Semantic IDs improve item recommendation by tracing interests
From Interests to Semantic IDs: Retrieval-Grounded Credit Assignment for Generative Recommendation
Information RetrievalArtificial Intelligence
Summary
The paper focuses on improving how recommendation systems predict the next item a user might want to see, like in shopping or streaming services. Traditional methods only reward correct item predictions but ignore the reasoning process that led to those predictions, which limits learning. The authors introduce a method that breaks down the reasoning into smaller parts and checks which parts actually helped find the right item, giving more precise feedback. This approach helps the system learn better to recommend items by connecting interests to actual results.
What this means in practice
- •For e-commerce platform engineers: Create more accurate product recommendation models by attributing credit to specific interest-based queries that lead to successful item predictions.
- •For video streaming service developers: Enhance next-item content recommendations using semantic IDs linked to user interest queries that improve recall and ranking.
Authors
Mengdan Zhu, Yufan Zhao, Yao Zhao, Sophie Di, Tao Di, Yulan Yan, Sridhar Iyer, Liang Zhao
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.