Ranking items using compact semantic codes matches long text methods
Reasoning Reranking with Semantic IDs as the Sole Item Representation
Information RetrievalComputation and LanguageMachine Learning
Summary
Many recommendation systems rely on reading detailed descriptions of items to decide which to suggest. This paper tests what happens if the system knows only short semantic codes representing those items, without reading full text. The authors find that with special training, models can rank items well just from these codes, saving space and sometimes using hidden user behavior signals. However, models that rank well from codes do not always describe items accurately in their explanations, suggesting more work is needed to improve reasoning quality.
What this means in practice
- •For recommender system developers: Use compact semantic IDs instead of full text to efficiently rerank item lists with similar accuracy but shorter input prompts.
- •For mobile app engineers: Implement item ranking that requires less memory and bandwidth by relying on short semantic codes instead of long item descriptions.
Authors
Zhenwei Tang, Jintao Liu, Di Bai, Xidong Wu, Xinwu Cheng
Abstract
Reasoning rerankers write a rationale before ranking a candidate slate, and existing systems read item metadata to do so, even though generative recommenders increasingly represent items as compact Semantic IDs (SIDs). Ranking from SIDs alone has two appeals: SID prompts are far shorter than the matched item text, and although item codes are fixed from content, the backbone learns their token embeddings in part from user-item interaction sequences, so the tokens can carry behavioral signal that item text lacks. That training targets next-item generation, and whether it supports ranking a supplied slate from SIDs alone is untested. We build ReRankID, a controlled testbed in which models reason and rank from a history and slate shown only as SIDs, varying item rendering, history length, and slate size one at a time. On this testbed, such models often copy the input order instead of reranking, and readable metadata does not repair the failure. We compare supervised fine-tuning, rejection-sampled self-training, on-policy distillation, and verifier-based reinforcement learning as post-training methods and find that they elicit the missing behavior. The strongest post-trained model, the 8B model trained with a graded ranking reward, gives the best SID-only ranking and matches strong text-reading LLMs whose prompts are about 20x longer. It ranks above random with no history at all, where a caption-reading reference does not, and it keeps its gains at long histories, where text prompts pass 50k tokens and the caption-reading reference stops improving. In a paired rationale audit, this model argues trade-offs better than that reference but describes items less accurately. The common post-training methods we test thus elicit ranking from SIDs but not faithful item descriptions, which calls for post-training that improves the reasoning itself, not only the ranking it produces.