pith. machine review for the scientific record. sign in

arxiv: 2605.05096 · v1 · submitted 2026-05-06 · 💻 cs.IR

Recognition: unknown

CapsID: Soft-Routed Variable-Length Semantic IDs for Generative Recommendation

Hang Zheng, Jianguo Lou, Menghang Gong, Qixin Guo, Wenzhuo Cheng, Zhaobin Yang, Zhengwei Zheng

Authors on Pith no claims yet

Pith reviewed 2026-05-08 15:51 UTC · model grok-4.3

classification 💻 cs.IR
keywords generative recommendationsemantic IDscapsule routingvector quantizationvariable-length tokensautoregressive retrievalinformation retrieval
0
0 comments X

The pith

Soft capsule routing in CapsID produces variable-length semantic IDs that raise recall while cutting inference cost in generative recommendation.

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

The paper shows that hard residual vector quantization collapses multi-faceted item semantics at cluster boundaries and propagates early errors into later ID positions, forcing generative recommenders to rely on extra dense vectors that increase latency. CapsID replaces the hard assignments with probabilistic capsule routing at each layer so an item can draw from several semantic capsules, the residual is reconstructed from the routed combination rather than a single code, and the ID sequence ends once the active capsule reaches high confidence. SemanticBPE then merges adjacent tokens into reusable subwords based on both co-occurrence and embedding similarity. On Amazon Beauty, Sports, Toys and a large industrial catalog this yields 9.6 percent higher Recall@10 than the strongest single-representation baseline while matching or beating hybrid sparse-dense systems at 51 percent of their latency.

Core claim

Replacing hard nearest-neighbor residual quantization with capsule routing, in which each item layer routes probabilistically to multiple semantic capsules, updates the residual with the routed reconstruction, and terminates once capsule confidence is high enough, produces semantic IDs that preserve multi-faceted item semantics without auxiliary dense representations.

What carries the argument

Capsule routing for semantic ID generation: at each quantization layer an item is softly assigned to several capsules, the residual is updated by their weighted reconstruction, and the sequence length is set by a confidence threshold.

Load-bearing premise

Soft routing via capsules and confidence-driven length termination resolve the multi-faceted semantics and error propagation issues of hard residual quantization without introducing new biases or requiring extensive hyperparameter tuning.

What would settle it

An experiment that applies CapsID to the same Amazon and industrial benchmarks but removes the soft-routing component and observes Recall@10 fall back to the hard-quantization baseline level would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.05096 by Hang Zheng, Jianguo Lou, Menghang Gong, Qixin Guo, Wenzhuo Cheng, Zhaobin Yang, Zhengwei Zheng.

Figure 1
Figure 1. Figure 1: Overview of CAPSID+SEMANTICBPE. Top: item features pass through a stack of capsule layers with confidence-driven early stopping, SEMANTICBPE merges semantically compatible adjacent tokens, and an autoregressive Transformer generates the next item’s SID via trie-constrained beam search. Bottom left: inside one CAPSID layer, soft routing replaces hard arg max; votes are reconciled over a few agreement iterat… view at source ↗
Figure 2
Figure 2. Figure 2: Variable-length behaviour of CAPSID. (a) SID length distribution per dataset: mode L= 3, mean L¯ ∈ [3.41, 3.89]. (b) Fractions of the three stopping rules that fire per dataset; the hard cap Lmax accounts for ≤10% of items, so the confidence and residual rules dominate. (c) Recall@10 by popularity tier on Beauty, relative to TIGER (log scale): gains scale from head (+19%) through torso (+30%) to tail (+140… view at source ↗
Figure 3
Figure 3. Figure 3: Tokenizer diagnostics on Beauty. (a) Code collision: view at source ↗
Figure 4
Figure 4. Figure 4: Variable-length behaviour of CAPSID across Beauty, Sports, Toys, and the industrial catalog. (a) SID length distribution: mode is L = 3 on every dataset; mean L¯ ranges from 3.41 (Beauty) to 3.89 (Toys), with the industrial catalog at 3.8 (cf view at source ↗
Figure 5
Figure 5. Figure 5: Codebook geometry and per-position prediction accuracy on Beauty. view at source ↗
read the original abstract

Generative recommendation maps each item to a sequence of Semantic IDs (SIDs) and recasts retrieval as autoregressive token generation. In this paradigm the main bottleneck is the tokenizer rather than the Transformer: residual vector quantization with a hard nearest-neighbor assignment at every layer collapses multi-faceted item semantics at cluster boundaries and propagates early errors to later SID positions. A common workaround is to append a dense vector or attribute prefix to the SID, but this dual-representation design inflates inference cost and gives up the simplicity of a generative interface. We address the bottleneck at the tokenizer itself. CAPSID replaces hard residual quantization with capsule routing: at each layer an item probabilistically routes to several semantic capsules, the residual is updated by the routed reconstruction rather than by a single winning code, and the SID terminates once the active capsule's confidence is high enough. On top of CAPSID, SEMANTICBPE composes adjacent SID tokens into reusable subwords by combining their co-occurrence with their embedding compatibility. On Amazon Beauty, Sports, Toys, and a 35M-item proprietary industrial catalog, CAPSID+SEMANTICBPE improves Recall at 10 by 9.6% on average over ReSID, the strongest single-representation baseline, and matches or exceeds a COBRA-style sparse-dense system on every public benchmark while running at 51% of its inference latency. Ablations show that soft routing, iterative agreement, and confidence-driven length each contribute independently, and the gains are largest on tail items where boundary semantics dominate.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The manuscript proposes CapsID, which replaces hard residual vector quantization in semantic ID tokenization for generative recommendation with soft capsule routing: items are probabilistically assigned to multiple semantic capsules per layer, residuals are updated via routed reconstructions, and SID length terminates based on capsule confidence. It adds SemanticBPE to compose adjacent tokens into subwords using co-occurrence and embedding compatibility. On Amazon Beauty/Sports/Toys and a 35M-item industrial catalog, CapsID+SemanticBPE reports 9.6% average Recall@10 gains over ReSID (strongest single-representation baseline), matches or exceeds COBRA-style sparse-dense systems at 51% latency, with ablations claiming independent contributions from soft routing, iterative agreement, and confidence-driven termination (largest gains on tail items).

Significance. If the empirical results and ablations hold, the work offers a targeted improvement to the core tokenizer bottleneck in generative recommenders, enabling better handling of multi-faceted semantics and reduced error propagation without hybrid representations or added inference cost. The efficiency gains and tail-item focus are practically relevant; the ablations isolating mechanism contributions provide a useful strength for validating the design choices.

minor comments (2)
  1. [Abstract] The abstract and experimental sections should report the number of runs, standard deviations, or error bars for the Recall@10 figures (including the 9.6% average) to allow assessment of statistical reliability.
  2. Provide additional implementation details or pseudocode for SemanticBPE's composition rule (co-occurrence combined with embedding compatibility) to ensure reproducibility.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary of CapsID, the recognition of its targeted improvement to the semantic ID tokenizer, and the recommendation for minor revision. We will incorporate clarifications and any minor adjustments to improve readability and completeness.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces CapsID as a capsule-routing replacement for hard residual quantization in semantic ID tokenization, followed by SemanticBPE for subword composition. All load-bearing claims (Recall@10 gains, latency reductions, ablation contributions) are tied to experimental comparisons against external baselines (ReSID, COBRA-style systems) and internal ablations on public and proprietary datasets. No equations or steps reduce by construction to fitted parameters renamed as predictions, no uniqueness theorems are imported from self-citations, and the method description does not define outputs in terms of themselves. The central argument remains an empirical proposal validated outside its own fitted values.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 2 invented entities

The approach introduces new algorithmic components (capsules, routing, BPE variant) whose effectiveness is demonstrated empirically but without external independent validation or formal proofs in the provided abstract.

free parameters (2)
  • number of semantic capsules
    Determines branching for probabilistic routing per layer; specific value not provided in abstract.
  • confidence threshold for SID termination
    Controls variable length of generated IDs; chosen to balance semantic coverage and efficiency.
axioms (1)
  • domain assumption Item semantics are multi-faceted and can be probabilistically assigned to multiple capsules without loss of information
    Core premise enabling replacement of hard nearest-neighbor assignment with soft routing.
invented entities (2)
  • semantic capsules no independent evidence
    purpose: To enable soft probabilistic routing of item features across multiple semantic groups
    New component introduced in CapsID to address quantization boundaries and error propagation.
  • SEMANTICBPE no independent evidence
    purpose: To compose adjacent SID tokens into reusable subwords based on co-occurrence and embedding compatibility
    Novel subword tokenization method built on top of CapsID outputs.

pith-pipeline@v0.9.0 · 5598 in / 1562 out tokens · 45442 ms · 2026-05-08T15:51:26.659540+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 21 canonical work pages · 6 internal anchors

  1. [1]

    Glass: A generative recommender for long-sequence modeling via sid-tier and semantic search.arXiv preprint arXiv:2602.05663, 2026

    Shiteng Cao, Junda She, Ji Liu, Bin Zeng, Chengcheng Guo, Kuo Cai, Qiang Luo, Ruiming Tang, Han Li, Kun Gai, et al. Glass: A generative recommender for long-sequence modeling via sid-tier and semantic search.arXiv preprint arXiv:2602.05663, 2026

    work page arXiv 2026

  2. [2]

    Sinkhorn distances: Lightspeed computation of optimal transport.Advances in neural information processing systems, 26, 2013

    Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport.Advances in neural information processing systems, 26, 2013

    2013

  3. [3]

    Autoregressive entity retrieval.arXiv preprint arXiv:2010.00904, 2020

    Nicola De Cao, Gautier Izacard, Sebastian Riedel, and Fabio Petroni. Autoregressive entity retrieval.arXiv preprint arXiv:2010.00904, 2020

    work page arXiv 2010

  4. [4]

    OneRec: Unifying Retrieve and Rank with Generative Recommender and Iterative Preference Alignment

    Jiaxin Deng, Shiyao Wang, Kuo Cai, Lejian Ren, Qigen Hu, Weifeng Ding, Qiang Luo, and Guorui Zhou. Onerec: Unifying retrieve and rank with generative recommender and iterative preference alignment.arXiv preprint arXiv:2502.18965, 2025

    work page internal anchor Pith review arXiv 2025

  5. [5]

    Differentiable Semantic ID for Generative Recommendation

    Junchen Fu, Xuri Ge, Alexandros Karatzoglou, Ioannis Arapakis, Suzan Verberne, Joemon M Jose, and Zhaochun Ren. Differentiable semantic id for generative recommendation.arXiv preprint arXiv:2601.19711, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  6. [6]

    Hinton, Sara Sabour, and Nicholas Frosst

    Geoffrey E. Hinton, Sara Sabour, and Nicholas Frosst. Matrix capsules with em rout- ing. InInternational Conference on Learning Representations, 2018. URL https://api. semanticscholar.org/CorpusID:65203110

    2018

  7. [7]

    Learning vector-quantized item representation for transferable sequential recommenders

    Yupeng Hou, Zhankui He, Julian McAuley, and Wayne Xin Zhao. Learning vector-quantized item representation for transferable sequential recommenders. InProceedings of the ACM Web Conference 2023, pages 1162–1171, 2023

    2023

  8. [8]

    Actionpiece: Contextually tokeniz- ing action sequences for generative recommendation,

    Yupeng Hou, Jianmo Ni, Zhankui He, Noveen Sachdeva, Wang-Cheng Kang, Ed H Chi, Julian McAuley, and Derek Zhiyuan Cheng. Actionpiece: Contextually tokenizing action sequences for generative recommendation.arXiv preprint arXiv:2502.13581, 2025

    work page arXiv 2025

  9. [9]

    Stop treating collisions equally: Qualification-aware semantic id learning for recommendation at industrial scale.arXiv preprint arXiv:2603.00632, 2026

    Zheng Hu, Yuxin Chen, Yongsen Pan, Xu Yuan, Yuting Yin, Daoyuan Wang, Boyang Xia, Zefei Luo, Hongyang Wang, Songhao Ni, et al. Stop treating collisions equally: Qualification-aware semantic id learning for recommendation at industrial scale.arXiv preprint arXiv:2603.00632, 2026

    work page arXiv 2026

  10. [10]

    Categorical Reparameterization with Gumbel-Softmax

    Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. arXiv preprint arXiv:1611.01144, 2016

    work page internal anchor Pith review arXiv 2016

  11. [11]

    Generative recommendation with semantic ids: A practitioner’s handbook

    Clark Mingxuan Ju, Liam Collins, Leonardo Neves, Bhuvesh Kumar, Louis Yufeng Wang, Tong Zhao, and Neil Shah. Generative recommendation with semantic ids: A practitioner’s handbook. InProceedings of the 34th ACM International Conference on Information and Knowledge Management, pages 6420–6425, 2025

    2025

  12. [12]

    Self-attentive sequential recommendation.2018 IEEE International Conference on Data Mining (ICDM), pages 197–206, 2018

    Wang-Cheng Kang and Julian McAuley. Self-attentive sequential recommendation.2018 IEEE International Conference on Data Mining (ICDM), pages 197–206, 2018. URL https: //api.semanticscholar.org/CorpusID:52127932

    2018

  13. [13]

    Sentencepiece: A simple and language independent subword tokenizer and detokenizer for neural text processing

    Taku Kudo and John Richardson. Sentencepiece: A simple and language independent subword tokenizer and detokenizer for neural text processing. InProceedings of the 2018 conference on empirical methods in natural language processing: System demonstrations, pages 66–71, 2018

    2018

  14. [14]

    Li, Zhiyuan Liu, Mengmeng Wu, Yuchi Xu, Pipei Huang, Huan Zhao, Guoliang Kang, Qiwei Chen, Wei Li, and Lee

    C. Li, Zhiyuan Liu, Mengmeng Wu, Yuchi Xu, Pipei Huang, Huan Zhao, Guoliang Kang, Qiwei Chen, Wei Li, and Lee. Multi-interest network with dynamic routing for recommendation at tmall.Proceedings of the 28th ACM International Conference on Information and Knowledge Management, 2019. URLhttps://api.semanticscholar.org/CorpusID:119120813

    2019

  15. [15]

    Rethinking generative recommender to- kenizer: Recsys-native encoding and semantic quantization beyond llms.arXiv preprint arXiv:2602.02338, 2026

    Yu Liang, Zhongjin Zhang, Yuxuan Zhu, Kerui Zhang, Zhiluohan Guo, Wenhang Zhou, Zonqi Yang, Kangle Wu, Yabo Ni, Anxiang Zeng, et al. Rethinking generative recommender to- kenizer: Recsys-native encoding and semantic quantization beyond llms.arXiv preprint arXiv:2602.02338, 2026. 13

    work page arXiv 2026

  16. [16]

    Generative recommender with end-to-end learnable item tokenization

    Enze Liu, Bowen Zheng, Cheng Ling, Lantao Hu, Han Li, and Wayne Xin Zhao. Generative recommender with end-to-end learnable item tokenization. InProceedings of the 48th Interna- tional ACM SIGIR Conference on Research and Development in Information Retrieval, pages 729–739, 2025

    2025

  17. [17]

    The concrete distribution: A continuous relaxation of discrete random variables

    Chris J Maddison, Andriy Mnih, and Yee Whye Teh. The concrete distribution: A continuous relaxation of discrete random variables.arXiv preprint arXiv:1611.00712, 2016

    work page arXiv 2016

  18. [18]

    Image- based recommendations on styles and substitutes.Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2015

    Julian McAuley, Christopher Targett, Javen Qinfeng Shi, and Anton van den Hengel. Image- based recommendations on styles and substitutes.Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2015. URL https://api.semanticscholar.org/CorpusID:1012652

    2015

  19. [19]

    Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of machine learning research, 21(140):1–67, 2020

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of machine learning research, 21(140):1–67, 2020

    2020

  20. [20]

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

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan H. Keshavan, Trung Hieu Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Maheswaran Sathiamoorthy. Recommender systems with generative retrieval. ArXiv, abs/2305.05065, 2023. URL https://api.semanticscholar.org/CorpusID: 258564854

    work page arXiv 2023

  21. [21]

    Sara Sabour, Nicholas Frosst, and Geoffrey E. Hinton. Dynamic routing between capsules. ArXiv, abs/1710.09829, 2017. URL https://api.semanticscholar.org/CorpusID: 3603485

    work page arXiv 2017

  22. [22]

    Methods and metrics for cold-start recommendations

    Andrew I Schein, Alexandrin Popescul, Lyle H Ungar, and David M Pennock. Methods and metrics for cold-start recommendations. InProceedings of the 25th annual international ACM SIGIR conference on Research and development in information retrieval, pages 253–260, 2002

    2002

  23. [23]

    Neural machine translation of rare words with subword units

    Rico Sennrich, Barry Haddow, and Alexandra Birch. Neural machine translation of rare words with subword units. InProceedings of the 54th annual meeting of the association for computational linguistics (volume 1: long papers), pages 1715–1725, 2016

    2016

  24. [24]

    Vectorizing the trie: Efficient constrained decoding for llm-based generative retrieval on accelerators.arXiv preprint arXiv:2602.22647, 2026

    Zhengyang Su, Isay Katsman, Yueqi Wang, Ruining He, Lukasz Heldt, Raghunandan Keshavan, Shao-Chuan Wang, Xinyang Yi, Mingyan Gao, Onkar Dalal, et al. Vectorizing the trie: Efficient constrained decoding for llm-based generative retrieval on accelerators.arXiv preprint arXiv:2602.22647, 2026

    work page arXiv 2026

  25. [25]

    Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer

    Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang. Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer. InPro- ceedings of the 28th ACM international conference on information and knowledge management, pages 1441–1450, 2019

    2019

  26. [26]

    Transformer memory as a differentiable search index.Advances in neural information processing systems, 35:21831–21843, 2022

    Yi Tay, Vinh Tran, Mostafa Dehghani, Jianmo Ni, Dara Bahri, Harsh Mehta, Zhen Qin, Kai Hui, Zhe Zhao, Jai Gupta, et al. Transformer memory as a differentiable search index.Advances in neural information processing systems, 35:21831–21843, 2022

    2022

  27. [27]

    Neural discrete representation learning.Advances in neural information processing systems, 30, 2017

    Aaron Van Den Oord, Oriol Vinyals, et al. Neural discrete representation learning.Advances in neural information processing systems, 30, 2017

    2017

  28. [28]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

    2017

  29. [29]

    Towards Efficient and Generalizable Retrieval: Adaptive Semantic Quantization and Residual Knowledge Transfer

    Huimu Wang, Xingzhi Yao, Yiming Qiu, Qinghong Zhang, Haotian Wang, Yufan Cui, Songlin Wang, Sulong Xu, and Mingming Li. Towards efficient and generalizable retrieval: Adaptive semantic quantization and residual knowledge transfer.arXiv preprint arXiv:2602.23978, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  30. [30]

    Learnable item tokenization for generative recommendation.Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, 2024

    Wenjie Wang, Honghui Bao, Xinyu Lin, Jizhi Zhang, Yongqi Li, Fuli Feng, See-Kiong Ng, and Tat-Seng Chua. Learnable item tokenization for generative recommendation.Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, 2024. URLhttps://api.semanticscholar.org/CorpusID:269757237. 14

    2024

  31. [31]

    UniRec: Bridging the Expressive Gap between Generative and Discriminative Recommendation via Chain-of-Attribute

    Ziliang Wang, Gaoyun Lin, Xuesi Wang, Shaoqiang Liang, Yili Huang, and Weijie Bian. Unirec: Bridging the expressive gap between generative and discriminative recommendation via chain-of-attribute.arXiv preprint arXiv:2604.12234, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  32. [32]

    Cofirec: Coarse-to-fine tokenization for generative recommendation.arXiv preprint arXiv:2511.22707, 2025

    Tianxin Wei, Xuying Ning, Xuxing Chen, Ruizhong Qiu, Yupeng Hou, Yan Xie, Shuang Yang, Zhigang Hua, and Jingrui He. Cofirec: Coarse-to-fine tokenization for generative recommendation.arXiv preprint arXiv:2511.22707, 2025

    work page arXiv 2025

  33. [33]

    On the convergence properties of the em algorithm.The Annals of statistics, pages 95–103, 1983

    CF Jeff Wu. On the convergence properties of the em algorithm.The Annals of statistics, pages 95–103, 1983

    1983

  34. [34]

    Mmq-v2: Align, denoise, and amplify: Adaptive behavior mining for semantic ids learning in recommendation.arXiv preprint arXiv:2510.25622, 2025

    Yi Xu, Moyu Zhang, Chaofan Fan, Jinxin Hu, Xiaochen Li, Yu Zhang, Xiaoyi Zeng, and Jing Zhang. Mmq-v2: Align, denoise, and amplify: Adaptive behavior mining for semantic ids learning in recommendation.arXiv preprint arXiv:2510.25622, 2025

    work page arXiv 2025

  35. [35]

    Generative recommendation for large-scale advertising.arXiv preprint arXiv:2602.22732, 2026

    Ben Xue, Dan Liu, Lixiang Wang, Mingjie Sun, Peng Wang, Pengfei Zhang, Shaoyun Shi, Tianyu Xu, Yunhao Sha, Zhiqiang Liu, et al. Generative recommendation for large-scale advertising.arXiv preprint arXiv:2602.22732, 2026

    work page arXiv 2026

  36. [36]

    Merge: Next-generation item indexing paradigm for large-scale streaming recommendation.arXiv preprint arXiv:2601.20199, 2026

    Jing Yan, Yimeng Bai, Zongyu Liu, Yahui Liu, Junwei Wang, Jingze Huang, Haoda Li, Sihao Ding, Shaohui Ruan, and Yang Zhang. Merge: Next-generation item indexing paradigm for large-scale streaming recommendation.arXiv preprint arXiv:2601.20199, 2026

    work page arXiv 2026

  37. [37]

    Unifying generative and dense retrieval for sequential recommendation.arXiv preprint arXiv:2411.18814, 2024

    Liu Yang, Fabian Paischer, Kaveh Hassani, Jiacheng Li, Shuai Shao, Zhang Gabriel Li, Yun He, Xue Feng, Nima Noorshams, Sem Park, et al. Unifying generative and dense retrieval for sequential recommendation.arXiv preprint arXiv:2411.18814, 2024

    work page arXiv 2024

  38. [38]

    Sparse meets dense: Unified generative recommendations with cascaded sparse-dense representations.arXiv preprint arXiv:2503.02453, 2025

    Yuhao Yang, Zhi Ji, Zhaopeng Li, Yi Li, Zhonglin Mo, Yue Ding, Kai Chen, Zijian Zhang, Jie Li, Shuanglong Li, et al. Sparse meets dense: Unified generative recommendations with cascaded sparse-dense representations.arXiv preprint arXiv:2503.02453, 2025

    work page arXiv 2025

  39. [39]

    Das: Dual- aligned semantic ids empowered industrial recommender system

    Wencai Ye, Mingjie Sun, Shaoyun Shi, Peng Wang, Wenjin Wu, and Peng Jiang. Das: Dual- aligned semantic ids empowered industrial recommender system. InProceedings of the 34th ACM International Conference on Information and Knowledge Management, pages 6217–6224, 2025

    2025

  40. [40]

    Align 3gr: Unified multi-level alignment for llm-based generative recommendation

    Wencai Ye, Mingjie Sun, Shuhang Chen, Wenjin Wu, and Peng Jiang. Align 3gr: Unified multi-level alignment for llm-based generative recommendation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 16154–16162, 2026

    2026

  41. [41]

    Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers for Generative Recommendations

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

    work page internal anchor Pith review arXiv 2024

  42. [42]

    Retention

    Bowen Zheng, Yupeng Hou, Hongyu Lu, Yu Chen, Wayne Xin Zhao, and Ji rong Wen. Adapting large language models by integrating collaborative semantics for recommendation.2024 IEEE 40th International Conference on Data Engineering (ICDE), pages 1435–1448, 2023. URL https://api.semanticscholar.org/CorpusID:265213194. A Implementation Details This appendix spec...

    2024