This work addresses the credit-assignment gap in SID recommendation with retrieval-grounded query attribution, and reveals the potential of interest-conditioned SID decoding: selecting the target-relevant query among generated interests improves both recall and ranking.
Abstract
Semantic IDs (SIDs) encode each catalog item as a short token sequence, enabling generative recommenders to predict the next item autoregressively. Reasoning-enhanced variants, an increasingly common extension, first generate a textual trace and then decode a next-item SID by beam search. Such recommenders are commonly trained with group-relative policy optimization under an exact-match SID reward, which is sparse in large catalogs. Two failure modes follow. When all rollouts in a group miss the target, the group yields zero advantage and no learning signal. Rollouts sharing the same SID reward receive identical advantages, however much their traces differ. In both cases the reward reflects only the decoded SID, never the reasoning that produced it. This creates a credit-assignment gap. We address this gap with retrieval-grounded query attribution. Each trace is structured into a history summary, a set of interest hypotheses, and a final SID. A frozen retriever executes every hypothesis as a catalog query, so that each hypothesis becomes independently verifiable rather than judged only through the final SID. A rollout is rewarded when any of its queries retrieves the target within the \mbox{top-$K$}, and per-query hit indicators localize that reward to individual hypotheses. Credit is thus assigned at the span level: only hypotheses that individually hit receive positive retrieval advantage, while the retrieval channel never updates the final SID span. Rollouts that share a SID reward can therefore receive different updates. Across experiments on three Amazon Reviews datasets, this yields consistent improvements in SID recommendation. On Video Games, an oracle analysis further reveals the potential of interest-conditioned SID decoding: selecting the target-relevant query among generated interests improves both recall and ranking.
Difficulty-Aware Semantic-ID Optimization (DASO), a tree-aware post-training method that addresses failure mode as an online rollout-allocation problem and improves over MiniOneRec-style GRPO on 11 of 12 metrics and achieves the best result on 9 of 12 metrics.
Generative recommenders retrieve items by autoregressively decoding semantic IDs (SIDs). The standard autoregressive SID interface (SID-AR) represents each history item with K code tokens, expanding a T-item history to TK tokens, while trie-constrained beam search repeatedly re-enters the full backbone during generatio...
Danil Gusak, A. Volodkevich, Evgeny Frolov· Proceedings of the 20th ACM...· 3 citations
The experiments show single-head beam search consistently beats MTP on most product-critical metrics: ranking and content QA, and trains at roughly 2.6x lower dollar cost on Qwen3-8B scale; MTP wins only on raw inference throughput.
Sejoon Oh, Fernando Amat Gil, Evan Smothers et al.· 0 citations
Evo-Rec is proposed, a three-stage framework for learning better reasoning and further enhancing it through reinforcement learning that consistently outperforms discriminative, generative, and reasoning-enhanced recommenders across all evaluation metrics.
Meng-Dan Zhu, Yu-Fan Zhao, Sophie Di et al.· 0 citations
This work asks a simple question: can the model's own SID tree serve as the action abstraction for that OPE, and explains how resolution depth is the operative knob and a conditional bias bound links the coarsening bias to the quantizer's worst-case reconstruction residual and the target-logging divergence.
Generative recommendation (GR) of social media platforms commonly retrieves items by autoregressively decoding discrete Semantic IDs (SIDs) with finite-width beam search. This work examines how this output interface constrains whether held-out targets enter finite returned lists, while distinguishing catalog-wide score...
Yu Sun, Xiang Ren, Zuo-Dong Xiang et al.· 1 citation