From Interests to Semantic IDs: Retrieval-Grounded Credit Assignment for Generative Recommendation

2026-09-24

AI Summary

This study addresses the challenge of sparse rewards in generative recommendation, which leads to inadequate credit assignment by overlooking discrepancies across reasoning trajectories. To this end, it proposes a retrieval-grounded query attribution framework that decomposes each reasoning trajectory into historical summarization, interest hypotheses, and final semantic IDs (SIDs). A frozen retriever independently verifies each hypothesis, enabling span-level fine-grained credit assignment via group relative policy optimization. This approach overcomes the reward bottleneck inherent in relying solely on final SIDs by rendering intermediate steps verifiable. Extensive experiments on three Amazon datasets demonstrate significant improvements in recommendation performance. Furthermore, oracle analyses confirm that interest-conditioned decoding effectively enhances both recall and ranking metrics.

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.

Problem

• Generative Recommendation
• Semantic IDs
• Credit Assignment
• Policy Optimization
• Reasoning Traces

Innovation

• Semantic IDs
• Generative Recommendation
• Credit Assignment
• Retrieval-Grounded Attribution
• Group-Relative Policy Optimization