Pith. sign in

REVIEW 4 major objections 7 minor 42 references

GRACE: Generative Recommendation via Journey-Aware Sparse Attention on Chain-of-Thought Tokenization

T0 review · 4 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read A generative recommender that reasons category-price-brand before item tokens nearly doubles ranking accuracy on sparse e-commerce histories.

desk verdict Solid industrial generative-rec paper with a real but incremental contribution; the +106% gain is plausible but rests on unmeasured CoT token accuracy, and a few tables need fixing. read the letter →

arxiv 2507.14758 v1 pith:GFTZ5W3D submitted 2025-07-19 cs.CL cs.AIcs.IR

classification cs.CLcs.AIcs.IR
keywords multi-behaviorsequentialrecommendationgenerativechain-of-thoughttokenizationsparseattentionproductknowledgegraphsemanticjourney-awaree-commerce
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

GRACE tries to establish that a generative, transformer-based recommender can be made more accurate and cheaper if each item is tokenized as a coarse-to-fine reasoning path over product knowledge graph attributes, and if attention is restricted to journey-aware parts of the user's history. The paper reports that prepending three deterministic attribute tokens - product type, price band, brand - to the existing semantic item tokens raises next-item ranking by up to +106.9% HR@10 and +106.7% NDCG@10 over the strongest generative baseline on a sparse Home dataset, and by +22.1% HR@10 on Electronics, while a four-component sparse attention cut activated attention parameters by up to 48% at long sequences. If the paper is right, explicit structured knowledge and journey-level sparse attention are the levers that let generative recommenders scale to long, behavior-rich histories rather than richer semantic embeddings alone.

What carries the argument

The load-bearing objects are (1) Chain-of-Thought (CoT) tokenization, which turns each item into a fixed coarse-to-fine attribute prefix [PT, PRICE, BRAND] followed by T semantic IDs from balanced RQ-VAE and K-means tokenization, inserted after behavior tokens; and (2) Journey-Aware Sparse Attention (JSA), a gated combination of four sparse attention strategies over the tokenized user sequence: multi-journey compression via an MLP on blocks, top-N intra-journey block selection, inter-journey attention using only the first CoT and semantic token per item, and a truncated current-context window. The CoT prefix does the work of pruning the candidate space early in generation, while JSA makes long multi-behavior sequences tractable and multi-scale in a single trainable mechanism.

What would settle it

Measure the per-position accuracy of the three CoT attribute tokens during beam-search decoding on the Home and Electronics test sets, and the coverage of PT/PRICE/BRAND attributes in the product knowledge graph for the 48k and 32k items. If attribute-token accuracy is low yet HR@10 remains high, or if corrupting the target item's attribute tokens at inference does not substantially drop HR@10, then the CoT tokenization is not the load-bearing cause of the reported gains.

Watch

Extended reading notes

Core claim

The central claim is that the bottleneck for multi-behavior generative recommendation is not only token expressiveness but also the absence of deterministic reasoning and the quadratic cost of full attention. GRACE augments balanced semantic IDs (derived from RQ-VAE plus K-means) with a chain-of-thought trajectory - product type, price band, brand - read off the product knowledge graph, so decoding produces the behavior token, then the three attribute tokens, then the semantic item tokens, with the attribute prefix narrowing the candidate space before item identity is predicted. On the attention side, GRACE replaces full attention with a gated mixture of four sparse strategies: compressed journey blocks, top-N intra-journey block selection, coarse-grained inter-journey tokens, and a truncated current-context window. The paper reports state-of-the-art HR@10 and NDCG@10 on two real-world e-commerce datasets and up to a 48% reduction in activated attention parameters as sequence length grows.

Load-bearing premise

The gains depend on the generated product-type, price-band, and brand tokens being correct and on those attributes being present and accurate for every item in the product knowledge graph; a wrong attribute prefix prunes the candidate space so the target item cannot be recovered, and the paper does not report CoT token accuracy or PKG coverage.

Editorial extensions

If this is right

  • Explicit product-graph attributes can be generated before item identity, so the decoder's search space is pruned early; a beam of 10 candidates then suffices for top-10 ranking.
  • Journey-level sparse attention with compression, intra-, inter-, and current-context scopes matches or beats full attention while cutting activated parameters by 32%, 43%, and 48% at sequence lengths 50, 100, and 200.
  • The gains concentrate on diverse and sparse catalogs and on high-intent behaviors: +62.5% NDCG@10 for add-to-cart on Home and more than doubling Like NDCG@10 on Electronics.
  • Removing CoT tokens or any single JSA strategy degrades all three prediction tasks, with the largest drops on target add-to-cart prediction when compression and intra-journey selection are removed.
  • The CoT prefix acts as a semantic filter that localizes decoding to specific semantic clusters, reducing token dispersion in the generated item space.

Reading between the lines

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

  • The paper does not measure CoT attribute-token accuracy or product-knowledge-graph coverage; a natural extension would test whether the gains survive when attributes are missing, noisy, or predicted incorrectly.
  • The same tokenization idea could transfer to other structured spaces - sizes, colors, sellers, or multi-hop category paths - and to session-based recommendation where short histories make category-level priors especially valuable.
  • Because the efficiency comparison reports activated parameters rather than wall-clock time, the practical latency and throughput benefit on long real-world histories is still to be demonstrated.
  • If the attribute-prefix reasoning is truly load-bearing, the method suggests a testable design rule: generative recommenders for structured catalogs should generate an explicit coarse-to-fine index path before item semantics.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. GRACE is an encoder-decoder generative recommender for multi-behavior sequential recommendation. It augments behavior and semantic item tokens (from item text via RQ-VAE/K-means) with deterministic Chain-of-Thought tokens obtained by traversing a product knowledge graph in coarse-to-fine order (product type, price band, brand), and replaces full attention with a gated Journey-Aware Sparse Attention (JSA) over four scopes: compressed journey blocks, top-N intra-journey blocks, inter-journey coarse tokens, and a truncated current-journey window. On two proprietary Walmart datasets (Home, Electronics), GRACE is compared with eight baselines on target-behavior, behavior-specific, and behavior-item prediction. The reported gains over the strongest generative baseline MBGen reach +106.9% HR@10 on Home and +22.1% on Electronics, with attention computation reduced by up to 48% at long sequence lengths. Ablations remove CoT tokens and JSA components, and a hyper-parameter study fixes window size, top-N, and beam size.

Significance. If the empirical results hold, GRACE is a significant advance in generative recommendation: it shows that inserting explicit PKG attributes as intermediate generated tokens can yield large accuracy gains over a strong baseline, and JSA addresses the attention cost of dense tokenization. The paper includes a useful sanity check: the 'w/o CoT & JSA' ablation reproduces MBGen's NDCG@10 of 4.80, and evaluation ranks against all items rather than sampled negatives. However, the headline accuracy gain is conditional on the model generating correct CoT tokens and on complete PKG coverage, neither of which is measured; Section 3.6 contains numerical inconsistencies; and the efficiency metric in Table 4 is undefined. These are fixable, but they are central to the paper's claims.

major comments (4)
  1. [2.5; 3.7] Section 2.5 states that at inference the decoder generates all CoT tokens (product type, price band, brand) before the semantic item tokens, and Section 2.3.2 defines these tokens from the item's PKG attributes. If a generated CoT token is wrong, the target item's semantic tokens may fall outside the surviving beam, so the large accuracy gains reported in Table 2 are contingent on CoT token generation accuracy and on PKG completeness. The paper reports neither the accuracy of PT/PRICE/BRAND token prediction nor PKG coverage (the fraction of items for which all three attributes exist), and Section 3.7's co-occurrence heatmap only shows alignment between product-type and semantic clusters, not generation accuracy. Please report per-attribute CoT token accuracy, PKG coverage, and an error analysis of end-to-end recall conditioned on whether the generated CoT tokens match the target item's attributes.
  2. [3.6; Table 5] Section 3.6's behavior-level claims are inconsistent with Table 5. The text says Electronic Like NDCG@10 improves from 9.00 to 21.22, but Table 5 reports 16.97; it reports Home ATC improvement of +62.5% and Click +16.8%, whereas Table 5 implies +106.7% (4.80 to 9.92) and +40.5% (7.46 to 10.48). Table 5 also shows GRACE below MBGen on Home Like (18.76 vs 28.22), Home Remove (27.03 vs 33.67), and Electronic Remove (41.07 vs 46.74), a fact the text does not discuss beyond 'most behaviors.' These discrepancies must be corrected and the underperforming behaviors analyzed before the robustness claim can be accepted.
  3. [Table 4] Table 4's efficiency numbers are not interpretable. The column heading 'Activated Parameters' is not defined, and the reported values (63,504, 252,004, 1,004,004) are not consistent with parameter counts for the attention layers described in Section 2.4 (hidden size 256, six heads of dimension 64). Clarify what is being counted (e.g., attended key-value pairs, FLOPs, or activated parameters), state whether the compression MLPs, selection scoring, and gating are included, and report measured runtime or FLOPs. The 48% reduction is a headline contribution and needs a defensible measurement.
  4. [Table 2] The significance stars in Table 2 are unsupported. No standard deviations, number of independent runs, or description of the statistical test is provided, and the p-value annotation appears only in the table caption. Report variance over multiple seeds or bootstrapped confidence intervals and specify the test used; otherwise the 'significantly outperforms' wording in Section 3.2 is not evaluable.
minor comments (7)
  1. [2.3.2] The text refers to 'h-hop attributes' but the implementation fixes three attributes (PT, PRICE, BRAND); clarify whether the method generalizes to arbitrary h and how the order is chosen.
  2. [3.1.4] All sequences are truncated to 50 for training/evaluation, while Table 1 reports average sequence lengths of 171.54 and 85.43; the truncation means the efficiency gains at lengths 100/200 in Table 4 are not exercised in the reported accuracy evaluation, and the effect of truncation on accuracy is not studied.
  3. [3.3.2] The claim that directly applying NSA overlooks the complexity of user-item interactions is not tested; an NSA baseline on the same data would make the comparison concrete.
  4. [References] References [32] and [33] are the same NSA technical report; they should be merged.
  5. [Figure 2] Figure 2 is difficult to parse, in particular the token-order labels ('321321b'); please enlarge and annotate the exact token sequence layout.
  6. [3.4] The hyper-parameter analysis does not state whether the reported numbers are from a validation set or the test set; please clarify to rule out selection on the test set.
  7. [3.6] Domain names are inconsistent: the text uses 'Electronic' while tables use 'Electronics'; please unify the terminology.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: CoT generation and JSA are genuine predictive components; the single self-citation is not load-bearing.

full rationale

The claimed derivation chain is self-contained against external benchmarks. GRACE's CoT tokens are deterministic PKG attributes (product type, price band, brand) of each item, but at inference they are generated by the decoder from the encoded user history, not copied from the target; the target item's attribute sequence is not an input to the encoder. Predicting these tokens is therefore a genuine next-token prediction task, not a fitted quantity renamed as a prediction. The JSA mechanism is an adaptation of the external DeepSeek NSA work, with four newly designed scopes; its efficiency numbers (Table 4) are direct counts of activated key/value parameters for the described sparse patterns. The ablation in Table 3 explicitly equates 'w/o CoT & JSA' with MBGen and reports the same NDCG@10 (4.80) as the independently run MBGen baseline in Table 2, so the +106% improvement is a measured comparison under matched conditions rather than a construction. The only self-citation is [2], used to support the existence/construction of the product knowledge graph, an input resource; it does not carry the accuracy claim or impose a forced choice. The paper does not report CoT token accuracy or PKG coverage, and Table 5 has a discrepancy (text says Electronic Like 21.22 vs table 16.97), but these are verification and correctness concerns, not circular reductions. No equation in the paper reduces to its own input.

Assumptions & free parameters 8 free parameters · 4 assumptions · 0 invented entities

The central claim rests primarily on the PKG attribute token design and the sparse attention pattern, both of which involve hand-chosen hyperparameters that are not all reported. No mathematical derivation is offered; the paper is an empirical architecture comparison. No invented physical or conceptual entities are introduced.

free parameters (8)
  • PKG attribute set and order = [PT, PRICE, BRAND]
    Hand-chosen design; no ablation over order or attribute alternatives.
  • Number of price bands = 5
    Price discretization chosen ad hoc; no sensitivity analysis reported.
  • Semantic token levels T and codebook sizes = T=3, codebook 64, K=64
    Inherited from MBGen, still free design choices affecting the token space.
  • JSA compression block size l and stride d = not reported
    Critical parameters for compression and block segmentation; absent from implementation details.
  • Inter-journey kept tokens M_g and M_s = not reported (example M_g=M_s=1)
    Defines the coarse sequence for inter-journey attention; never specified in experiments.
  • top-N intra-journey blocks = 3
    Chosen via sensitivity analysis on the same datasets.
  • Current-journey window size w = 10
    Chosen via sensitivity analysis; trade-off with compute.
  • Beam size at inference = 10
    Chosen via sensitivity analysis; larger beams gave no gain.
assumptions (4)
  • ad hoc to paper PKG traversal [product type, price band, brand] in coarse-to-fine order provides useful deterministic context for next-item prediction and mirrors user reasoning.
    Invoked in Section 2.3.2. The attribute set and order are chosen by hand and drive most of the reported accuracy gain, but are never ablated over alternatives.
  • domain assumption Behavior merging with priority ATC > Like > Click yields a fair, non-trivial prediction task.
    Section 3.1.1. This preprocessing removes same-item chains and changes the task distribution; no comparison to other merging rules is given.
  • domain assumption The RQ-VAE/K-means semantic IDs from MBGen are faithful, generatable item identifiers.
    Section 2.3.1 and 3.1.3. The tokenization is inherited from prior work and not re-validated on these datasets.
  • domain assumption Sparse attention over compressed blocks, top-N blocks, coarse tokens, and a local window preserves enough information for accurate sequence modeling.
    Sections 2.4.1 to 2.4.4. Standard NSA-style assumption applied to recommender sequences; supported only by ablation on one dataset.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GRACE: Generative Recommendation via Journey-Aware Sparse Attention on Chain-of-Thought Tokenization." pith.science (2026). https://pith.science/paper/GFTZ5W3D

@misc{pith2026250714758,
  author       = {Pith},
  title        = {Pith review of: GRACE: Generative Recommendation via Journey-Aware Sparse Attention on Chain-of-Thought Tokenization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GFTZ5W3D}},
  note         = {Machine review of arXiv:2507.14758}
}
read the original abstract

Generative models have recently demonstrated strong potential in multi-behavior recommendation systems, leveraging the expressive power of transformers and tokenization to generate personalized item sequences. However, their adoption is hindered by (1) the lack of explicit information for token reasoning, (2) high computational costs due to quadratic attention complexity and dense sequence representations after tokenization, and (3) limited multi-scale modeling over user history. In this work, we propose GRACE (Generative Recommendation via journey-aware sparse Attention on Chain-of-thought tokEnization), a novel generative framework for multi-behavior sequential recommendation. GRACE introduces a hybrid Chain-of-Thought (CoT) tokenization method that encodes user-item interactions with explicit attributes from product knowledge graphs (e.g., category, brand, price) over semantic tokenization, enabling interpretable and behavior-aligned generation. To address the inefficiency of standard attention, we design a Journey-Aware Sparse Attention (JSA) mechanism, which selectively attends to compressed, intra-, inter-, and current-context segments in the tokenized sequence. Experiments on two real-world datasets show that GRACE significantly outperforms state-of-the-art baselines, achieving up to +106.9% HR@10 and +106.7% NDCG@10 improvement over the state-of-the-art baseline on the Home domain, and +22.1% HR@10 on the Electronics domain. GRACE also reduces attention computation by up to 48% with long sequences.

Figures

Figures reproduced from arXiv: 2507.14758 by the authors.

Figure 1
Figure 1. Comparison with existing generative recommenda [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of GRACE framework with Hybrid Tokenization and Journey-aware Sparse Attention (JSA). Hybrid [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Inference of GRACE. 2.4.3 Inter-journey Transition. Inter-journey is the transition be￾tween intra-journeys, e.g., the customer first views soccer balls and later TV tables. Inter-journey transition becomes a useful context when we focus on the high-level journey transition and correlation without getting too many details at the item-level, and learn more generic shopping patterns under the multi-behavior setting. T… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Hyper-parameter analysis range of hyper-parameters, with optimal results obtained at 𝑤=10, top-𝑁=3, and beam size=10. 3.5 Computational Efficiency To evaluate the efficiency benefits of our journal-aware sparse at￾tention mechanism, we compare GRACE’s attention layer a…
Figure 5
Figure 5. Figure 5: CoT-PT and L1 tokens co-occurrence heatmap. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 9 canonical work pages

  1. [1]

    Jianxin Chang, Chen Gao, Yu Zheng, Yiqun Hui, Yanan Niu, Yang Song, Depeng Jin, and Yong Li. 2021. Sequential recommendation with graph neural networks. InProceedings of the 44th international ACM SIGIR conference on research and development in information retrieval. 378–387

  2. [2]

    Jiao Chen, Luyi Ma, Xiaohan Li, Jianpeng Xu, Jason HD Cho, Kaushiki Nag, Evren Korpeoglu, Sushant Kumar, and Kannan Achan. 2024. Relation labeling in product knowledge graphs with large language models for e-commerce.International Journal of Machine Learning and Cybernetics15, 12 (2024), 5725–5743

  3. [3]

    Gabriel de Souza Pereira Moreira, Sara Rabhi, Jeong Min Lee, Ronay Ak, and Even Oldridge. 2021. Transformers4rec: Bridging the gap between nlp and sequential/session-based recommendation. InProceedings of the 15th ACM con- ference on recommender systems. 143–153

  4. [4]

    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 preprint arXiv:2502.18965(2025)

  5. [5]

    William Fedus, Barret Zoph, and Noam Shazeer. 2022. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity.Journal of Machine Learning Research23, 120 (2022), 1–39

  6. [6]

    Yulong Gu, Zhuoye Ding, Shuaiqiang Wang, Lixin Zou, Yiding Liu, and Dawei Yin

  7. [7]

    Long Guo, Lifeng Hua, Rongfei Jia, Binqiang Zhao, Xiaobo Wang, and Bin Cui. 2019. Buying or browsing?: Predicting real-time purchasing intent us- ing attention-based deep network with multiple behavior. InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. 1984–1992

  8. [8]

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

Show all 42 references
  1. [9]

    Yupeng Hou, Binbin Hu, Zhiqiang Zhang, and Wayne Xin Zhao. 2022. Core: simple and effective session-based recommendation within consistent represen- tation space. InProceedings of the 45th international ACM SIGIR conference on research and development in information retrieval....

  2. [10]

    Yupeng Hou, Jianmo Ni, Zhankui He, Noveen Sachdeva, Wang-Cheng Kang, Ed H Chi, Julian McAuley, and Derek Zhiyuan Cheng. 2025. ActionPiece: Contextually Tokenizing Action Sequences for Generative Recommendation.arXiv preprint arXiv:2502.13581(2025)

  3. [11]

    Bowen Jin, Hansi Zeng, Guoyin Wang, Xiusi Chen, Tianxin Wei, Ruirui Li, Zhengyang Wang, Zheng Li, Yang Li, Hanqing Lu, et al. 2023. Language models as semantic indexers.arXiv preprint arXiv:2310.07815(2023)

  4. [12]

    Wang-Cheng Kang and Julian McAuley. 2018. Self-attentive sequential recom- mendation. In2018 IEEE international conference on data mining (ICDM). IEEE, 197–206

  5. [13]

    Jinming Li, Wentao Zhang, Tian Wang, Guanglei Xiong, Alan Lu, and Gerard Medioni. 2023. GPT4Rec: A generative framework for personalized recommen- dation and user interests interpretation.arXiv preprint arXiv:2304.03879(2023)

  6. [14]

    Han Liu, Yinwei Wei, Xuemeng Song, Weili Guan, Yuan-Fang Li, and Liqiang Nie. 2024. Mmgrec: Multimodal generative recommendation with transformer model.arXiv preprint arXiv:2404.16555(2024)

  7. [15]

    Zihan Liu, Yupeng Hou, and Julian McAuley. 2024. Multi-behavior generative recommendation. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management. 1575–1585

  8. [16]

    Jianmo Ni, Gustavo Hernandez Abrego, Noah Constant, Ji Ma, Keith B Hall, Daniel Cer, and Yinfei Yang. 2021. Sentence-t5: Scalable sentence encoders from pre-trained text-to-text models.arXiv preprint arXiv:2108.08877(2021)

  9. [17]

    Aleksandr V Petrov and Craig Macdonald. 2023. Generative sequential recom- mendation with gptrec.arXiv preprint arXiv:2306.11114(2023)

  10. [18]

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Tran, Jonah Samost, et al

  11. [19]

    Steffen Rendle, Christoph Freudenthaler, and Lars Schmidt-Thieme. 2010. Factor- izing personalized markov chains for next-basket recommendation. InProceedings of the 19th international conference on World wide web. 811–820

  12. [20]

    Jiajie Su, Chaochao Chen, Zibin Lin, Xi Li, Weiming Liu, and Xiaolin Zheng

  13. [21]

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

  14. [22]

    Jiaxi Tang and Ke Wang. 2018. Personalized top-n sequential recommenda- tion via convolutional sequence embedding. InProceedings of the eleventh ACM international conference on web search and data mining. 565–573

  15. [23]

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need.Advances in neural information processing systems30 (2017)

  16. [24]

    InProceedings of the 31st ACM international conference on multimedia

    Personalized behavior-aware transformer for multi-behavior sequential recommendation. InProceedings of the 31st ACM international conference on multimedia. 6321–6331

  17. [25]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models.Advances in neural information processing systems35 (2022), 24824–24837

  18. [26]

    Shu Wu, Yuyuan Tang, Yanqiao Zhu, Liang Wang, Xing Xie, and Tieniu Tan. 2019. Session-based recommendation with graph neural networks. InProceedings of the AAAI conference on artificial intelligence, Vol. 33. 346–353

  19. [27]

    Lianghao Xia, Chao Huang, Yong Xu, and Jian Pei. 2022. Multi-behavior sequen- tial recommendation with temporal graph transformer.IEEE Transactions on Knowledge and Data Engineering35, 6 (2022), 6099–6112

  20. [28]

    Longtao Xiao, Haozhao Wang, Cheng Wang, Linfei Ji, Yifan Wang, Jieming Zhu, Zhenhua Dong, Rui Zhang, and Ruixuan Li. 2025. Progressive Collaborative and Semantic Knowledge Fusion for Generative Recommendation.arXiv preprint arXiv:2502.06269(2025)

  21. [29]

    Ye Wang, Jiahao Xun, Minjie Hong, Jieming Zhu, Tao Jin, Wang Lin, Haoyuan Li, Linjun Li, Yan Xia, Zhou Zhao, et al. 2024. EAGER: Two-Stream Generative Recommender with Behavior-Semantic Collaboration. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and D...

  22. [30]

    Zhiyu Yao, Xinyang Chen, Sinan Wang, Qinyan Dai, Yumeng Li, Tanchao Zhu, and Mingsheng Long. 2024. Recommender transformers with behavior pathways. InProceedings of the ACM Web Conference 2024. 3643–3654

  23. [31]

    Enming Yuan, Wei Guo, Zhicheng He, Huifeng Guo, Chengkai Liu, and Ruiming Tang. 2022. Multi-behavior sequential transformer recommender. InProceedings of the 45th international ACM SIGIR conference on research and development in information retrieval. 1642–1652

  24. [33]

    Jingyang Yuan, Huazuo Gao, Damai Dai, Junyu Luo, Liang Zhao, Zhengyan Zhang, Zhenda Xie, YX Wei, Lean Wang, Zhiping Xiao, et al. 2025. Native sparse attention: Hardware-aligned and natively trainable sparse attention, 2025.URL https://arxiv. org/abs/2502.11089(2025)

  25. [34]

    Yuhao Yang, Chao Huang, Lianghao Xia, Yuxuan Liang, Yanwei Yu, and Chen- liang Li. 2022. Multi-behavior hypergraph-enhanced transformer for sequential recommendation. InProceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining. 2263–2274

  26. [35]

    Neil Zeghidour, Alejandro Luebs, Ahmed Omran, Jan Skoglund, and Marco Tagliasacchi. 2021. Soundstream: An end-to-end neural audio codec.IEEE/ACM Transactions on Audio, Speech, and Language Processing30 (2021), 495–507

  27. [36]

    Jiaqi Zhai, Lucy Liao, Xing Liu, Yueming Wang, Rui Li, Xuan Cao, Leon Gao, Zhao- jie Gong, Fangda Gu, Michael He, et al. 2024. Actions speak louder than words: Trillion-parameter sequential transducers for generative recommendations.arXiv preprint arXiv:2402.17152(2024)

  28. [37]

    Tingting Zhang, Pengpeng Zhao, Yanchi Liu, Victor S Sheng, Jiajie Xu, Deqing Wang, Guanfeng Liu, Xiaofang Zhou, et al . 2019. Feature-level deeper self- attention network for sequential recommendation.. InIJCAI. 4320–4326

  29. [38]

    Kun Zhou, Hui Wang, Wayne Xin Zhao, Yutao Zhu, Sirui Wang, Fuzheng Zhang, Zhongyuan Wang, and Ji-Rong Wen. 2020. S3-rec: Self-supervised learning for se- quential recommendation with mutual information maximization. InProceedings of the 29th ACM international conference on inf...

  30. [39]

    Nasser Zalmout, Chenwei Zhang, Xian Li, Yan Liang, and Xin Luna Dong. 2021. All you need to know to build a product knowledge graph. InProceedings of the 27th ACM SIGKDD Conference on knowledge discovery & data mining. 4090–4091

  31. [2015]

    Session-based recommendations with recurrent neural networks.arXiv preprint arXiv:1511.06939(2015)

  32. [2019]

    InProceedings of the 28th ACM international conference on information and knowledge management

    BERT4Rec: Sequential recommendation with bidirectional encoder rep- resentations from transformer. InProceedings of the 28th ACM international conference on information and knowledge management. 1441–1450

  33. [2020]

    InProceedings of the 29th ACM international conference on information & knowledge management

    Deep multifaceted transformers for multi-objective ranking in large-scale e-commerce recommender systems. InProceedings of the 29th ACM international conference on information & knowledge management. 2493–2500

  34. [2023]

    Recommender systems with generative retrieval.Advances in Neural Information Processing Systems36 (2023), 10299–10315

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.