. GenRec employs a Page-wise NTP training objective for denser gradient signals, an asymmetric linear Token Merger to compress multi-token Semantic IDs, and GRPO-SR, a reinforcement learning method combining Group Relative Policy Optimization with NLL regularization and hybrid rewards. Online A/B tests on the JD App show GenRec achieves significant improvements in click and transaction counts compared to the existing pipeline.
GenRec proves that generative recommendation can beat existing pipelines in a large-scale industrial setting, achieving nearly 10% gains in key metrics by focusing on preference alignment and efficient sequence encoding.
Generative Retrieval (GR) offers a promising paradigm for recommendation through next-token prediction (NTP). However, scaling it to large-scale industrial systems introduces three challenges: (i) within a single request, the identical model inputs may produce inconsistent outputs due to the pagination request mechanism; (ii) the prohibitive cost of encoding long user behavior sequences with multi-token item representations based on semantic IDs, and (iii) aligning the generative policy with nuanced user preference signals. We present GenRec, a preference-oriented generative framework deployed on the JD App that addresses above challenges within a single decoder-only architecture. For training objective, we propose Page-wise NTP task, which supervises over an entire interaction page rather than each interacted item individually, providing denser gradient signal and resolving the one-to-many ambiguity of point-wise training. On the prefilling side, an asymmetric linear Token Merger compresses multi-token Semantic IDs in the prompt while preserving full-resolution decoding, reducing input length by ~2X with negligible accuracy loss. To further align outputs with user satisfaction, we introduce GRPO-SR, a reinforcement learning method that pairs Group Relative Policy Optimization with NLL regularization for training stability, and employs Hybrid Rewards combining a dense reward model with a relevance gate to mitigate reward hacking. In month-long online A/B tests serving production traffic, GenRec achieves 9.5% improvement in click count and 8.7% in transaction count over the existing pipeline.
