Pith. sign in

REVIEW 4 major objections 6 minor 48 references

HCGRec: Hint-Conditioned Generative Recommendation with Semantic IDs

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read A one-token prefix hint, used only in training, restores reward signal for hard semantic-ID examples and cuts dead groups from 70% to 20%.

desk verdict A clean, domain-specific hinting scheme for unreachable GRPO groups in semantic-ID recommenders, but the empirical case is undermined by unquantified handling of unresolved instances. read the letter →

arxiv 2608.11980 v1 pith:FVAFHQT5 submitted 2026-08-12 cs.IR cs.AI

classification cs.IRcs.AI
keywords generativerecommendationsequentialsemanticIDsreward-basedpost-trainingGRPOreachabilitydiagnosisprefixhintscreditdecomposition
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

The paper argues that reward-based post-training of semantic-ID generative recommenders — systems that represent each item as a short sequence of discrete semantic tokens and generate the next item by autoregressive decoding — fails not mainly because rewards are sparse but because many training instances are unreachable under a finite rollout budget: when an early token selects the wrong branch of the item-token prefix tree, all sampled completions receive identical zero rewards and group-relative optimization has no signal. Hint-Conditioned Generative Recommendation (HCGRec) diagnoses each instance with the current checkpoint's own rollouts and, only for instances that fail unhinted generation, prepends the shortest target-ID prefix that makes the target reachable; the model then generates the remaining suffix, converting inactive zero-reward groups into informative comparisons over item-token completions. Because hinted prefix tokens are oracle-provided context while suffix tokens are sampled actions, the method splits credit: a supervised prefix-anchoring loss stabilizes the hinted branch, and GRPO optimizes only the sampled suffix. The reported experiments show that this training-only intervention improves ranking metrics over supervised fine-tuning and vanilla reward-based post-training, especially at deeper cutoffs, and reduces the zero-advantage training-sample fraction from over 70 percent to below 20 percent.

What carries the argument

The load-bearing mechanism is reachability-aware hint conditioning on the semantic-ID prefix tree: a diagnostic pass evaluates, for each instance and each hint length $h$, whether a rollout group conditioned on the target prefix $\mathbf{p}^*_h$ contains an exact match, and sets the hint to the shortest such length, $h^*(x,y^*)=\min C(x,y^*)$. This converts an unhinted generation problem that yields identical zero rewards into a suffix-generation problem on a reachable branch, so group-relative advantages become informative. The companion mechanism is hint-aware credit decomposition: the hinted prefix tokens receive the supervised anchoring loss $\mathcal{L}_{\text{prefix}}^{\text{SFT}}$, since they are oracle-provided item context, while the sampled suffix tokens receive GRPO advantages computed from group rewards on the completed IDs; the two are combined in $\mathcal{L}_{\text{HCGRec}}$ with a calibration weight $\lambda$.

What would settle it

Run the offline diagnostic pass and report the fraction of training instances with $C(x,y^*)=\emptyset$; if that fraction is substantial and a variant of HCGRec that excludes or defaults those instances shows no ranking gain over the unhinted baseline on genuinely unreachable instances, the reachability-recovery explanation would be refuted.

Watch

Extended reading notes

Core claim

The central claim is that finite-rollout unreachability in the item-token prefix tree is the main bottleneck in semantic-ID reward-based post-training, and that supplying a minimal target-prefix hint at training time restores reachability without changing inference. For each training instance, HCGRec runs an offline diagnostic pass with the SFT checkpoint: it tests hint depths from 0 up to $H_{\max}$ and accepts the shortest length $h^*(x,y^*)$ for which at least one diagnostic rollout conditioned on that prefix exactly recovers the target Semantic ID. Post-training then proceeds under the hinted branch: the model samples only the suffix, rewards are computed on the completed identifiers, and the objective $\mathcal{L}_{\text{HCGRec}} = \mathcal{L}_{\text{suffix}}^{\text{GRPO}} + \lambda \mathcal{L}_{\text{prefix}}^{\text{SFT}}$ assigns group-relative policy credit to sampled suffix tokens and supervised semantic-anchoring credit to the oracle-provided prefix. The paper reports that this recovers learning signal for hard instances, improves ranking metrics — with the clearest gains at deeper cutoffs — and lowers the fraction of zero-advantage training groups from over 70 percent to below 20 percent.

Load-bearing premise

The method's gains rest on the assumption that nearly every hard training instance has a target-prefix hint within the diagnostic budget that makes the item reachable, and that instances without such a hint can be dropped or given a default hint without biasing the training distribution — yet the paper never reports how many instances are unresolved.

Editorial extensions

If this is right

  • Training with minimal target-prefix hints cuts the fraction of zero-advantage rollout groups from over 70 percent to below 20 percent, so the same reward function yields far more usable group-relative learning signal.
  • Because the hint is chosen once by the SFT checkpoint and fixed during post-training, the offline minimal-hint policy is more stable and cheaper than recomputing hint depth online, and it avoids the objective drift of dynamic hinting.
  • Supervised credit on the hinted prefix only — not the full sequence — gives the strongest results, implying that once prefix alignment is restored, suffix-level policy optimization should be left to GRPO rather than diluted by full-sequence supervised loss.
  • The prefix-anchoring weight $\lambda$ behaves as a calibration knob: moderate values around 0.001–0.01 are best, while $\lambda=0.1$ consistently harms quality, so the prefix/suffix division of labor must be preserved.
  • Inference never receives the hint, so HCGRec changes only the training procedure and can be applied to an existing semantic-ID recommender without altering the deployment interface.

Reading between the lines

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

  • An untested implication is that instances with no reachable hint ($C(x,y^*)=\emptyset$) are censored or given the default depth-3 hint, and the paper does not report their fraction; if that fraction is large, part of the measured gain could come from excluding the hardest examples rather than from hinting, so reporting the unresolved ratio is the natural next step.
  • The same diagnosis-and-hint recipe should transfer to other tree-structured generative retrieval settings, such as hierarchical document-ID search, where an early-token error also collapses finite-rollout reward signal; a testable extension would apply HCGRec's diagnostic pass to such tasks.
  • Because the hint is selected by the SFT checkpoint, the method inherits the checkpoint's blind spots: instances the SFT stage never learned may be systematically hinted, potentially biasing the policy toward already-reachable branches; measuring per-instance hint depth by user or item subgroup would reveal such bias.
  • The headline reduction in zero-advantage groups depends on the diagnostic budget (beam size 16, $H_{\max}=3$), so varying that budget and re-measuring both the hint-length distribution and final ranking would test the sensitivity of the central result.
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

4 major / 6 minor

Summary. The manuscript proposes HCGRec, a training-time method for semantic-ID generative recommenders that addresses finite-rollout unreachability in reward-based post-training. Before GRPO, an offline diagnostic pass rolls out from an SFT checkpoint with target-prefix hints of increasing length and selects the shortest prefix that makes the ground-truth item reachable within a fixed budget. During post-training, HCGRec supplies that oracle prefix as context, applies GRPO only to the sampled suffix, and adds a supervised prefix-anchoring loss. Experiments on three Amazon datasets compare HCGRec with sequential recommenders, supervised generative baselines, and reward-based post-training variants, and report improvements on several ranking cutoffs together with a large reduction in zero-variance rollout groups.

Significance. If the empirical claims are supported, the paper makes a useful contribution: it identifies a concrete failure mode of GRPO over multi-token Semantic IDs, proposes a simple and interpretable training-time intervention, and introduces a token-source-aware credit decomposition that is well motivated by the distinction between oracle-provided context and sampled actions. The release of code is a practical strength. However, the central empirical claim currently rests on unresolved-instance handling that is unspecified and possibly inconsistent, on comparisons without uncertainty quantification, and on an interpretation of the zero-gradient reduction that is confounded by direct supervision over oracle target prefixes. These issues are load-bearing because the paper's causal story is that reachability-aware hinting, rather than target leakage or data censoring, produces the gains.

major comments (4)
  1. [Section 4.1.5 and Algorithm 1] The treatment of unresolved instances is contradictory and unreported. Algorithm 1 (lines 9-11, 14) marks instances with C(x,y*) = empty as unresolved and trains only on resolved instances, whereas Section 4.1.5 states that the diagnostic pass uses a 'default unsolved depth 3'. With M=4 and H_max=3, these two statements are mutually exclusive: an instance that fails the B_h=1 test at h=3 cannot legally receive h=3 under Eq. (12). The paper never reports (i) the fraction of unresolved instances per dataset and task, (ii) whether unresolved instances enter the post-training pool or are dropped, or (iii) their held-out and zero-gradient statistics. If unresolved instances are dropped, the training set is censored toward easier examples, and the reported reduction in zero-advantage groups could be caused by deletion rather than by recovered learning signal. This is a load-bearing issue for the abstract's central claim, and it must be resolved by specifying and measuring the unresolved treatment.
  2. [Section 4.2, Table 2] The main empirical comparisons are reported without error bars, multiple seeds, or significance tests, and several headline differences are very small. On Instruments, HR@10 improves from 0.1179 (GRPO Rule-only) to 0.1180 (HCGRec), while NDCG@10 decreases from 0.0960 to 0.0945; on Arts, the NDCG@10 gain is 0.0956 vs. 0.0952. Without uncertainty quantification, the claim that HCGRec 'substantially improves' over vanilla reward-based post-training is not established for these cutoffs. In addition, the abstract's 'over 70%' zero-advantage reduction is not directly visible in Figure 2, whose end-of-training values are about 0.55 and 0.63 on Arts and Instruments; please state whether the 70% figure refers to a different reporting point and provide the corresponding curves and confidence intervals.
  3. [Section 3.4, Eq. (16)-(17), and Figure 2] The interpretation that hinting 'recovers learning signal' is confounded by direct supervision over oracle target prefixes. The objective in Eq. (17) includes L_SFT_prefix, which teacher-forces the model to produce the target prefix tokens from the original context. A reduction in zero-variance rollout groups is therefore expected even if no additional reward signal is recovered, because the model is explicitly trained to start completions in the correct branch. The offline-hint variant in Section 4.2 separates credit decomposition from hinting, but it does not separate the effect of exposing the correct target prefix from the effect of reachability-corrected reward optimization. A control that removes oracle target information while keeping the hint-length distribution, such as random prefixes of the same length or prefixes selected without consulting y*, is needed to support the causal claim.
  4. [Section 3.3, Eq. (10)-(12)] The diagnostic guarantee is checkpoint-specific and finite-budget, but the central claim concerns learning during post-training when the policy is changing. Eq. (10) establishes that at least one target hit exists for π_sft under budget G_d, while Eq. (13) samples from π_θold with a possibly different group size G. The paper does not report how often hinted prompts actually produce a target hit under the evolving policy, nor whether G equals G_d. Without this information, the suffix GRPO loss in Eq. (15) may still be operating on groups with all-zero rewards for many 'recovered' instances, and the zero-gradient ratio in Figure 2 could improve for reasons unrelated to reachability. Please report the target-hit rate of hinted groups during training alongside the zero-gradient ratio.
minor comments (6)
  1. [Section 4.1.5] The phrase 'default unsolved depth 3' is undefined; if it is a code-level parameter, specify its exact role and reconcile it with Algorithm 1's resolved-only training loop.
  2. [Section 4.2] Table 2 reports single checkpoints selected by NDCG@10. State explicitly whether each number is a single run and whether checkpoint selection is made on the validation set as claimed, because this affects the interpretation of the reported gains.
  3. [Figure 2] The smoothing applied to the zero-gradient ratio curves is not described. Specify the smoothing method, the number of logged points, and whether the curves are from a single training run or averaged over seeds.
  4. [Section 1 and Conclusion] The abstract and conclusion describe the results as 'substantially improves' over supervised fine-tuning and vanilla reward-based post-training, while Section 4.2 itself notes that HCGRec is not uniformly best and leaves several cutoffs to baselines. Please calibrate the strength of the claims to the actual table.
  5. [Section 4.3, Figure 4] The comparison between offline minimal hinting and dynamic hinting does not state the computation budget of the dynamic variant. Since the paper argues that offline hinting is cheaper, please report the additional diagnostic and training cost of dynamic hinting.
  6. [Affiliations] The affiliation 'Mei Tuan' appears to be a typo for 'Meituan'; please correct it.

Circularity Check

1 steps flagged · score 6.0 of 10

The headline zero-advantage reduction is largely by construction: h* is chosen so that B_h=1, which is the same condition as a non-zero-reward-variance group under exact-match rewards.

  1. fitted input called prediction [Section 3.3, Eqs. (10)-(12); Section 4.3, Figure 2; Eq. (18)]
    "We say that length h passes the reachability test if at least one diagnostic completion exactly recovers the target identifier: B_h(x,y*) = I[∃ j ∈ {1,...,G_d} s.t. \tilde{y}^{(h)}_j = y*]. ... The hint used by HCGRec is the shortest prefix in this set: h*(x,y*) = min C(x,y*). ... The guarantee is therefore finite-budget and checkpoint-specific: the selected prompt has already produced at least one target hit under π_sft and the diagnostic budget."

    With the exact-target-match reward of Eq. (18), B_h=1 means at least one diagnostic completion equals y*, so the diagnostic reward set contains a 1 and has nonzero variance; hence the instance is not a zero-advantage group under Eq. (7) with τ=0. The abstract's headline metric ('reducing zero-advantage training samples from over 70% to below 20%') is exactly the frequency of zero-variance groups, and Eq. (12) selects the shortest prefix that makes that variance nonzero by construction. Reporting the drop in Figure 2 as evidence that 'prefix hints are not a superficial input modification' presents a selection-rule consequence as an emergent training effect.

full rationale

The one demonstrable circularity is the zero-advantage reduction claim. In Section 3.3, B_h is defined by the existence of a diagnostic completion exactly equal to y*, and h* is chosen as min C(x,y*). Under the exact-match reward of Eq. (18), B_h=1 is equivalent to nonzero reward variance, i.e., the negation of the zero-advantage/unreachable condition formalized in Eq. (7) with τ=0. Thus every accepted hint is, by selection, an instance moved out of the zero-advantage regime at the SFT checkpoint. The prefix-anchoring loss of Eq. (16) further supervises oracle target tokens, so part of the 'recovered learning signal' is teacher-forced label tokens by design. I do not raise the score to 8-10 because the held-out ranking results in Table 2 remain genuine, non-circular evidence: inference uses no hint, all RL variants start from the same SFT checkpoint, and HCGRec does not uniformly dominate, which is inconsistent with a fully forced result. The unresolved-instance handling is a serious rigor concern rather than circularity: Algorithm 1 trains only resolved instances, while Section 4.1.5 mentions a 'default unsolved depth 3' that conflicts with the formal B_h=1 acceptance rule, and unresolved fractions are never reported; this could bias gains through training-set censoring, but it is a validity issue, not a derivation that reduces to its own inputs. No load-bearing self-citation or imported uniqueness theorem appears, so the score reflects the one structural reduction in the paper's own diagnostic metric.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on three assumptions beyond standard GRPO: the semantic-ID prefix tree is semantically meaningful, exact-match reward is a sufficient training signal, and the finite diagnostic budget correctly classifies reachable instances. The method-specific free parameters, lambda, H_max, G_d, and default unsolved depth, are chosen by hand and not derived. No new entities are postulated, so the method is a training-time mechanism rather than a new physical or mathematical object.

free parameters (5)
  • prefix anchoring weight lambda = 0.005 default, ablated over 0, 0.001, 0.005, 0.01, 0.1
    Controls the weight of the supervised prefix-anchoring loss in Eq (17). Chosen by hand and validated in RQ4, not derived.
  • maximum hint depth H_max = 3 (M-1 for 4-token IDs)
    Limits the hint search space in Eq (11). Chosen by hand; the paper assumes the first three tokens are sufficient for reachability.
  • diagnostic rollout budget G_d = beam size 16
    Number of diagnostic completions per instance in Eq (9)-(10). Chosen by hand; affects whether an instance is classified as reachable.
  • default unsolved depth = 3
    Mentioned in Section 4.1.5 for instances where no hint passes the reachability test. Its role is ambiguous: either those instances are excluded from training or trained with h=3. This is a free modeling choice that affects the training distribution.
  • zero-variance threshold tau = 0 (for exact-match reward)
    Threshold in the unreachable definition of Eq (7). Set to 0 when rewards are binary, so it is a chosen hyperparameter.
assumptions (4)
  • standard math The GRPO group-relative objective in Eq (3)-(6) is a valid optimizer for reward-based post-training.
    The paper builds on DeepSeek-Math's GRPO as an established baseline without re-deriving it.
  • domain assumption Items are represented as fixed-length four-token semantic IDs organized as a coarse-to-fine prefix tree.
    The method's premise is that early semantic tokens define a meaningful coarse branch and that revealing the true prefix relocates generation to the correct region. Invoked throughout Section 3.
  • domain assumption Exact target match is a sufficient reward signal for post-training.
    The paper uses only the exact-match reward of Eq (18) to drive training; it assumes binary exact match is a reasonable proxy for recommendation quality. No reward shaping or user feedback is used.
  • ad hoc to paper The diagnostic pass with beam size 16 and H_max=3 correctly classifies reachable instances.
    Eq (7)-(12) define reachability with respect to a finite diagnostic budget. The method assumes this budget is enough to cover the useful hard instances and that the treatment of unresolved instances does not bias training. The paper does not report the unresolved fraction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HCGRec: Hint-Conditioned Generative Recommendation with Semantic IDs." pith.science (2026). https://pith.science/paper/FVAFHQT5

@misc{pith2026260811980,
  author       = {Pith},
  title        = {Pith review of: HCGRec: Hint-Conditioned Generative Recommendation with Semantic IDs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FVAFHQT5}},
  note         = {Machine review of arXiv:2608.11980}
}
read the original abstract

Semantic-ID generative recommenders represent each item as a short sequence of discrete semantic tokens and predict the next item by autoregressively generating this token sequence. This paradigm enables a unified generation interface for item IDs, histories, and item text, but it also creates a structured optimization bottleneck during reward-based post-training: when an early semantic token enters the wrong branch of the item-token space, finite rollout groups rarely reach the ground-truth item, so group-relative optimization receives identical zero rewards and produces no useful advantage. We propose Hint-Conditioned Generative Recommendation (HCGRec), a semantic-ID generative recommendation framework that recovers learning signal for such hard training instances. HCGRec diagnoses each instance with checkpoint rollouts and supplies a minimal target-prefix hint only when the current generator cannot reach the correct item. The model then generates the unhinted suffix under the hinted semantic branch, turning zero-reward groups into informative comparisons over item-token completions. Hinting also changes token identity: hinted prefix tokens are oracle-provided item context, while unhinted suffix tokens are sampled generation actions. We therefore introduce hint-aware credit decomposition, using supervised learning to preserve item-semantic and prefix-structure alignment for hinted tokens and GRPO to optimize the sampled suffix. Experiments on sequential recommendation benchmarks show that HCGRec substantially improves over supervised fine-tuning and vanilla reward-based post-training, while reducing zero-advantage training samples from over 70% to below 20%. The code is accessible at https://github.com/WncFht/GRec.

Figures

Figures reproduced from arXiv: 2608.11980 by the authors.

Figure 1
Figure 1. Overview of HCGRec. (A) In semantic-ID generative recommendation, unhinted rollouts may enter wrong prefix [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Fraction of rollout groups without gradient signal [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Task-wise fixed-hint depth distributions on Instru [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Task-scope trajectories on Instruments under the [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: RQ4 credit-weight comparison on Instruments and [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 8 canonical work pages

  1. [1]

    Keqin Bao, Jizhi Zhang, Yang Zhang, Wenjie Wang, Fuli Feng, and Xiangnan He

  2. [2]

    Jiaxin Deng, Shiyao Wang, Kuo Cai, Lejian Ren, Qigen Hu, Weifeng Ding, Qiang Luo, and Guorui Zhou. 2025. OneRec: Unifying Retrieve and Rank with Generative Recommender and Iterative Preference Alignment. arXiv:2502.18965 [cs.IR] https://arxiv.org/abs/2502.18965

  3. [3]

    Zezhong Fan, Ziheng Chen, Luyi Ma, Jin Huang, Lalitesh Morishetti, Kaushiki Nag, Sushant Kumar, and Kannan Achan. 2026. CRAB: Codebook Rebalancing for Bias Mitigation in Generative Recommendation. arXiv:2604.05113 [cs.IR] https://arxiv.org/abs/2604.05113

  4. [4]

    Dengzhao Fang, Jingtong Gao, Chengcheng Zhu, Yu Li, Xiangyu Zhao, and Yi Chang. 2025. HiD-VAE: Interpretable Generative Recommendation via Hi- erarchical and Disentangled Semantic IDs. arXiv:2508.04618 [cs.IR] https: //arxiv.org/abs/2508.04618

  5. [5]

    Jose, and Zhaochun Ren

    Junchen Fu, Xuri Ge, Alexandros Karatzoglou, Ioannis Arapakis, Suzan Verberne, Joemon M. Jose, and Zhaochun Ren. 2026. Differentiable Semantic ID for Genera- tive Recommendation. arXiv:2601.19711 [cs.IR] https://arxiv.org/abs/2601.19711

  6. [6]

    Gao, Chen Xue, Marc Versage, Xie Zhou, Zhongruo Wang, Chao Li, Yeon Seonwoo, Nan Chen, Zhen Ge, Gourab Kundu, Weiqi Zhang, Tian Wang, Qingjun Cui, and Trishul Chilimbi

    Vianne R. Gao, Chen Xue, Marc Versage, Xie Zhou, Zhongruo Wang, Chao Li, Yeon Seonwoo, Nan Chen, Zhen Ge, Gourab Kundu, Weiqi Zhang, Tian Wang, Qingjun Cui, and Trishul Chilimbi. 2025. SynerGen: Contextualized Generative Recommender for Unified Search and Recommendation. arXiv:2509.21777 [cs.CL] https://arxiv.org/abs/2509.21777

  7. [7]

    Shijie Geng, Shuchang Liu, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2023. Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5). arXiv:2203.13366 [cs.IR] https://arxiv.org/abs/ 2203.13366

  8. [8]

    Balázs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk

Show all 48 references
  1. [9]

    Jianchao Ji, Zelong Li, Shuyuan Xu, Wenyue Hua, Yingqiang Ge, Juntao Tan, and Yongfeng Zhang. 2023. GenRec: Large Language Model for Generative Recommendation. arXiv:2307.00457 [cs.IR] https://arxiv.org/abs/2307.00457

  2. [10]

    Jie Jiang, Yangru Huang, Zeyu Wang, Changping Wang, Yuling Xiong, Jun Zhang, and Huan Yu. 2026. Spend Search Where It Pays: Value-Guided Structured Sam- pling and Optimization for Generative Recommendation. arXiv:2602.10699 [cs.AI] https://arxiv.org/abs/2602.10699

  3. [11]

    Clark Mingxuan Ju, Liam Collins, Leonardo Neves, Bhuvesh Kumar, Louis Yufeng Wang, Tong Zhao, and Neil Shah. 2025. Generative Recommendation with Semantic IDs: A Practitioner’s Handbook. arXiv:2507.22224 [cs.IR] https://arxiv. org/abs/2507.22224

  4. [12]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-Attentive Sequential Recom- mendation. arXiv:1808.09781 [cs.IR] https://arxiv.org/abs/1808.09781

  5. [13]

    Kirill Khrylchenko. 2026. Variable-Length Semantic IDs for Recommender Sys- tems. arXiv:2602.16375 [cs.IR] https://arxiv.org/abs/2602.16375

  6. [14]

    Xiaoyu Kong, Leheng Sheng, Junfei Tan, Yuxin Chen, Jiancan Wu, An Zhang, Xiang Wang, and Xiangnan He. 2025. MiniOneRec: An Open-Source Framework for Scaling Generative Recommendation. arXiv:2510.24431 [cs.IR] https://arxiv. org/abs/2510.24431

  7. [15]

    Yifan Liu, Kangning Zhang, Xiangyuan Ren, Yanhua Huang, Jiarui Jin, Yingjie Qin, Ruilong Su, Ruiwen Xu, Yong Yu, and Weinan Zhang. 2024. AlignRec: Aligning and Training in Multimodal Recommendations. InProceedings of the 33rd ACM International Conference on Information and Kno...

  8. [16]

    Jianmo Ni, Jiacheng Li, and Julian McAuley. 2019. Justifying recommendations using distantly-labeled reviews and fine-grained aspects. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural...

  9. [17]

    Gustavo Penha, Edoardo D’Amico, Marco De Nadai, Enrico Palumbo, Alexandre Tamborrino, Ali Vardasbi, Max Lefarov, Shawn Lin, Timothy Heath, Francesco Fabbri, and Hugues Bouchard. 2025. Semantic IDs for Joint Generative Search and Recommendation. arXiv:2508.10478 [cs.IR] https:/...

  10. [18]

    Gustavo Penha, Ali Vardasbi, Enrico Palumbo, Marco de Nadai, and Hugues Bouchard. 2024. Bridging Search and Recommendation in Generative Retrieval: Does One Task Help the Other? arXiv:2410.16823 [cs.IR] https://arxiv.org/abs/ 2410.16823

  11. [19]

    Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan H. Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Maheswaran Sathiamoorthy. 2023. Recommender Systems with Generative Retrieval. arXiv:2305.05065 [cs.IR] htt...

  12. [20]

    Shuai Shao, Kangning Zhang, Qingyao Li, Shijian Wang, Hao Wang, Wenxiang Jiao, Yuan Lu, Yi Guo, Weiwen Liu, and Weinan Zhang. 2026. Harness-R1: Learning to Edit Executable Runtime Harnesses from Agent Failure Trajectories. arXiv:2608.02276 [cs.AI] doi:10.48550/arXiv.2608.02276

  13. [21]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y. K. Li, Y. Wu, and Daya Guo. 2024. DeepSeek- Math: Pushing the Limits of Mathematical Reasoning in Open Language Models. arXiv:2402.03300 [cs.CL] https://arxiv.org/abs/2402.03300

  14. [22]

    Zihua Si, Zhongxiang Sun, Jiale Chen, Guozhang Chen, Xiaoxue Zang, Kai Zheng, Yang Song, Xiao Zhang, Jun Xu, and Kun Gai. 2024. Generative Re- trieval with Semantic Tree-Structured Item Identifiers via Contrastive Learning. arXiv:2309.13375 [cs.IR] https://arxiv.org/abs/2309.13375

  15. [23]

    Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang

  16. [24]

    Junfei Tan, Yuxin Chen, An Zhang, Junguang Jiang, Bin Liu, Ziru Xu, Han Zhu, Jian Xu, Bo Zheng, and Xiang Wang. 2025. Reinforced Preference Optimization for Recommendation. arXiv:2510.12211 [cs.IR] https://arxiv.org/abs/2510.12211

  17. [25]

    Jiaxi Tang and Ke Wang. 2018. Personalized Top-N Sequential Recommendation via Convolutional Sequence Embedding. arXiv:1809.07426 [cs.IR] https://arxiv. org/abs/1809.07426

  18. [26]

    Yibiao Wei, Jie Zou, Pengfei Zhang, Xiao Ao, Weikang Guo, Zeyu Ma, and Yang Yang. 2026. CARD: Non-Uniform Quantization of Visual Semantic Unit for Generative Recommendation. arXiv:2604.26427 [cs.IR] https://arxiv.org/abs/ 2604.26427

  19. [27]

    Ming Xia, Zhiqin Zhou, Guoxin Ma, and Dongmin Huang. 2026. Un- leash the Potential of Long Semantic IDs for Generative Recommendation. arXiv:2602.13573 [cs.IR] https://arxiv.org/abs/2602.13573

  20. [28]

    Yu Xia, Canwen Xu, Zhewei Yao, Julian McAuley, and Yuxiong He. 2026. Learning to Hint for Reinforcement Learning. arXiv:2604.00698 [cs.LG] https://arxiv.org/ abs/2604.00698

  21. [29]

    Haibo Xing, Hao Deng, Lingyu Mu, Jinxin Hu, Yu Zhang, Xiaoyi Zeng, and Jing Zhang. 2026. Learning to Reflect and Correct: Towards Better Decoding Tra- jectories for Large-Scale Generative Recommendation. arXiv:2602.23639 [cs.IR] https://arxiv.org/abs/2602.23639

  22. [30]

    Liu Yang, Fabian Paischer, Kaveh Hassani, Jiacheng Li, Shuai Shao, Zhang Gabriel Li, Yun He, Xue Feng, Nima Noorshams, Sem Park, Bo Long, Robert D Nowak, Xiaoli Gao, and Hamid Eghbalzadeh. 2024. Unifying Generative and Dense Retrieval for Sequential Recommendation. arXiv:2411....

  23. [31]

    Kangning Zhang, Jiarui Jin, Yingjie Qin, Ruilong Su, Jianghao Lin, Yong Yu, and Weinan Zhang. 2024. MOTOR: Learning ID-free Item Representa- tion with Token Crossing for Embedding-based Multimodal Recommendation. arXiv:2410.19276 [cs.IR] doi:10.48550/arXiv.2410.19276 CIKM ’26,...

  24. [32]

    Kangning Zhang, Yixing Li, Shuai Shao, Qingyao Li, Zhengxi Lu, Zhiyuan Yao, Jianghao Lin, Wenxiang Jiao, Yuan Lu, Weiwen Liu, Weinan Zhang, and Yong Yu

  25. [33]

    Kangning Zhang, Weiwen Liu, Wenxiang Jiao, Kounianhua Du, Yuan Lu, Weinan Zhang, and Yong Yu. 2026. LoopTool: Closing the Data–Training Loop for Robust LLM Tool Calls. InProceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pap...

  26. [34]

    Kangning Zhang, Yingjie Qin, Weinan Zhang, Yong Yu, and Jianghao Lin. 2026. DiffCold: A Diffusion-based Generative Model for Cold-Start Item Recommenda- tion. arXiv:2606.12245 [cs.IR] doi:10.48550/arXiv.2606.12245

  27. [35]

    Kangning Zhang, Shuai Shao, Qingyao Li, Jianghao Lin, Lingyue Fu, Shi- jian Wang, Wenxiang Jiao, Yuan Lu, Weiwen Liu, Weinan Zhang, and Yong Yu. 2026. MMSkills: Towards Multimodal Skills for General Visual Agents. arXiv:2605.13527 [cs.AI] doi:10.48550/arXiv.2605.13527

  28. [36]

    Xichen Zhang, Sitong Wu, Yinghao Zhu, Haoru Tan, Shaozuo Yu, Ziyi He, and Jiaya Jia. 2026. Scaf-GRPO: Scaffolded Group Relative Policy Optimization for Enhancing LLM Reasoning. arXiv:2510.19807 [cs.CL] https://arxiv.org/abs/2510. 19807

  29. [37]

    Bowen Zheng, Yupeng Hou, Hongyu Lu, Yu Chen, Wayne Xin Zhao, Ming Chen, and Ji-Rong Wen. 2023. Adapting Large Language Models by Integrating Collaborative Semantics for Recommendation. arXiv:2311.09049 [cs.IR] https: //arxiv.org/abs/2311.09049

  30. [38]

    Bowen Zheng, Enze Liu, Zhongfu Chen, Zhongrui Ma, Yue Wang, Wayne Xin Zhao, and Ji-Rong Wen. 2025. Pre-training Generative Recommender with Multi- Identifier Item Tokenization. arXiv:2504.04400 [cs.IR] https://arxiv.org/abs/2504. 04400

  31. [39]

    Guorui Zhou, Jiaxin Deng, Jinghao Zhang, Kuo Cai, Lejian Ren, Qiang Luo, Qianqian Wang, Qigen Hu, Rui Huang, Shiyao Wang, Weifeng Ding, Wuchao Li, Xinchen Luo, Xingmei Wang, Zexuan Cheng, Zixing Zhang, Bin Zhang, Boxuan Wang, Chaoyi Ma, Chengru Song, Chenhui Wang, Di Wang, Don...

  32. [40]

    Guorui Zhou, Hengrui Hu, Hongtao Cheng, Huanjie Wang, Jiaxin Deng, Jinghao Zhang, Kuo Cai, Lejian Ren, Lu Ren, Liao Yu, Pengfei Zheng, Qiang Luo, Qianqian Wang, Qigen Hu, Rui Huang, Ruiming Tang, Shiyao Wang, Shujie Yang, Tao Wu, Wuchao Li, Xinchen Luo, Xingmei Wang, Yi Su, Yu...

  33. [41]

    Jieming Zhu, Mengqun Jin, Qijiong Liu, Zexuan Qiu, Zhenhua Dong, and Xiu Li

  34. [42]

    Yaochen Zhu, Harald Steck, Dawen Liang, Yinhan He, Vito Ostuni, Jundong Li, and Nathan Kallus. 2026. Rank-GRPO: Training LLM-based Conversational Recommender Systems with Reinforcement Learning. arXiv:2510.20150 [cs.IR] https://arxiv.org/abs/2510.20150

  35. [43]

    Yanyan Zou, Junbo Qi, Lunsong Huang, Yu Li, Kewei Xu, Jiabao Gao, Bin- glei Zhao, Xuanhua Yang, Sulong Xu, and Shengjie Li. 2026. GenRec: A Preference-Oriented Generative Framework for Large-Scale Recommendation. arXiv:2604.14878 [cs.IR] https://arxiv.org/abs/2604.14878

  36. [2016]

    arXiv:1511.06939 [cs.LG] https://arxiv.org/abs/1511.06939

    Session-based Recommendations with Recurrent Neural Networks. arXiv:1511.06939 [cs.LG] https://arxiv.org/abs/1511.06939

  37. [2019]

    arXiv:1904.06690 [cs.IR] https://arxiv.org/abs/ 1904.06690

    BERT4Rec: Sequential Recommendation with Bidirectional Encoder Rep- resentations from Transformer. arXiv:1904.06690 [cs.IR] https://arxiv.org/abs/ 1904.06690

  38. [2023]

    arXiv:2305.00447 [cs.IR] doi:10.1145/ 3604915.3608857

    TALLRec: An Effective and Efficient Tuning Framework to Align Large Language Model with Recommendation. arXiv:2305.00447 [cs.IR] doi:10.1145/ 3604915.3608857

  39. [2024]

    arXiv:2404.14774 [cs.IR] https://arxiv.org/abs/2404.14774

    CoST: Contrastive Quantization based Semantic Tokenization for Genera- tive Recommendation. arXiv:2404.14774 [cs.IR] https://arxiv.org/abs/2404.14774

  40. [2026]

    arXiv:2607.28590 [cs.CV] doi:10.48550/arXiv.2607.28590

    VAD: Attributing Visual Evidence for Target Reconstruction in Multimodal On-Policy Distillation. arXiv:2607.28590 [cs.CV] doi:10.48550/arXiv.2607.28590

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.