pith. sign in

arxiv: 2604.04427 · v1 · submitted 2026-04-06 · 💻 cs.IR · cs.CL

FAVE: Flow-based Average Velocity Establishment for Sequential Recommendation

Ke Shi , Yao Zhang , Feng Guo , Jinyuan Zhang , JunShuo Zhang , Shen Gao , Shuo Shang This is my paper

Pith reviewed 2026-05-10 20:29 UTC · model grok-4.3

classification 💻 cs.IR cs.CL
keywords sequential recommendationflow-based generationone-step inferencevelocity fieldsemantic anchorgenerative modelsinference efficiency
0
0 comments X

The pith

FAVE learns a single average velocity vector from a user-history semantic anchor to enable accurate one-step generative sequential recommendation.

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

The paper introduces FAVE to overcome the inefficiencies of flow-based generative models in sequential recommendation, which typically start from random noise and require multiple iterative steps. By using a two-stage training process, it first creates a stable preference space through semantic alignment at both ends of the sequence, then initializes the flow with a masked user history embedding and learns a global average velocity. A consistency constraint based on Jacobian-vector products ensures the trajectory is straight enough for single-step generation. This matters because it promises to make high-quality generative recommendations fast enough for real-world use where latency matters.

Core claim

FAVE addresses prior mismatch and linear redundancy in flow models by establishing a semantic anchor prior from user interaction history and learning a global average velocity that consolidates the recovery trajectory into one displacement vector, enforced by a JVP-based consistency constraint, resulting in one-step generation that maintains recommendation accuracy.

What carries the argument

The average velocity vector, which replaces multi-step iterative solving by representing the entire flow trajectory as a single displacement from the semantic anchor prior, with straightness enforced by the JVP consistency constraint.

If this is right

  • One-step inference replaces multi-step recovery without loss of accuracy.
  • Inference speed improves by an order of magnitude on standard benchmarks.
  • Generative recommendation becomes practical for latency-sensitive applications.
  • State-of-the-art performance is achieved across three recommendation datasets.
  • The two-stage training prevents representation collapse while enabling efficient generation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar velocity-based consolidation could simplify other flow or diffusion models in sequential tasks.
  • Semantic anchors from history might improve efficiency in non-recommendation generative modeling.
  • Further work could test if the average velocity approximates a closed-form solution for preference transitions.
  • The approach may extend to real-time personalization systems where speed is critical.

Load-bearing premise

The JVP-based consistency constraint combined with the semantic anchor prior is enough to make a single learned velocity vector produce accurate next-item predictions equivalent to full trajectory recovery.

What would settle it

Running the full multi-step flow model and the one-step FAVE on the same test set and checking if their recommendation metrics like NDCG or Hit Rate differ significantly.

Figures

Figures reproduced from arXiv: 2604.04427 by Feng Guo, Jinyuan Zhang, JunShuo Zhang, Ke Shi, Shen Gao, Shuo Shang, Yao Zhang.

Figure 1
Figure 1. Figure 1: Comparison of generative recommendation paradigms. (Left) Existing “Noise-to-Data” generative meth￾ods suffer from prior mismatch and linear redundancy. (Right) Our proposed Fave employs direct trajectory trans￾port from a semantic anchor prior. 1 Introduction The core objective of sequential recommendation (SR) is to accu￾rately model the dynamic evolution of user intent, reflected in user interaction seq… view at source ↗
Figure 2
Figure 2. Figure 2: The overall framework of Fave which adopts a two-stage training strategy. It first constructs a basic manifold by [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of embedding distributions on ML [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of inference trajectories for Fave (left) [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance of different hyperparameters. (a) [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
read the original abstract

Generative recommendation has emerged as a transformative paradigm for capturing the dynamic evolution of user intents in sequential recommendation. While flow-based methods improve the efficiency of diffusion models, they remain hindered by the ``Noise-to-Data'' paradigm, which introduces two critical inefficiencies: prior mismatch, where generation starts from uninformative noise, forcing a lengthy recovery trajectory; and linear redundancy, where iterative solvers waste computation on modeling deterministic preference transitions. To address these limitations, we propose a Flow-based Average Velocity Establishment (Fave) framework for one-step generation recommendation that learns a direct trajectory from an informative prior to the target distribution. Fave is structured via a progressive two-stage training strategy. In Stage 1, we establish a stable preference space through dual-end semantic alignment, applying constraints at both the source (user history) and target (next item) to prevent representation collapse. In Stage 2, we directly resolve the efficiency bottlenecks by introducing a semantic anchor prior, which initializes the flow with a masked embedding from the user's interaction history, providing an informative starting point. Then we learn a global average velocity, consolidating the multi-step trajectory into a single displacement vector, and enforce trajectory straightness via a JVP-based consistency constraint to ensure one-step generation. Extensive experiments on three benchmarks demonstrate that Fave not only achieves state-of-the-art recommendation performance but also delivers an order-of-magnitude improvement in inference efficiency, making it practical for latency-sensitive scenarios.

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

2 major / 2 minor

Summary. The paper introduces FAVE, a flow-based framework for sequential recommendation that enables one-step generation by learning a global average velocity from an informative semantic anchor prior. It employs a two-stage training process: Stage 1 uses dual-end semantic alignment to stabilize the preference space, and Stage 2 incorporates a masked history embedding as prior, learns the average velocity, and applies a JVP-based consistency constraint to straighten the trajectory for efficient inference. The authors claim state-of-the-art performance and an order-of-magnitude improvement in inference efficiency on three benchmarks.

Significance. If the empirical claims hold, FAVE would represent a meaningful advance in generative recommendation models by substantially reducing inference latency while preserving recommendation quality. The use of a single displacement vector instead of iterative sampling addresses a key practical limitation of flow and diffusion models in real-time recommendation systems. The two-stage approach with semantic anchors and JVP constraints offers a novel way to handle prior mismatch and trajectory redundancy.

major comments (2)
  1. [Method (Stage 2)] The description of the global average velocity and the JVP-based consistency constraint lacks a formal derivation or error analysis showing that a single learned vector suffices to approximate the next-item distribution accurately for sparse user sequences; the skeptic concern that residual curvature in discrete trajectories may require additional steps is not addressed quantitatively.
  2. [Experiments] The abstract and experimental claims assert SOTA results and efficiency gains without providing quantitative metrics, ablation studies on the JVP term, or comparisons to multi-step baselines; this makes it impossible to verify whether the one-step approximation maintains performance across varying sequence lengths and sparsity levels.
minor comments (2)
  1. [Abstract] The abstract mentions 'three benchmarks' but does not name them; including dataset names would improve clarity.
  2. [Notation] The term 'JVP-based consistency constraint' is introduced without prior definition or reference to the Jacobian-vector product computation in the context of flow models.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the potential practical impact of FAVE on inference latency in generative recommendation. We respond to each major comment below and outline the revisions we will make.

read point-by-point responses

  1. Referee: [Method (Stage 2)] The description of the global average velocity and the JVP-based consistency constraint lacks a formal derivation or error analysis showing that a single learned vector suffices to approximate the next-item distribution accurately for sparse user sequences; the skeptic concern that residual curvature in discrete trajectories may require additional steps is not addressed quantitatively.

    Authors: We agree that the current presentation would benefit from greater theoretical grounding. The JVP consistency term is introduced to penalize deviations from a straight trajectory under the semantic anchor prior, but we acknowledge the absence of a formal error bound. In the revised manuscript we will add a derivation section that shows the one-step displacement approximates the target distribution with error controlled by the residual curvature and the strength of the consistency loss. We will also report quantitative curvature measurements on sparse subsequences from the three benchmarks to directly address whether additional steps would be required. revision: yes

  2. Referee: [Experiments] The abstract and experimental claims assert SOTA results and efficiency gains without providing quantitative metrics, ablation studies on the JVP term, or comparisons to multi-step baselines; this makes it impossible to verify whether the one-step approximation maintains performance across varying sequence lengths and sparsity levels.

    Authors: The experimental section already contains tables with concrete HR@K, NDCG@K, and wall-clock inference times on the three benchmarks, together with comparisons against both diffusion and flow baselines. Nevertheless, we accept that an explicit ablation isolating the JVP term and stratified analyses by sequence length and sparsity are missing. In the revision we will insert an ablation table that removes the JVP constraint, plus additional results broken down by sequence-length quartiles and user-interaction sparsity levels, including direct head-to-head numbers against the corresponding multi-step flow variants. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The provided abstract and outline describe a two-stage training process introducing dual-end semantic alignment in Stage 1 and a semantic anchor prior plus JVP-based consistency constraint in Stage 2 to enable one-step generation via a learned global average velocity. No equations, self-citations, or definitions are shown that reduce the velocity vector or consistency loss to a tautological fit of the target distribution by construction. The central claims rest on empirical results from three benchmarks rather than any self-referential derivation, satisfying the criteria for a self-contained method.

Axiom & Free-Parameter Ledger

1 free parameters · 0 axioms · 1 invented entities

The abstract introduces several new modeling choices whose grounding is not visible: the semantic anchor prior, the global average velocity, and the JVP consistency constraint. No explicit free parameters or axioms are stated.

free parameters (1)
  • global average velocity
    Learned displacement vector that replaces the multi-step trajectory; its value is fitted during Stage 2 training.
invented entities (1)
  • semantic anchor prior no independent evidence
    purpose: Informative starting embedding obtained by masking user history to avoid uninformative noise
    Introduced to solve prior mismatch; no independent evidence of correctness is provided in the abstract.

pith-pipeline@v0.9.0 · 5568 in / 1276 out tokens · 32208 ms · 2026-05-10T20:29:10.128670+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

50 extracted references · 50 canonical work pages

  1. [1]

    Albergo and Eric Vanden-Eijnden

    Michael S. Albergo and Eric Vanden-Eijnden. 2023. Building Normalizing Flows with Stochastic Interpolants. InThe Eleventh International Conference on Learning Representations, ICLR 2023. OpenReview.net

    work page 2023

  2. [2]

    Chu, Usman Naseem, Yang Wang, and Fang Chen

    Zhuo Cai, Shoujin Wang, Victor W. Chu, Usman Naseem, Yang Wang, and Fang Chen. 2025. Unleashing the Potential of Diffusion Models Towards Diversified Sequential Recommendations. InProceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2025. ACM, 1476–1486

    work page 2025

  3. [3]

    Jialei Chen, Yuanbo Xu, and Yiheng Jiang. 2025. Unlocking the Power of Diffusion Models in Sequential Recommendation: A Simple and Effective Approach. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining, V.2, KDD 2025, Toronto ON, Canada, August 3-7, 2025. ACM, 155–166

    work page 2025

  4. [4]

    Mingyue Cheng, Qi Liu, Wenyu Zhang, Zhiding Liu, Hongke Zhao, and Enhong Chen. 2024. A general tail item representation enhancement framework for sequential recommendation.Frontiers Comput. Sci.18, 6 (2024), 186333

    work page 2024

  5. [5]

    Bronstein, and Avishek Joey Bose

    Oscar Davis, Samuel Kessler, Mircea Petrache, İsmail İlkan Ceylan, Michael M. Bronstein, and Avishek Joey Bose. 2024. Fisher Flow Matching for Generative Modeling over Discrete Data. InProceedings of the 38th NeurIPS Conference on Neural Information Processing Systems

    work page 2024

  6. [6]

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, Dustin Podell, Tim Dockhorn, Zion English, and Robin Rombach. 2024. Scaling rectified flow transformers for high-resolution image synthesis. InProceedings of the 41st International Conference on Machine Learning...

    work page 2024

  7. [7]

    Ziwei Fan, Zhiwei Liu, Yu Wang, Alice Wang, Zahra Nazari, Lei Zheng, Hao Peng, and Philip S. Yu. 2022. Sequential Recommendation via Stochastic Self-Attention. InProceedings of the 31st The Web Conference on WWW. 2036–2047

    work page 2022

  8. [8]

    Junchen Fu, Xuri Ge, Xin Xin, Alexandros Karatzoglou, Ioannis Arapakis, Jie Wang, and Joemon M. Jose. 2024. IISAN: Efficiently Adapting Multimodal Repre- sentation for Sequential Recommendation with Decoupled PEFT. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval. ACM, 687–697

    work page 2024

  9. [9]

    Ruining He, Wang-Cheng Kang, and Julian J. McAuley. 2017. Translation-based Recommendation. InProceedings of the 11th ACM Conference on Recommender Systems, Paolo Cremonesi, Francesco Ricci, Shlomo Berkovsky, and Alexander Tuzhilin (Eds.). ACM, 161–169

    work page 2017

  10. [10]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yong-Dong Zhang, and Meng Wang. 2020. LightGCN: Simplifying and Powering Graph Convolution Net- work for Recommendation. InProceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval, SIGIR. ACM, 639–648

    work page 2020

  11. [11]

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

    work page

  12. [12]

    InPro- ceedings of the 4th International Conference on Learning Representations Conference on Learning Representations

    Session-based Recommendations with Recurrent Neural Networks. InPro- ceedings of the 4th International Conference on Learning Representations Conference on Learning Representations

    work page

  13. [13]

    Yangqin Jiang, Yuhao Yang, Lianghao Xia, and Chao Huang. 2024. DiffKG: Knowledge Graph Diffusion Model for Recommendation. InProceedings of the 17th WSDM Conference on Web Search and Data Mining. 313–321

    work page 2024

  14. [14]

    Wang-Cheng Kang and Julian J. McAuley. 2018. Self-Attentive Sequential Rec- ommendation. InProceedings of the 18th IEEE International Conference on Data Mining Conference, ICDM 2018. 197–206

    work page 2018

  15. [15]

    Minu Kim, Yongsik Lee, Sehyeok Kang, Jihwan Oh, Song Chong, and Se-Young Yun. 2024. Preference Alignment with Flow Matching. InProceedings of the 38th Neural Information Processing Systems Conference on NeurIPS Conference on 2024

    work page 2024

  16. [16]

    Gasnikov, and Alexander Korotin

    Nikita Kornilov, Petr Mokrov, Alexander V. Gasnikov, and Alexander Korotin

    work page

  17. [17]

    InProceedings of the 38th NeurIPS Conference on Neural Information Processing Systems

    Optimal Flow Matching: Learning Straight Trajectories in Just One Step. InProceedings of the 38th NeurIPS Conference on Neural Information Processing Systems

    work page

  18. [18]

    Hanyu Li, Jiayu Li, Weizhi Ma, Peijie Sun, Haiyang Wu, Jingwen Wang, Yuekui Yang, Min Zhang, and Shaoping Ma. 2025. CD-CDR: Conditional Diffusion- based Item Generation for Cross-Domain Recommendation. InProceedings of the 48th SIGIR Conference on Research and Development in Information Retrieval. 1789–1798

    work page 2025

  19. [19]

    Li Li, Mingyue Cheng, Yuyang Ye, Zhiding Liu, and Enhong Chen. 2025. Prefer- ence Trajectory Modeling via Flow Matching for Sequential Recommendation. CoRRabs/2508.17618 (2025)

    work page arXiv 2025

  20. [20]

    Zihao Li, Aixin Sun, and Chenliang Li. 2024. DiffuRec: A Diffusion Model for Sequential Recommendation.ACM Trans. Inf. Syst.42, 3 (2024), 66:1–66:28

    work page 2024

  21. [21]

    Xiao Lin, Xiaokai Chen, Chenyang Wang, Hantao Shu, Linfeng Song, Biao Li, and Peng Jiang. 2024. Discrete Conditional Diffusion for Reranking in Recom- mendation. InCompanion Proceedings of the ACM on Web Conference 2024, WWW 2024, Singapore, Singapore, May 13-17, 2024. ACM, 161–169

    work page 2024

  22. [22]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. 2023. Flow Matching for Generative Modeling. InProceedings of the 11th International Conference on Learning Representations ICRL 2023. Open- Review.net

    work page 2023

  23. [23]

    Chengyi Liu, Xiao Chen, Shijie Wang, Wenqi Fan, and Qing Li. 2025. Continuous- time Discrete-space Diffusion Model for Recommendation.CoRRabs/2511.12114 (2025)

    work page arXiv 2025

  24. [24]

    Chengkai Liu, Yangtian Zhang, Jianling Wang, Rex Ying, and James Caverlee

    work page

  25. [25]

    InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Flow Matching for Collaborative Filtering. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining. ACM, 1765–1775

    work page

  26. [26]

    Feng Liu, Lixin Zou, Xiangyu Zhao, Min Tang, Liming Dong, Dan Luo, Xiangyang Luo, and Chenliang Li. 2025. Flow Matching Based Sequential Recommender Model. InProceedings of the 34th International Joint Conference on Artificial Intelligence, IJCAI 2025. ijcai.org, 3108–3116

    work page 2025

  27. [27]

    Pan Liu, Xiaohua Tian, and Zhouhan Lin. 2024. Enable Fast Sampling for Seq2Seq Text Diffusion. InProceedings of the 2024 EMNLP Conference on Findings. 8495– 8505

    work page 2024

  28. [28]

    Sijia Liu, Jiahao Liu, Hansu Gu, Dongsheng Li, Tun Lu, Peng Zhang, and Ning Gu. 2023. AutoSeqRec: Autoencoder for Efficient Sequential Recommendation. InProceedings of the 32nd ACM International Conference on Information and Knowledge Management Conference, CIKM 2023. 1493–1502

    work page 2023

  29. [29]

    Julian McAuley, Christopher Targett, Qinfeng Shi, and Anton Van Den Hengel

    work page

  30. [30]

    InProceedings of the 38th international ACM SIGIR conference on research and development in information retrieval

    Image-based recommendations on styles and substitutes. InProceedings of the 38th international ACM SIGIR conference on research and development in information retrieval. 43–52

    work page

  31. [31]

    Preetum Nakkiran, Arwen Bradley, Hattie Zhou, and Madhu Advani. 2025. Step- by-Step Diffusion: An Elementary Tutorial.Found. Trends Comput. Graph. Vis. 17, 1 (2025), 1–75

    work page 2025

  32. [32]

    Yong Niu, Xing Xing, Zhichun Jia, Ruidi Liu, Mindong Xin, and Jianfu Cui. 2024. Diffusion Recommendation with Implicit Sequence Influence. InProceedings of the 33rd ACM Web Conference on WWW 2024. 1719–1725

    work page 2024

  33. [33]

    Yifang Qin, Wei Ju, Hongjun Wu, Xiao Luo, and Ming Zhang. 2024. Learning Graph ODE for Continuous-Time Sequential Recommendation.IEEE Trans. Knowl. Data Eng.36, 7 (2024), 3224–3236

    work page 2024

  34. [34]

    Ruihong Qiu, Zi Huang, Hongzhi Yin, and Zijian Wang. 2022. Contrastive Learn- ing for Representation Degeneration Problem in Sequential Recommendation. InProceedings of the 15th ACM International Conference on Web Search and Data Mining, WSDM 2022, K. Selcuk Candan, Huan Liu, Leman Akoglu, Xin Luna Dong, and Jiliang Tang (Eds.). ACM, 813–823

    work page 2022

  35. [35]

    Tran, Jonah Samost, Maciej Kula, Ed H

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Mahesh Sathiamoorthy. 2023. Recommender Systems with Generative Retrieval. InProceedings of the 36th NeurIPS Conference on Neural Information Processing Systems

    work page 2023

  36. [36]

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

    work page

  37. [37]

    InProceedings of the 28th ACM Conference on Information and Knowledge Management ,CIKM 2019

    BERT4Rec: Sequential Recommendation with Bidirectional Encoder Rep- resentations from Transformer. InProceedings of the 28th ACM Conference on Information and Knowledge Management ,CIKM 2019. 1441–1450

    work page 2019

  38. [38]

    Jiaxi Tang and Ke Wang. 2018. Personalized Top-N Sequential Recommenda- tion via Convolutional Sequence Embedding. InProceedings of the 11th ACM International Conference on Web Search and Data Mining Conference on WSDM. 565–573

    work page 2018

  39. [39]

    Dongsheng Wang, Yuxi Huang, Shen Gao, Yifan Wang, Chengrui Huang, and Shuo Shang. 2025. Generative Next POI Recommendation with Semantic ID. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining. ACM, 2904–2914

    work page 2025

  40. [40]

    Yifan Wang, Shen Gao, Jiabao Fang, Lisi Chen, Peng Han, and Shuo Shang. 2025. DRE: Generating Recommendation Explanations by Aligning Large Language Models at Data-Level. InDatabase Systems for Advanced Applications - Proceedings of the 30th International Conference, DASFAA 2025 (Lecture Notes in Computer Science). Springer, 286–302. FAVE: Flow-based Aver...

    work page 2025

  41. [41]

    Yifan Wang, Shen Gao, Jiabao Fang, Rui Yan, Billy Chiu, and Shuo Shang. 2025. CESRec: Constructing Pseudo Interactions for Sequential Recommendation via Conversational Feedback. InFindings of the Association for Computational Lin- guistics: EMNLP. Association for Computational Linguistics, 16227–16239

    work page 2025

  42. [42]

    Yuhao Wang, Ziru Liu, Yichao Wang, Xiangyu Zhao, Bo Chen, Huifeng Guo, and Ruiming Tang. 2024. Diff-MSR: A Diffusion Model Enhanced Paradigm for Cold-Start Multi-Scenario Recommendation. InProceedings of the 17th ACM International Conference on Web Search and Data Mining Conference on WSDM. 779–787

    work page 2024

  43. [43]

    Yuanbo Xu, En Wang, Yongjian Yang, and Hui Xiong. 2024. GS-RS: A Generative Approach for Alleviating Cold Start and Filter Bubbles in Recommender Systems. IEEE Trans. Knowl. Data Eng.36, 2 (2024), 668–681

    work page 2024

  44. [44]

    Yiyan Xu, Wenjie Wang, Fuli Feng, Yunshan Ma, Jizhi Zhang, and Xiangnan He

    work page

  45. [45]

    InProceedings of the 47th SIGIR Conference on Research and Development in Information Retrieval

    Diffusion Models for Generative Outfit Recommendation. InProceedings of the 47th SIGIR Conference on Research and Development in Information Retrieval. 1350–1359

    work page

  46. [46]

    Zhengyi Yang, Jiancan Wu, Zhicai Wang, Xiang Wang, Yancheng Yuan, and Xiangnan He. 2023. Generate What You Prefer: Reshaping Sequential Recom- mendation via Guided Diffusion. InProceedings of the 36th Neural Information Processing Systems Conference on Neural Information Processing Systems, nips 2023

    work page 2023

  47. [47]

    Yi Zhang, Yiwen Zhang, Yu Wang, Tong Chen, and Hongzhi Yin. 2025. Towards Distribution Matching between Collaborative and Language Spaces for Genera- tive Recommendation. InProceedings of the 48th SIGIR Conference on Research and Development in Information Retrieval. 2006–2016

    work page 2025

  48. [48]

    Jujia Zhao, Wenjie Wang, Yiyan Xu, Teng Sun, Fuli Feng, and Tat-Seng Chua

    work page

  49. [49]

    InProceedings of the 47th SIGIR Conference on Research and Development in Information Retrieval

    Denoising Diffusion Recommender Model. InProceedings of the 47th SIGIR Conference on Research and Development in Information Retrieval. 1370–1379

    work page

  50. [50]

    Yunqin Zhu, Chao Wang, Qi Zhang, and Hui Xiong. 2024. Graph Signal Diffusion Model for Collaborative Filtering. InProceedings of the 47th SIGIR Conference on Research and Development in Information Retrieval. 1380–1390

    work page 2024