Skip to content

Off-Policy Evaluation for Semantic ID Recommenders: Does the Model's Own Code Hierarchy Help?

Aug 2026 · 1 citation · 17 references
Computer Science

TL;DR

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.

Abstract

Generative recommenders increasingly emit semantic IDs (SIDs): each item is a short sequence of hierarchical discrete codes from a residual quantizer, decoded autoregressively. Before spending scarce A/B-test, a team may decide offline which decoder or reranking variants are worth testing - a job for off-policy evaluation (OPE). We ask a simple question: can the model's own SID tree serve as the action abstraction for that OPE? Our answer has three parts. (i) Under the near-argmax logging real recommenders use, per-item OPE is hopeless - as item-level effective sample size is usually small on production logs - but marginalizing items to code-prefix clusters restores estimable support and cuts error. (ii) This gain is thanks to coarsening, not to the hierarchy specifically; but the SID tree is what makes coarsening feasible in a generative system - each cluster's mass is exactly and cheaply returned by the decoder, whereas flat clustering requires enumerating item/leaf masses that a code-only decoder does not directly expose. (iii) Resolution depth is the operative knob - coarser under scarce support - and a conditional bias bound links the coarsening bias to the quantizer's worst-case reconstruction residual and the target-logging divergence.

View source

Similar papers

#artificial intelligence Review Sep 2026

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

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.

Meng-Dan Zhu, Yu-Fan Zhao, Yao Zhao et al. · 0 citations
Preprint Aug 2026

Difficulty-Aware Semantic-ID Optimization for Generative Recommendation

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.

Xin Yu, Stephen Li, S. Aghaei et al. · 1 citation
Book Open access Sep 2026

CoSID: Concept-Conditioned Semantic-ID Decoding for Efficient Generative Recommendation

CoSID is introduced, a concept-conditioned SID decoder that encodes the history once into a compact next-item concept and delegates the entire beam search to a lightweight KV-cached decoder.

Danil Gusak, A. Volodkevich, Evgeny Frolov · 1 citation
#artificial intelligence Book Open access Aug 2026

Pairwise Ranking Outperforms Single-Action RL for Offline Explanation Selection: A Practical Lesson

This work separates generation from selection: explanations are produced ahead of time as a frozen candidate pool (six prompt styles, two commodity LLMs), and a small CPU-resident selector picks one at request time, which needs no GPU and returns in under 100 ms.

T. Chowdhury, Saeideh Shahrokh Esfahani · 0 citations
Book Open access Sep 2026

Not All Branches Are Equal: Adaptive Semantic ID Construction for Generative Recommendation

Generative recommendation models represent items as sequences of discrete tokens known as semantic IDs. Existing approaches derive these identifiers using residual or hierarchical vector quantization, assuming that tokens carry independent semantic meaning across quantization levels. Revisiting this assumption, we argu...

Guy Hadad, Haggai Roitman, Bracha Shapira · 0 citations
Preprint Aug 2026

SWIM: Step-Wise Integrated Measure for Session-supervised List Evaluation in Generative Re-ranking

SWIM (Step-Wise Integrated Measure), a list-level evaluator that models user behaviors as a finite-horizon prefix session-level survival process, and efficiently estimates continuation probabilities and utilities in parallel, satisfying strict industrial latency constraints.

Yuan Pu, Cheng-Hao Zhang, Chao Feng et al. · 0 citations

Related blog posts

Microsoft Research Blog Nov 16, 2026

TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient

AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.