Reasoning over Semantic IDs Enhances Generative Recommendation
Authors: Yingzhi He, Yan Sun, Junfei Tan, Yuxin Chen, Xiaoyu Kong, Chunxu Shen, Xiang Wang, An Zhang, Tat-Seng Chua
Organizations: National University of Singapore Singapore, Singapore · University of Science and Technology of China Hefei, China · Tencent Inc. Shenzhen, China
Recent advances in generative recommendation have leveraged pretrained LLMs by formulating sequential recommendation as autoregressive generation over a unified token space comprising language tokens and itemic identifiers, where each item is represented by a compact sequence of discrete tokens, namely Semantic IDs (SIDs). This SID-based formulation enables efficient decoding over large-scale item corpora and provides a natural interface for LLM-based recommenders to leverage rich world knowledge. Meanwhile, breakthroughs in LLM reasoning motivate reasoning-enhanced recommendation, yet effective reasoning over SIDs remains underexplored and challenging. Itemic tokens are not natively meaningful to LLMs; moreover, recommendation-oriented SID reasoning is hard to evaluate, making high-quality supervision scarce. To address these challenges, we propose SIDReasoner, a two-stage framework that elicits reasoning over SIDs by strengthening SID--language alignment to unlock transferable LLM reasoning, rather than relying on large amounts of recommendation-specific reasoning traces. Concretely, SIDReasoner first enhances SID-language alignment via multi-task training on an enriched SID-centered corpus synthesized by a stronger teacher model, grounding itemic tokens in diverse semantic and behavioral contexts. Building on this enhanced alignment, SIDReasoner further improves recommendation reasoning through outcome-driven reinforced optimization, which guides the model toward effective reasoning trajectories without requiring explicit reasoning annotations. Extensive experiments on three real-world datasets demonstrate the effectiveness of our reasoning-augmented SID-based generative recommendation. Beyond accuracy, the results highlight the broader potential of large reasoning models for generative recommendation, including improved interpretability and cross-domain generalization.
Figures & tables
Figure 1. Illustration of the overall framework of our proposed SIDReasoner.
Figure 2. Illustration of enriched alignment corpus.
Dataset
#Items
#Interactions
Train
Val&Test
Games
3,858
61,417
49,133
6,142
Office
3,459
48,656
38,924
4,866
Industrial
3,686
45,325
36,259
4,533
Table 1. The statistics of recommendation datasets.
Models
Games
Office
Industrial
R@5
N@5
R@10
N@10
R@5
N@5
R@10
N@10
R@5
N@5
R@10
N@10
Caser
0.0376
0.0241
0.0659
0.0332
0.0880
0.0663
0.1114
0.0738
0.0664
0.0528
0.0852
0.0588
GRU4Rec
0.0329
0.0219
0.0599
0.0305
0.0682
0.0480
0.0974
0.0574
0.0788
0.0578
0.1030
0.0649
SASRec
0.0501
0.0345
0.0723
0.0416
0.1019
0.0824
0.1167
0.0871
0.0807
0.0647
0.0964
0.0697
TIGER
0.0489
0.0300
0.0763
0.0402
0.1270
0.1037
0.1429
0.1121
0.1003
0.0823
0.1325
0.0924
HSTU
0.0539
0.0396
0.0746
0.0462
0.1204
0.1069
0.1323
0.1107
0.1008
0.0898
0.1138
0.0940
Table 2. Performance comparison over different baseline methods. R is shorts for Recall, N is short for NDCG.
Figure 3. Performance comparison on cross-domain recommendation setting.
Figure 4. Ablation performance of different alignment strategies with Best-of- N reasoning selection.
Models
Games
Office
R@10
N@10
R@10
N@10
S1 = Multi-task Alignment
0.0414
0.0206
0.0388
0.0201
- After RL
0.0741
0.0355
0.1539
0.0970
S2 = S1 + Enriched Alignment
0.0632
0.0320
0.0619
0.0363
- After RL
0.0957
0.0470
0.1607
0.1178
S3 = S2 + General Reasoning
0.0806
0.0450
0.1231
0.0883
Table 3. Ablation performance of different alignment strategies before and after reinforcement learning optimization.
Models
MMLU
IFEVAL
GSM8K
Vanilla Qwen3-1.7B
0.6085
0.1793
0.6850
S1 = Multi-task Alignment
0.2760
0.0906
0.0060
S2 = S1 + Enriched Alignment
0.4464
0.0739
0.0330
S3 = S2 + General Reasoning
0.5580
0.1497
0.5430
Table 4. Performance comparison over general abilities.
Enrichment
R@10
N@10
Metadata
0.0703
0.0466
Qwen3-8B
0.0727
0.0480
Qwen3-32B
0.0858
0.0498
GPT-4o-mini
0.1031
0.0563
Table 5. Performance comparison over different teacher models for corpus enrichment on Games dataset.
Figure 5. Case study of how explicit reasoning yields effective and explainable recommendations.
Figure 6. Reasoning length and performance change during RL Training on Games Dataset.
Large language model-based recommendation uses language models to personalize ranking, retrieval, and conversational suggestions from text and behavior.
Generative recommendation reformulates item retrieval as sequence generation, allowing a unified model to directly generate the next item from a user's interaction history. Semantic IDs further make this paradigm effective and scalable by representing each item as discrete codes, enabling knowledge sharing among semantically related items. Recent studies introduce explicit reasoning before Semantic-ID generation, helping models summarize user interests and infer possible preference transitions. However, reasoning is not inherently beneficial: Inaccurate or uninformative reasoning may mislead subsequent item generation and ultimately degrade recommendation performance. This raises a central challenge: how can a recommender select and learn effective reasoning traces and progressively evolve toward better reasoning from its own generations? In this work, we propose Evo-Rec, a three-stage framework for learning better reasoning and further enhancing it through reinforcement learning. First, we align Semantic IDs with their textual and behavioral contexts, enabling the model to understand and generate item identifiers. Second, we sample multiple candidate reasoning traces and retain those that improve the prediction of the ground-truth item, providing a stronger reasoning initialization through supervised fine-tuning. Third, we further optimize the reasoning policy through reinforcement learning with catalog-constrained item generation and ranking-aware recommendation feedback. Experiments on three Amazon Review benchmarks show that Evo-Rec consistently outperforms discriminative, generative, and reasoning-enhanced recommenders across all evaluation metrics. These results demonstrate the effectiveness of our framework in learning better reasoning for SID-based generative recommendation.
Mengdan Zhu, Yufan Zhao, Sophie Di +5
Emory University Atlanta, GA, USA · Microsoft Redmond, WA, USA · Cornell University Ithaca, NY, USA
Semantic IDs (SIDs) are now a central component of generative recommendation. Current SID-based systems assign three roles to the same token sequence. Shared prefixes are intended to organize related items, the complete SID identifies an individual item, and each generated token narrows the items that can still be returned. We systematically investigate SIDs from item encoding and SID construction to autoregressive generation and final recommendation. We examine how SID construction changes item representations and how those changes affect generation. Across three Amazon domains and eight SID constructions, SID neighborhoods recover only 32.2% of the encoder's ten nearest neighbors on average. Alternative item descriptions still retrieve the corresponding item first in 99.57% of controlled cases, yet change 38.4% of exact SIDs. These results show that SIDs retain broad organization but lose much of the encoder's fine local structure, while their exact tokens are not determined by item meaning alone. This loss becomes consequential during generation. After the final semantic token, TIGER retains only 29.9% of held-out targets that were plausible recommendations before SID filtering. Motivated by these findings, we propose Item-Supported Decoding (ISD), a lightweight inference-time method that allows a user-specific item ranking to support corresponding SID prefixes before beam search discards them. The same ranking then orders the generated items. ISD requires no additional parameters or retraining of the SID constructor or decoder. We empirically show that ISD improves NDCG@10 over the corresponding SID backbone in every evaluated setting, with relative gains of up to 31.2%. Our results show that SIDs provide useful coarse item organization, but their fine boundaries should not alone determine which items remain available during generation.
Junting Wang, Xinrui He, Yunzhe Li +1
University of Illinois Urbana-Champaign Urbana, Illinois, USA
Generative recommendation commonly represents items using fixed-length semantic identifiers (SIDs) constructed through clustering and quantization. However, these artificial codes may overcompress item semantics, remain misaligned with pretrained LLM vocabularies, and require costly autoregressive decoding. In light of this, we propose VaLiDRec, a generative recommendation framework based on variable-length, LLM-aligned semantic identifiers. VaLiDRec constructs SIDs directly from informative native LLM vocabulary tokens via token importance estimation, semantic-quality-aware pruning, and collision-aware refinement, allowing identifier lengths to adapt to item semantic complexity. To model user preferences, VaLiDRec incorporates graph-aware soft prompts and reformulates recommendation as token-set prediction with token-level item scoring, eliminating autoregressive SID generation and beam search. Experiments on four real-world datasets show that VaLiDRec consistently outperforms strong sequential and generative recommendation baselines across all evaluation metrics. It further achieves superior zero-shot item cold-start performance and 87.49× faster inference than LC-Rec. These results demonstrate that LLM-native variable-length semantic identifiers provide a more expressive and efficient paradigm for generative recommendation.
Shutong Qiao, Wei Yuan, Tong Chen +3
University of Queensland Brisbane, Australia · Computer Network Information Center, Chinese Academy of Sciences Beijing, China · Griffith University Gold Coast, Australia