Pith. sign in

REVIEW 3 major objections 5 minor 58 references

SmartGR: Hierarchy and Beam-Aware Knowledge Distillation for Generative Recommendation

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read SmartGR distills a large generative recommender into a small one with two added losses—learnable depth weights over semantic-ID levels and a pairwise beam-ranking term—and reports an average 8.6% quality gain with inference speed intact.

desk verdict Solid, well-ablated distillation paper for generative recommendation; the beam-aware loss mechanism is less clean than claimed, but the empirical package earns a serious look. read the letter →

arxiv 2608.02048 v1 pith:EXU7YC4E submitted 2026-08-03 cs.IR

Ziheng Zhang , Yu Cui , Bohao Wang , Yong He , Chao Yu , Chuan Yuan , Wujie Sun , Can Wang
show 1 more author
Jiawei Chen
This is my paper · ORCID
classification cs.IR
keywords generativerecommendationknowledgedistillationsemanticIDsbeamsearchhierarchy-awareweightingrankingautoregressivemodelcompression
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Generative recommenders represent items as hierarchical semantic IDs and rank candidates with beam search, and they obey a scaling law: larger models are more accurate but too slow for production. This paper claims that most of that accuracy gap can be transferred to a small model without changing its architecture or inference cost, using two loss terms placed exactly where the teacher's advantage lives. The first term weights per-level distillation by a learned monotone function of semantic-ID depth, because the teacher's edge grows at deeper, finer-grained levels. The second term teaches the student the teacher's ranking of neighboring prefixes during beam search, so a high-scoring item is not pruned because one intermediate prefix looked weak. Across four datasets, the distilled student improves all sixteen metrics by 2.3%–17.5% over the base student, averages 8.6% higher quality, and keeps a 1.92×–2.89× inference speedup over the much larger teacher.

What carries the argument

The load-bearing object is the cached teacher-beam set $\mathcal{B}(x)$ together with a compatibility mask over semantic-ID levels. For each training context $x$, SmartGR picks the teacher beam $b_{k^*}$ that shares the longest prefix with the target SID, and defines $\mathcal{L}(x)=\{1,\dots,L^*\}$ as the levels of that shared prefix; distillation is restricted to these levels because only they carry teacher signal consistent with the target. Hierarchy-Aware SID Distillation computes a normalized, monotone depth weight $w_\ell(x)=\exp(a_\ell)/\sum_{j=1}^{L^*}\exp(a_j)$ with $a_\ell=f_\theta(\ell/L)$ and $f_\theta(d)=\tanh(\theta d/\tau_{\mathrm{lev}})/\tanh(\theta/\tau_{\mathrm{lev}})$, then applies it as a weight on a KL divergence between teacher and student conditional SID distributions at each level in $\mathcal{L}(x)$. Beam-Aware Ranking Distillation takes the selected beam $b_{k^*}$ as the positive and its next lower-ranked beam $b_{k^*+1}$ as the hard negative; at each shared level $\ell$ it turns their cumulative prefix scores into teacher and student preference distributions via softmax after dividing by $\ell$, and minimizes $\tau^2\,\mathrm{KL}(\pi^\ell_T\,\|\,\pi^\ell_S)$. Dividing by $\ell$ removes the length effect of accumulated log-probabilities, and the adjacent negative keeps the contrast hard while preserving the teacher's ordering. The full objective is hard cross-entropy on the target SID plus $\lambda_{\mathrm{SID}}L_{\mathrm{SID}} + \lambda_{\mathrm{BEAM}}L_{\mathrm{BEAM}}$.

What would settle it

A decisive test: on a new dataset, compute the correlation between the teacher's cumulative beam scores and whether those prefixes lead to correctly retrieved target items; SmartGR predicts that distillation gains track this calibration, so a dataset where teacher rankings are poorly calibrated—or a shuffled-score ablation—should show the two losses providing little or no gain over hard supervision alone.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that the two failures of generic distillation for generative recommendation are measurable and correctable: teachers gain most at deep semantic-ID levels, and per-position distribution matching does not protect a high-final-scoring item from being pruned on a weak prefix. SmartGR responds with two objectives that use the same cached teacher beams. A hierarchy-aware SID loss restricts itself to the longest teacher beam prefix compatible with the target SID and weighs the KL divergence at each level by a learned monotone function of normalized depth. A beam-aware ranking loss constructs preference distributions from cumulative prefix scores of that positive beam and its next lower-ranked neighbor, and matches those distributions at each shared level. The paper reports that the combined objective improves every one of the 16 evaluation metrics relative to the base student, wins 15 of 16 against the strongest distillation baselines, and does so while keeping the student's faster inference, so the teacher's beam-search ranking behavior is what transfers.

Load-bearing premise

The load-bearing premise is that the teacher's beam scores and rankings are a trustworthy source of supervision: if the longest prefix the teacher shares with the correct answer is often a low-quality path, or the adjacent beam is not a meaningful contrast, the distilled signal will mislead the student rather than help it.

Editorial extensions

If this is right

  • GR practitioners can compress a large teacher into a small autoregressive student without redesigning the decoder, since SmartGR only adds loss terms to the existing training objective.
  • An offline teacher cache suffices: caching the top-16 teacher token distributions and two scored beam prefixes per example avoids running the teacher during student training, and the BEAM loss forced-scores only two student sequences per example.
  • The method's gains grow with teacher beam width: widening the cached teacher beam from 2 to 16 raises recommendation quality by 8.5% on average while only 1.53x the distillation time.
  • The learned hierarchy weights are interpretable and monotone—deeper SID levels receive larger weights on both Amazon and Kuaishou—so the student is forced to pay more attention where the teacher's advantage is largest.
  • Because the student architecture is unchanged, the distilled model inherits the student's inference cost, giving a 1.92x–2.89x speedup over the 8B teacher depending on dataset.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A testable extension follows from the paper's own per-level-weighting ablation: pure per-level flexibility did not beat monotone depth weighting, which suggests the coarse-to-fine prior itself, not extra capacity, is the active ingredient; this predicts that monotone weighting will transfer to any hierarchy with ordered specificity, such as product taxonomies or code ASTs.
  • The coverage constraint bites on low-overlap domains: the paper reports that fewer than 5% of Kuaishou examples retain a target-compatible prefix of length three or more. An extension not explored here would synthesize target-compatible teacher prefixes when none exist, which could extend SmartGR's gains to low-coverage domains.
  • The adjacent-beam pairwise loss is a generic way to distill a teacher's ranking among hypotheses that are close in the beam, with cost independent of beam width, and could be tested on structured generation tasks where beam search is the inference bottleneck.
  • Because the method does not couple to tokenizer specifics, combining it with student-side quantization or pruning should compound the speedup; the losses operate on distributions over SID tokens and cumulative scores, both of which survive most compression schemes.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes SmartGR, a knowledge-distillation framework for generative recommendation (GR) that transfers knowledge from an OneRec-8B teacher to an OneRec-1.7B student without changing the student's architecture. Two auxiliary losses are added to hard supervision: Hierarchy-Aware SID Distillation, which applies learnable depth-monotone weights to a KL divergence at each SID level, and Beam-Aware Ranking Distillation, which matches teacher and student softmax preferences between the cumulative prefix scores of a selected positive beam and the next lower-ranked teacher beam. The authors report that SmartGR improves all 16 metrics over the original student by 2.3%–17.5% (8.6% average), outperforms the twelve considered KD baselines on 15 of 16 metrics, preserves the student's inference speed (2.39x average speedup over the teacher), and they provide ablations, hyperparameter sweeps, and a complexity analysis.

Significance. If the reported gains are robust, SmartGR makes a practical and conceptual contribution: it identifies two GR-specific distillation challenges, gives compact loss formulations for both, and demonstrates large efficiency-quality gains on four datasets without architectural changes. The paper's strengths include precise equations for both losses, component ablations on Amazon and Kuaishou (Tables 5 and 11), negative-beam comparisons (Table 9), weighting-scheme comparisons (Table 8), a teacher-cache complexity analysis (Section B.6), and an unusually frank reporting of coverage statistics (Table 12). Evaluation is against held-out metrics with external baselines, so there is no obvious circularity. However, the central mechanism of the beam-aware loss is not fully supported, and some reported margins over the best baseline are small; these concerns are detailed below.

major comments (3)
  1. [Section 3.2 (Eqs. 7-10); Appendix B.3] The hard negative k−=k∗+1 is selected by final teacher beam score, but beam-search pruning at level ℓ is determined by the top-K cumulative prefix scores at that level (Eq. 2), and these two orderings need not coincide. A prefix that is competitive at an intermediate level can be absent from the final cached beams, and because the cache stores Kbeam=16 final beams while Kuaishou inference uses beam width 32, the actual pruning boundary for a target-compatible prefix can be outside the cache. As a result, LBEAM in Eq. (10) does not directly train the student to keep the positive prefix above the level-wise pruning boundary invoked in Challenge 2. The ablation in Appendix B.3 compares only final-rank-based negatives (random, first, last, next-lower), so it cannot validate the pruning-avoidance mechanism. Please either derive negatives from the level-wise pruning boundary (e.g., cache the intermediate beam sets or use the K-th best prefix score at each level), provide evidence that final-rank-adjacent beams approximate the boundary for the selected positives, or reframe BEAM loss as generic pairwise ranking distillation.
  2. [Appendix C.1 (Table 12); Table 11] Under the selected min_lcp thresholds, fewer than 5% of Kuaishou examples are eligible for any prefix-based distillation, yet Table 11 shows large gains from the full method over the base student (Ad Pass@32 21.24 to 23.15; Video Recall@32 2.74 to 3.22). The paper's statement that effectiveness depends more on reliability than on coverage is plausible but unsupported: the gains could be driven by a small non-representative subset, or by an interaction with hard supervision over the four training epochs. Please report results separately for eligible versus non-eligible examples, and analyze how the learned hierarchy weights in Eq. (4) are estimated stably from this sparse coverage, especially at the deepest SID levels that Table 2 uses to motivate the approach.
  3. [Section 4.2 (Table 4)] The improvements over the best KD baseline are below one percent on several metrics (Beauty N@5 +0.3%, Ad P@16 +0.1%, Ad P@32 +0.4%, Ad R@16 +0.6%) and are negative on Video P@32 (−0.4%). All numbers come from a single run, with no confidence intervals or significance tests. Because the paper claims superiority over prior distillation methods, please add variance estimates or multiple-seed results, or moderate the claim to a consistent trend rather than a per-metric win on 15 of 16 metrics.
minor comments (5)
  1. [Table 4] In the SeqKD row, the Video P@16 and P@32 entries appear as '15.3718.72' with a missing space; please fix the typesetting.
  2. [Section 4.2] The Impr.b value for Video P@32 is −0.4%, meaning SmartGR does not beat the best baseline on that metric; this is not mentioned in the text and should be acknowledged.
  3. [Appendix A.2] FSQ (finite scalar quantization) is used without definition; define the term at first mention.
  4. [Figure 1(b)] The y-axis label reads 'Mean prefix rank' but the text describes the teacher and student as ranking the global Top-1 item; please clarify the aggregation and the definition of the Top-1 item.
  5. [Section 4.4] Table 6 reports training time for cache construction and training, but the hardware and software configuration for Table 6 is not stated; adding it would aid reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: SmartGR's gains are measured against independent held-out metrics, and its distillation losses are not fitted to those metrics.

full rationale

The paper's central claims are empirical: SmartGR is trained with two distillation losses and evaluated on held-out Recall, NDCG, and Pass metrics against the original student, the teacher, and external baselines. Nothing in the derivation chain defines the predicted metric in terms of the loss inputs. The teacher beam cache and logits are distillation targets by definition, but the reported 8.6% improvement is an independent measurement of the student's autoregressive outputs, not a restatement of the cache. The hierarchy-aware weights are learned jointly with the student, with checkpoints selected on validation hard loss, and the hyperparameters lambda_SID, lambda_BEAM, and min_lcp are swept on validation rather than fitted to test metrics. The paper contains self-citations to prior work by overlapping authors, such as OneRec, DLLM2Rec, and beam-search-aware optimization, but these are used as context or motivation and are not invoked as an unverified uniqueness theorem or as the load-bearing justification for the proposed losses. The choice of the next lower-ranked beam as the hard negative is a debatable modeling decision because beam pruning is decided by per-level cumulative prefix scores rather than final rank, but that is a correctness or specification risk, not circularity: the B.3 ablation compares alternative negative-selection strategies, and the final evaluation metrics remain independent of the choice. No equation reduces a predicted quantity to an input by construction, and no fitted parameter is renamed as a prediction. The evaluation is self-contained against external benchmarks, so the appropriate finding is no significant circularity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim depends on the teacher cache being a reliable source of supervision, on the specific hyperparameter choices, and on the structural prior that SID depth ordering should be monotone. The paper gives reasonable empirical support for these premises with ablations and sweeps, so the ledger is moderate and does not include invented entities.

free parameters (4)
  • theta_raw = Learned: theta = 0.998 (Amazon) and 0.992 (Kuaishou)
    A scalar controlling the monotone tanh mapping from SID depth to level weight, learned jointly with the student. It is a fitted parameter, though its value is mild (close to 1). Reported in Table 13.
  • lambda_SID = 0.6
    Hyperparameter for the SID loss weight, tuned on validation set. Reported sweeps show stability.
  • lambda_BEAM = 0.3
    Hyperparameter for the BEAM loss weight, tuned on validation set. Reported sweeps show stability.
  • min_lcp = 2 (Amazon), 3 (Kuaishou)
    Minimum shared-prefix length threshold that determines whether a sample is eligible for prefix-based distillation, tuned per dataset.
assumptions (3)
  • domain assumption The teacher's cached beam scores and rankings are reliable supervision signals for the student.
    Both losses are built from the OneRec-8B beam cache, including the selected positive beam and the pair of cumulative prefix scores. The paper's Table 12 shows that teacher-target prefix agreement is low on Kuaishou, so reliability of the teacher signal is a substantive premise.
  • domain assumption The longest-prefix teacher beam is the right distillation target and distilling all positions on that beam is harmful.
    This is an empirical modeling assumption validated by ablations (Tables 5 and 11), not a mathematical result. The paper treats it as a design principle.
  • domain assumption The two-stage SID pipeline (tokenizer plus autoregressive generator) with beam search is the right abstraction for the problem.
    The paper builds on the OneRec/OpenOneRec framework and does not question the tokenization or decoding assumptions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SmartGR: Hierarchy and Beam-Aware Knowledge Distillation for Generative Recommendation." pith.science (2026). https://pith.science/paper/EXU7YC4E

@misc{pith2026260802048,
  author       = {Pith},
  title        = {Pith review of: SmartGR: Hierarchy and Beam-Aware Knowledge Distillation for Generative Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EXU7YC4E}},
  note         = {Machine review of arXiv:2608.02048}
}
abstract

Generative recommendation (GR) has emerged as a promising paradigm for recommender systems. Scaling up GR models can improve recommendation performance, but it also substantially increases inference cost. Knowledge distillation provides a practical solution by transferring knowledge from a large GR model to a lightweight one. However, existing distillation methods do not account for two GR-specific challenges: imbalanced distillation difficulty across the semantic ID (SID) hierarchy and incorrect prefix pruning during beam search. To address these challenges, we propose SmartGR, a novel distillation framework that utilizes Hierarchy-Aware SID Distillation to transfer the teacher's modeling capability across the hierarchy and leverages Beam-Aware Ranking Distillation to distill the teacher's ranking preferences during beam search. Extensive experiments on four benchmark datasets demonstrate the effectiveness and efficiency of SmartGR, improving the performance by 8.6% while achieving a 2.39$\times$ inference speedup on average.

Figures

Figures reproduced from arXiv: 2608.02048 by the authors.

Figure 1
Figure 1. (a). We therefore compare the prefix-ranking capa￾bilities of the teacher and student [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the SmartGR framework. 2 Preliminary 2.1 Generative Recommendation This paper focuses on generative recommendation (GR), which formulates next-item recommendation as an autore￾gressive generation problem in a tokenized item space, rather than directly scoring candidate items (Rajput et al. 2023; Wang et al. 2024; Ju et al. 2025). Let I denote the item catalog. For a user u, let Hu = (i1, i2, . . . , in) … view at source ↗
Figure 3
Figure 3. Teacher gains over the original student and the [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Hyperparameter sensitivity on Beauty, Toys, Ad, and Video. Each row varies one hyperparameter, and the columns [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 47 canonical work pages

  1. [1]

    International Conference on Learning Representations , year =

    FitNets: Hints for Thin Deep Nets , author =. International Conference on Learning Representations , year =. 1412.6550 , archivePrefix =

    arXiv

  2. [2]

    International Conference on Learning Representations , year =

    Session-Based Recommendations with Recurrent Neural Networks , author =. International Conference on Learning Representations , year =

    work page

  3. [3]

    2018 IEEE International Conference on Data Mining , pages =

    Self-Attentive Sequential Recommendation , author =. 2018 IEEE International Conference on Data Mining , pages =. 2018 , doi =

    work page 2018

  4. [4]

    2019 , doi =

    Sun, Fei and Liu, Jun and Wu, Jian and Pei, Changhua and Lin, Xiao and Ou, Wenwu and Jiang, Peng , booktitle =. 2019 , doi =

    work page 2019

  5. [5]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

    Autoregressive Image Generation Using Residual Quantization , author =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

    work page

  6. [6]

    Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages =

    Topology Distillation for Recommender System , author =. Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages =. 2021 , doi =

    work page 2021

  7. [7]

    Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =

    Ranking Distillation: Learning Compact Ranking Models With High Performance for Recommender System , author =. Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =. 2018 , doi =

    work page 2018

  8. [8]

    Collaborative Distillation for Top-

    Lee, Jae-woong and Choi, Minjin and Lee, Jongwuk and Shim, Hyunjung , booktitle =. Collaborative Distillation for Top-. 2019 , doi =

    work page 2019

Show all 58 references
  1. [9]

    Kang, SeongKu and Hwang, Junyoung and Kweon, Wonbin and Yu, Hwanjo , booktitle =

  2. [10]

    Forouzandeh, Saman and Moradi, Parham and Jalili, Mahdi , booktitle =

  3. [11]

    Proceedings of the 18th ACM Conference on Recommender Systems , year =

    Distillation Matters: Empowering Sequential Recommenders to Match the Performance of Large Language Models , author =. Proceedings of the 18th ACM Conference on Recommender Systems , year =

  4. [12]

    2026 , doi =

    Yu, Yuanqing and Wang, Yifan and Ma, Weizhi and Guo, Zhiqiang and Zhang, Min , booktitle =. 2026 , doi =

    2026

  5. [13]

    2026 , doi =

    Yang, Weiqin and Wang, Bohao and Xu, Zhenxiang and Chen, Jiawei and Zhang, Shengjia and Chen, Jingbang and Jin, Canghong and Wang, Can , journal =. 2026 , doi =

    2026

  6. [14]

    Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages =

    Exploring Feature-based Knowledge Distillation for Recommender System: A Frequency Perspective , author =. Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages =. 2025 , doi =

    2025

  7. [15]

    IEEE Transactions on Knowledge and Data Engineering , volume =

    Preference-Consistent Knowledge Distillation for Recommender System , author =. IEEE Transactions on Knowledge and Data Engineering , volume =. 2025 , doi =

    2025

  8. [16]

    International Conference on Learning Representations , year =

    Rejuvenating Cross-Entropy Loss in Knowledge Distillation for Recommender Systems , author =. International Conference on Learning Representations , year =. 2509.20989 , archivePrefix =

  9. [17]

    Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing , pages =

    Sequence-Level Knowledge Distillation , author =. Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing , pages =. 2016 , doi =

    2016

  10. [18]

    Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages =

    f -Divergence Minimization for Sequence-Level Knowledge Distillation , author =. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages =. 2023 , doi =

    2023

  11. [19]

    arXiv preprint arXiv:1503.02531 , year =

    Distilling the Knowledge in a Neural Network , author =. arXiv preprint arXiv:1503.02531 , year =

    arXiv

  12. [20]

    2024 , eprint =

    Gu, Yuxian and Dong, Li and Wei, Furu and Huang, Minlie , booktitle =. 2024 , eprint =

    2024

  13. [21]

    International Conference on Learning Representations , year =

    On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes , author =. International Conference on Learning Representations , year =. 2306.13649 , archivePrefix =

    arXiv

  14. [22]

    Hybrid Policy Distillation for

    Zhu, Wenhong and Xie, Ruobing and Wang, Rui and Liu, Pengfei , booktitle =. Hybrid Policy Distillation for. 2026 , eprint =

    2026

  15. [23]

    The Bridge-Garden Dilemma in

    Wang, Guanghui and Kacuila, Kaiwen Lv and Yang, Zhiyong and Wang, Zitai and Wu, Jin-Wen and Huang, Longtao and Xu, Qianqian and Huang, Qingming , booktitle =. The Bridge-Garden Dilemma in. 2026 , eprint =

    2026

  16. [24]

    2025 , eprint =

    Ko, Jongwoo and Chen, Tianyi and Kim, Sungnyun and Ding, Tianyu and Liang, Luming and Zharkov, Ilya and Yun, Se-Young , booktitle =. 2025 , eprint =

    2025

  17. [25]

    International Conference on Learning Representations , year =

    Speculative Knowledge Distillation: Bridging the Teacher-Student Gap Through Interleaved Sampling , author =. International Conference on Learning Representations , year =

  18. [26]

    International Conference on Learning Representations , year =

    Explain in Your Own Words: Improving Reasoning via Token-Selective Dual Knowledge Distillation , author =. International Conference on Learning Representations , year =

  19. [27]

    2025 , doi =

    Chen, Xiao and Ma, Changyi and Fan, Wenqi and Zhang, Zhaoxiang and Qing, Li , booktitle =. 2025 , doi =

    2025

  20. [28]

    Findings of the Association for Computational Linguistics: ACL 2024 , year =

    Distillation Enhanced Generative Retrieval , author =. Findings of the Association for Computational Linguistics: ACL 2024 , year =. 2402.10769 , archivePrefix =

    2024 arXiv

  21. [29]

    Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Lightweight and Direct Document Relevance Optimization for Generative Information Retrieval , author =. Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2025 , doi =

    2025

  22. [30]

    International Conference on Learning Representations , year =

    Efficient Inference for Large Language Model-based Generative Recommendation , author =. International Conference on Learning Representations , year =. 2410.05165 , archivePrefix =

    arXiv

  23. [31]

    Guo, Zitian and Hou, Yupeng and Ju, Clark Mingxuan and Shah, Neil and McAuley, Julian , journal =

  24. [32]

    Prompt Distillation for Efficient

    Li, Lei and Zhang, Yongfeng and Chen, Li , booktitle =. Prompt Distillation for Efficient. 2023 , doi =

    2023

  25. [33]

    Ramos, Jerome and Wu, Bin and Lipani, Aldo , journal =

  26. [34]

    Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Generative Recommender with End-to-End Learnable Item Tokenization , author =. Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2025 , doi =

    2025

  27. [35]

    Zhou, Guorui and Bao, Honghui and Huang, Jiaming and others , journal =

  28. [36]

    2024 IEEE 40th International Conference on Data Engineering , pages =

    Adapting Large Language Models by Integrating Collaborative Semantics for Recommendation , author =. 2024 IEEE 40th International Conference on Data Engineering , pages =

    2024

  29. [37]

    Unleashing the Native Recommendation Potential:

    Zhang, Zhiyang and She, Junda and Cai, Kuo and Chen, Bo and Wang, Shiyao and Luo, Xinchen and Luo, Qiang and Tang, Ruiming and Li, Han and Gai, Kun and Zhou, Guorui , journal =. Unleashing the Native Recommendation Potential:

  30. [38]

    Proceedings of the 25th International Conference on World Wide Web , pages =

    Ups and Downs: Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering , author =. Proceedings of the 25th International Conference on World Wide Web , pages =

  31. [39]

    Advances in Neural Information Processing Systems , volume =

    Recommender Systems with Generative Retrieval , author =. Advances in Neural Information Processing Systems , volume =

  32. [40]

    Generative Recommendation with Semantic

    Ju, Clark Mingxuan and Collins, Liam and Neves, Leonardo and Kumar, Bhuvesh and Wang, Louis Yufeng and Zhao, Tong and Shah, Neil , journal =. Generative Recommendation with Semantic

  33. [41]

    and Pasumarthi, Rama Kumar and Menon, Aditya Krishna and Rawat, Ankit Singh and Yu, Felix and Kim, Seungyeon and Veit, Andreas and Kumar, Sanjiv , booktitle =

    Reddi, Sashank J. and Pasumarthi, Rama Kumar and Menon, Aditya Krishna and Rawat, Ankit Singh and Yu, Felix and Kim, Seungyeon and Veit, Andreas and Kumar, Sanjiv , booktitle =

  34. [42]

    Recommendation as Language Processing (

    Geng, Shijie and Liu, Shuchang and Fu, Zuohui and Ge, Yingqiang and Zhang, Yongfeng , booktitle =. Recommendation as Language Processing (. 2022 , doi =

    2022

  35. [43]

    and Yi, Xinyang , booktitle =

    Singh, Anima and Vu, Trung and Mehta, Nikhil and Keshavan, Raghunandan and Sathiamoorthy, Maheswaran and Zheng, Yilin and Hong, Lichan and Heldt, Lukasz and Wei, Li and Tandon, Devansh and Chi, Ed H. and Yi, Xinyang , booktitle =. Better Generalization with Semantic. 2024 , doi =

    2024

  36. [44]

    Proceedings of the 33rd ACM International Conference on Information and Knowledge Management , pages =

    Learnable Item Tokenization for Generative Recommendation , author =. Proceedings of the 33rd ACM International Conference on Information and Knowledge Management , pages =. 2024 , doi =

    2024

  37. [45]

    and McAuley, Julian and Cheng, Derek Zhiyuan , journal =

    Hou, Yupeng and Ni, Jianmo and He, Zhankui and Sachdeva, Noveen and Kang, Wang-Cheng and Chi, Ed H. and McAuley, Julian and Cheng, Derek Zhiyuan , journal =

  38. [46]

    arXiv preprint arXiv:2504.04405 , year =

    Universal Item Tokenization for Transferable Generative Recommendation , author =. arXiv preprint arXiv:2504.04405 , year =

    arXiv

  39. [47]

    and Ren, Zhaochun , journal =

    Fu, Junchen and Ge, Xuri and Karatzoglou, Alexandros and Arapakis, Ioannis and Verberne, Suzan and Jose, Joemon M. and Ren, Zhaochun , journal =. Differentiable Semantic

  40. [48]

    arXiv preprint arXiv:2402.17152 , year =

    Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations , author =. arXiv preprint arXiv:2402.17152 , year =

    arXiv

  41. [49]

    Han, Ruidong and Yin, Bin and Chen, Shangyu and Jiang, He and Jiang, Fei and Li, Xiang and Ma, Chi and Huang, Mincong and Li, Xiaoguang and Jing, Chunzhen and Han, Yueming and Zhou, Menglei and Yu, Lei and Liu, Chuan and Lin, Wei , journal =

  42. [50]

    Kong, Xiaoyu and Sheng, Leheng and Tan, Junfei and Chen, Yuxin and Wu, Jiancan and Zhang, An and Wang, Xiang and He, Xiangnan , journal =

  43. [51]

    arXiv preprint arXiv:2602.21677 , year =

    Trie-Aware Transformers for Generative Recommendation , author =. arXiv preprint arXiv:2602.21677 , year =

  44. [52]

    Distillation from Heterogeneous Models for Top-

    Kang, SeongKu and Kweon, Wonbin and Lee, Dongha and Lian, Jianxun and Xie, Xing and Yu, Hwanjo , booktitle =. Distillation from Heterogeneous Models for Top-. 2023 , doi =

    2023

  45. [53]

    2025 , url =

    Xu, Wujiang and Wu, Qitian and Liang, Zujie and Han, Jiaojiao and Ning, Xuying and Shi, Yunxiao and Lin, Wenfang and Zhang, Yongfeng , booktitle =. 2025 , url =

    2025

  46. [54]

    arXiv preprint arXiv:2502.15685 , year =

    Active Large Language Model-based Knowledge Distillation for Session-based Recommendation , author =. arXiv preprint arXiv:2502.15685 , year =

    arXiv

  47. [55]

    Wang, Xinfeng and Cui, Jin and Suzuki, Yoshimi and Fukumoto, Fumiyo , journal =

  48. [56]

    Zhang, Haoyi and Sun, Guohao and Lu, Jinhu and Liu, Guanfeng and Fang, Xiu Susie , journal =

  49. [57]

    Shahgir, Haz Sameen and Li, Yufei and Wei, Xiaohan and Pu, Yunchen and Tian, Fei and Sun, Chonglin and Shyu, Frank and Pandey, Sandeep and Simon, Luke and Dong, Yue and Liu, Xi , journal =

  50. [58]

    2025 , doi =

    Xie, Jiawen and Wu, Haiyang and Ji, Deyi and Yang, Yuekui and Ma, Shaoping , booktitle =. 2025 , doi =

    2025

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.