Mitigating Collaborative Semantic ID Staleness in Generative Retrieval

TL;DR

This paper addresses the problem of stale semantic IDs in generative retrieval systems caused by evolving user-item interactions, proposing a lightweight SID alignment update to improve retrieval performance and reduce retraining costs.

Contribution

It introduces a model-agnostic SID alignment update method that maintains compatibility with existing vocabularies, enabling efficient fine-tuning without full retraining.

Findings

Consistently improves Recall@K and nDCG@K across three benchmarks.

Reduces retriever training compute by approximately 8-9 times.

Effectively mitigates SID staleness caused by temporal drift.

Abstract

Generative retrieval with Semantic IDs (SIDs) assigns each item a discrete identifier and treats retrieval as a sequence generation problem rather than a nearest-neighbor search. While content-only SIDs are stable, they do not take into account user-item interaction patterns, so recent systems construct interaction-informed SIDs. However, as interaction patterns drift over time, these identifiers become stale, i.e., their collaborative semantics no longer match recent logs. Prior work typically assumes a fixed SID vocabulary during fine-tuning, or treats SID refresh as a full rebuild that requires retraining. However, SID staleness under temporal drift is rarely analyzed explicitly. To bridge this gap, we study SID staleness under strict chronological evaluation and propose a lightweight, model-agnostic SID alignment update. Given refreshed SIDs derived from recent logs, we align them to the existing SID vocabulary so the retriever checkpoint remains compatible, enabling standard warm-start fine-tuning without a full rebuild-and-retrain pipeline. Across three public benchmarks, our update consistently improves Recall@K and nDCG@K at high cutoffs over naive fine-tuning with stale SIDs and reduces retriever-training compute by approximately 8-9 times compared to full retraining.