Pith. sign in

REVIEW 6 major objections 4 minor 34 references

Diffusion Model for Interest Refinement in Multi-Interest Recommendation

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

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

Pith's one-line read This paper claims that coarse multi-interest vectors can be refined into more personalized representations by adding dimension-level Gaussian noise and denoising with cross-attention over pruned historical items, and reports gains on…

desk verdict DMI has a real, if incremental, idea—dimension-level diffusion refinement for multi-interest vectors—and honest offline experiments, but the pruning mechanism is under-tested and the paper has several numeric inconsistencies. read the letter →

arxiv 2502.05561 v2 pith:JAXJ3DY5 submitted 2025-02-08 cs.IR

Yankun Le , Haoran Li , Baoyuan Ou , Yingjie Qin , Zhixuan Yang , Ruilong Su , Fu Zhang This is my paper · ORCID
classification cs.IR
keywords multi-interestrecommendationdiffusionmodelcandidatematchinginterestrefinementitempruningcross-attentiondenoisingrepresentationlearningrecommendersystems
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

Multi-interest recommenders distill a user's history into several interest vectors by attention-weighting whole items, so each vector tends to carry attributes that are irrelevant to the user's actual interest. This paper proposes DMI, which treats those coarse vectors as noisy signals: it adds small dimension-level Gaussian noise, then iteratively reconstructs a cleaner vector with a cross-attention denoiser that attends only to the historical items the base extractor judged most relevant. The authors report that DMI outperforms every compared baseline on the Book, Beauty, and Gowalla benchmarks, and that replacing only the multi-interest model in a live mainfeed recommender raised user engagement by 0.59 percent over one week. The paper argues this makes diffusion-based denoising a practical plug-in for the candidate matching stage rather than a redesign of the recommender.

What carries the argument

The load-bearing machinery is the denoising module $f_\theta$: a cross-attention transformer that reconstructs a clean interest vector from a noisy one, conditioned on a pruned set of historical items. The forward process adds scheduled isotropic Gaussian noise at the dimension level of the interest vector $v_0$ (Eqs. 6-8), and the reverse process uses the transformer with the noisy vector as query and $\text{concat}(e_t, C)$ as keys and values, where $C$ selects the TopK historical item embeddings by the extractor's attention weights (Eq. 12). The item-pruning step does the conceptual work: it decides which collaborative information is interest-relevant before attention is applied, so the reconstruction is guided by targeted context rather than the full noisy history. A gradient stop and the fusion rule in Eq. 13 keep the base extractor trained by the recommendation loss while the diffusion loss refines representations.

What would settle it

Run DMI with the TopK pruning indices in Eq. 12 replaced by random subsets of historical items of the same size; if Recall@50 does not fall outside the reported variance, the attention-based pruning is not the mechanism producing the gain. A second check is to perturb the base attention weights used for pruning and observe whether DMI's advantage over its no-pruning variant shrinks, as the paper's calibration assumption predicts.

Watch

Extended reading notes

Core claim

The central claim is that the residual error left by item-level interest extraction lives partly in individual dimensions of the interest vector, and that a diffusion loop can remove it. Starting from an attention-aggregated vector $v_0$, DMI samples noisy states $v_t = \sqrt{\bar\alpha_t}\,v_0 + \sqrt{1-\bar\alpha_t}\,\varepsilon$ using a deliberately small noise schedule, then trains a denoising module $f_\theta$ to predict $v_0$ from $v_t$ via cross-attention over the pruned historical items $C = H[\,:\, \text{TopK}_{\gamma n}(\mathbf{a})]$. The final user representation is a weighted fusion $Z_u = \eta \tilde v_0 + (1-\eta) v$, and gradients from the reconstruction loss are stopped at the extractor so the diffusion module cannot trivialize the extractor's job. The paper reports consistent Recall, Hit Rate, and NDCG gains on three public datasets and a 0.59 percent engagement lift in an online A/B test, with ablations showing that removing the diffusion loop, the cross-attention denoiser, the item pruning, or the gradient stop each degrades performance.

Load-bearing premise

The load-bearing premise is that the base extractor's attention weights are calibrated well enough that pruning the lowest-attention historical items removes interest-irrelevant noise while preserving the collaborative context the denoiser needs; if those weights are miscalibrated, the pruned context inherits the very coarse-grained error DMI is meant to fix.

Editorial extensions

If this is right

  • Across Book, Beauty, and Gowalla, DMI improves Recall@20 by 12.8 to 16.2 percent over its strongest baseline, with similar gains in Hit Rate and NDCG, indicating the refinement transfers across domains and sparsity levels.
  • Ablations attribute the gain to all four design choices: the diffusion loop, the cross-attention denoiser over an MLP, the pruning of low-attention items, and the gradient stop that protects the extractor; removing any one degrades Recall@50 by up to roughly 77 percent.
  • Category-level analysis shows retrieved items become more concentrated in categories aligned with each interest while diversity among hit items increases, implying the refinement changes which categories are retrieved, not just the ranking within them.
  • In an online A/B test on a large mainfeed recommender, replacing only the multi-interest model with DMI raised engagement by 0.59 percent, category clicks by 0.31 percent, favorites by 1.09 percent, and comments by 0.64 percent, while DDIM kept inference affordable.

Reading between the lines

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

  • A testable extension the paper does not run is to mount the same dimension-level diffusion module on other attention-based multi-interest extractors; the mechanism predicts gains whenever the base attention weights carry signal, not only with REMI/ComiRec-style attention.
  • The pruning-ratio analysis hints that the optimal selection ratio depends on how much noise the base extractor injects; comparing sensitivity curves across extractors of different quality would directly test the paper's noise-as-irrelevant-attributes interpretation.
  • If interest noise concentrates in particular dimensions, an anisotropic or dimension-dependent noise schedule could beat the isotropic Gaussian schedule used here; this follows naturally from the paper's dimension-level framing but is not tested in it.
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

6 major / 4 minor

Summary. The paper proposes DMI (Diffusion Multi-Interest model), a plug-in diffusion module that refines coarse-grained user interest vectors in multi-interest candidate matching. A base multi-interest extractor (ComiRec-SA combined with REMI) produces K interest vectors by attentively aggregating historical item embeddings. DMI adds dimension-level Gaussian noise to the selected interest vector, then reconstructs a denoised vector with a Transformer cross-attention module conditioned on a diffusion-step embedding and on a pruned set of historical items selected by their attention weights. The final user representation is a convex combination of the refined and original vectors, and training is joint over a sampled-softmax recommendation loss and a Euclidean reconstruction loss. Experiments on Amazon Book, Amazon Beauty, and Gowalla report consistent improvements over baselines, ablations support the key components, and an online A/B test reports engagement gains.

Significance. If the results hold, DMI is a practically useful and relatively simple enhancement for multi-interest matching: it shows that a diffusion-style denoiser, guided by interest-specific collaborative signals, can improve retrieval quality over a strong REMI baseline, with consistent gains across three datasets and a reported real-world deployment. The paper's strengths include the breadth of offline evaluation, the inclusion of ablation and hyper-parameter studies, and the explicit commitment to code release upon acceptance. The main weaknesses are that the central pruning mechanism is not tested against random pruning, several equations and tables are internally inconsistent or underspecified, and the online A/B test lacks the statistical detail needed to assess the claimed improvement.

major comments (6)
  1. [3.2, Eq. (8)] The scaling factor s in Eq. (8) is never defined, yet it multiplies the entire linear noise schedule and directly controls the forward process. Please define s, state its value in the implementation settings, and clarify whether it is a tunable hyper-parameter. In addition, Eq. (9) writes p_theta(hat_v_{t-1} | hat_v_t) = N(hat_v_t; mu_theta, ...), but the distribution should be over hat_v_{t-1}, and Algorithm 1 step 5 says 'given v0 and T via q(v_T|v0)' although t was sampled uniformly in step 4; these need correction.
  2. [4.5.2, Table 5; 4.3, Table 3] There are internal numerical inconsistencies that undermine confidence in the reported results. For Gowalla with the default K=4, Table 5 reports Recall@50=0.2731 and NDCG@50=0.2308, whereas Table 2 reports Recall@50=0.2735 and NDCG@50=0.2263 for the same setting. In Table 3, the Gowalla DMI-IP recall is printed as '26.93' rather than a fraction, and several 'Improv.' percentages (e.g., DMI-GD on Book and DMI-IP on Gowalla) do not match the values in the preceding columns. Please provide corrected tables and explain any differences due to seeds or settings.
  3. [3.2, 4.3, 4.5.3] The claim that the item pruning strategy removes interest-irrelevant items is not tested against random pruning. The ablation DMI-IP removes pruning entirely, but that comparison conflates two effects: the selection of items by attention weights and the reduction of cross-attention context length. The analysis in Section 4.5.3 shows a non-monotonic relationship between the pruning ratio gamma and performance, which is not obviously explained by 'drop the irrelevant tail.' Please add a control that prunes the same number of items randomly, or by the lowest attention weights, to establish that attention-based selection itself is responsible for the observed improvement.
  4. [4.3, DMI-GD] The DMI-GD ablation changes two design choices simultaneously: it removes the stop-gradient between the diffusion module and the multi-interest extractor, and it also drops the fusion operation in Eq. (13), using the denoised vector directly as the final interest representation. The reported convergence failure cannot therefore be attributed specifically to the stop-gradient. Please run separate ablations for (i) keeping fusion but removing stop-gradient and (ii) removing fusion but keeping stop-gradient, so the two effects can be disentangled.
  5. [4.6, Table 6] The online A/B test is reported only as relative percentage changes over one week, with no number of users, fraction of traffic, confidence intervals, p-values, or variance information. The statement that 'even a 0.2% improvement in user engagement is considered statistically significant' is unsupported without details of the significance testing procedure. Please report the experimental setup and standard statistical quantities for the online test.
  6. [3.3, Eq. (18)] The reconstruction loss L_dm trains the denoiser to reproduce v0, which is the coarse-grained interest vector produced by the same extractor that DMI is supposed to refine. Thus the diffusion objective by itself does not define a 'clean' interest vector independent of the coarse extractor; any removal of irrelevant attributes must come from the recommendation loss L_S through the fusion in Eq. (13). Please clarify the identification of the clean target and provide evidence that the diffusion objective contributes beyond L_S, for example by including a lambda=0 ablation or by comparing v0 with an alternative target constructed from the positive item.
minor comments (4)
  1. [Throughout] There are repeated typographical errors, including 'denosing' for 'denoising,' 'precess' for 'process,' 'RALATED WORKS' in the related-work heading, and 'DDRM' in Section 4.5.3 where DMI is meant. These should be cleaned up.
  2. [Eq. (11)] The denoising module is described in the text as taking the noisy embedding v_t, but Eq. (11) uses hat_v_t. Please standardize the notation so that the training-time and inference-time inputs to the Transformer are clearly defined.
  3. [Table 4] The column header for Div(all) has an upward arrow, but the reported DMI value is lower than the REMI value and the text describes this as a slight reduction. Please correct the arrow or the interpretation.
  4. [5.2, Table 2] Since the paper positions DMI against existing plug-in diffusion recommenders such as DiffRec and the plug-in embedding-denoising work in [28,29], it would be helpful to state explicitly why these are not included in the offline comparison, or to add them as baselines.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DMI's reported gains rest on external offline and online evaluations, not on a self-referential derivation.

full rationale

DMI is primarily an empirical learning method. Its denoising module is trained to reconstruct the coarse interest vector v0 produced by the base multi-interest extractor (Eq. 18), and the final user representation is a convex combination of this reconstruction and the original vector (Eq. 13), so the refinement is anchored to the very representation it is meant to refine. This is, however, an autoencoding objective with an external conditioning signal (item pruning and cross-attention, Eqs. 11-12), not a derivation in which the conclusion is identical to an input by construction. The paper's central claims are validated on held-out test sets (Table 2), ablations (Table 3), and an online A/B test (Table 6), none of which are statistically forced by the model construction. The item pruning strategy uses attention weights from the same extractor that produces the coarse vectors, but the paper does not derive the benefit of pruning from that identity; it proposes the mechanism and tests it empirically, including against a no-pruning variant (DMI-IP). The hyperparameter analysis in Section 4.5.3 even reports non-monotonic behavior with the pruning ratio, which is a sign of an empirical trade-off rather than a circular guarantee. Self-citations are limited to unrelated prior work by one author (refs. [12,13]) and are not load-bearing; the diffusion formulations are cited to external prior work ([28], [29]). No reported result reduces to a fitted parameter renamed as a prediction, and no uniqueness or ansatz is imported from the authors' own prior work. The observed improvements are therefore external evidence, not circularity.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard diffusion mathematics plus two modeling assumptions: that dimension-level noise is Gaussian, and that the extractor's attention weights are reliable enough for item pruning. The chief fitted quantities are hyperparameters tuned on validation data; no new physical or conceptual entities are postulated.

free parameters (7)
  • eta (fusion weight) = 0.4
    Weights the refined vector against the original coarse vector in Eq. 13; chosen by hand.
  • lambda (diffusion loss weight) = tuned in {3,4,5,6,7}
    Balances reconstruction loss L_dm against recommendation loss in Eq. 19.
  • alpha_min and alpha_max (noise schedule endpoints) = 0.0001 fixed; alpha_max tuned in {0.001,0.002,0.004,0.008}
    Control the linear noise schedule in Eq. 8.
  • diffusion steps T = selected from {5,10,20,40,80}
    Number of forward/reverse steps; chosen per dataset in Figure 4.
  • gamma (pruning ratio) = not stated; analyzed in Figure 3
    Selects the fraction of historical items kept in Eq. 12.
  • s (noise schedule scaling) = not reported
    Appears in Eq. 8 multiplying the linear schedule but is never defined in the paper.
  • number of interest heads K = 4 (default), varied in {2,4,6,8}
    Number of user interest vectors from the extractor; chosen following REMI.
assumptions (4)
  • standard math Gaussian forward and reverse diffusion equations (Eqs. 6-10) and the variational bound justify using Euclidean distance as the reconstruction objective.
    Section 3.2 invokes standard DDPM machinery without proof; accepted background from [18, 22].
  • domain assumption Interest-irrelevant attributes in an aggregated interest vector behave like additive Gaussian noise that can be removed by denoising.
    Section 1 states that aggregating interest-irrelevant attributes can be considered as adding noise and that the reverse process removes it; this is the core modeling premise.
  • ad hoc to paper Attention weights from the coarse extractor rank items by interest relevance, so pruning the lowest-weight items in Eq. 12 keeps useful information and discards noise.
    Eq. 12 selects C = H[:, TopK(a)] using the extractor's attention weights; no independent calibration evidence is provided.
  • domain assumption Fusing the denoised vector with the original coarse vector via Eq. 13 with eta = 0.4 preserves personalization without causing reconstruction loss to collapse the extractor.
    Section 3.2 and the DMI-GD ablation show this design is necessary, but it is a heuristic rather than a derived result.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Diffusion Model for Interest Refinement in Multi-Interest Recommendation." pith.science (2026). https://pith.science/paper/JAXJ3DY5

@misc{pith2026250205561,
  author       = {Pith},
  title        = {Pith review of: Diffusion Model for Interest Refinement in Multi-Interest Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JAXJ3DY5}},
  note         = {Machine review of arXiv:2502.05561}
}
read the original abstract

Multi-interest candidate matching plays a pivotal role in personalized recommender systems, as it captures diverse user interests from their historical behaviors. Most existing methods utilize attention mechanisms to generate interest representations by aggregating historical item embeddings. However, these methods only capture overall item-level relevance, leading to coarse-grained interest representations that include irrelevant information. To address this issue, we propose the Diffusion Multi-Interest model (DMI), a novel framework for refining user interest representations at the dimension level. Specifically, DMI first introduces controllable noise into coarse-grained interest representations at the dimensional level. Then, in the iterative reconstruction process, DMI combines a cross-attention mechanism and an item pruning strategy to reconstruct the personalized interest vectors with the guidance of tailored collaborative information. Extensive experiments demonstrate the effectiveness of DMI, surpassing state-of-the-art methods on offline evaluations and an online A/B test. Successfully deployed in the real-world recommender system, DMI effectively enhances user satisfaction and system performance at scale, serving the major traffic of hundreds of millions of daily active users. \footnote{The code will be released for reproducibility once the paper is accepted.}

Figures

Figures reproduced from arXiv: 2502.05561 by the authors.

Figure 1
Figure 1. Each user’s historical behaviors include multiple at [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of DMI. Given user history behaviors, the multi-interest extraction generates attention weights and derives [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Analysis on different parameters. We show Recall@50 on three datasets. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Performance and inference time cost comparison [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 12 canonical work pages

  1. [1]

    Yukuo Cen, Jianwei Zhang, Xu Zou, Chang Zhou, Hongxia Yang, and Jie Tang

  2. [2]

    Zheng Chai, Zhihong Chen, Chenliang Li, Rong Xiao, Houyi Li, Jiawei Wu, Jingxu Chen, and Haihong Tang. 2022. User-Aware Multi-Interest Learning for Candidate Matching in Recommenders. In SIGIR ’22: The 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, Madrid, Spain, July 11 - 15, 2022 . ACM, 1326–1335. https://d...

    arXiv 2022

  3. [3]

    Gaode Chen, Xinghua Zhang, Yanyan Zhao, Cong Xue, and Ji Xiang. 2021. Explor- ing Periodicity and Interactivity in Multi-Interest Framework for Sequential Rec- ommendation. In Proceedings of the Thirtieth International Joint Conference on Ar- tificial Intelligence, IJCAI 2021, Virtual Event / Montreal, Canada, 19-27 August 2021, Zhi-Hua Zhou (Ed.). ijcai....

    work page doi:10.24963/ijcai.2021/197 2021

  4. [4]

    Eunjoon Cho, Seth A Myers, and Jure Leskovec. 2011. Friendship and mobility: user movement in location-based social networks. In Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining . 1082–1090

    2011

  5. [5]

    Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep Neural Networks for YouTube Recommendations. In Proceedings of the 10th ACM Conference on Recommender Systems, Boston, MA, USA, September 15-19, 2016 . ACM, 191–198. https://doi.org/10.1145/2959100.2959190

    arXiv 2016

  6. [6]

    Yingpeng Du, Ziyan Wang, Zhu Sun, Yining Ma, Hongzhi Liu, and Jie Zhang

  7. [7]

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

  8. [8]

    Jonathan Ho, Chitwan Saharia, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans. 2022. Cascaded diffusion models for high fidelity image generation. Journal of Machine Learning Research 23, 47 (2022), 1–33

    2022

Show all 34 references
  1. [9]

    Yangqin Jiang, Yuhao Yang, Lianghao Xia, and Chao Huang. 2024. DiffKG: Knowledge Graph Diffusion Model for Recommendation. In Proceedings of the 17th ACM International Conference on Web Search and Data Mining, WSDM 2024, Merida, Mexico, March 4-8, 2024 . ACM, 313–321

    2024

  2. [10]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. 2015. Adam: A Method for Stochastic Opti- mization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings

    2015

  3. [12]

    Haoran Li, Qiang Gao, Hongmei Wu, and Li Huang. 2024. Advancing Event Causality Identification via Heuristic Semantic Dependency Inquiry Network. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 1467–1478

    2024

  4. [13]

    Haoran Li, Junqi Liu, Zexian Wang, Shiyuan Luo, Xiaowei Jia, and Huaxiu Yao

  5. [14]

    Jiuqiang Li and Hongjun Wang. 2024. Graph Diffusive Self-Supervised Learning for Social Recommendation. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2024, Washington DC, USA, July 14-18, 2024 . ACM, 2...

    2024

  6. [15]

    Hashimoto

    Xiang Li, John Thickstun, Ishaan Gulrajani, Percy Liang, and Tatsunori B. Hashimoto. 2022. Diffusion-LM Improves Controllable Text Generation. In Advances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022...

    2022

  7. [16]

    Fuyu Lv, Taiwei Jin, Changlong Yu, Fei Sun, Quan Lin, Keping Yang, and Wilfred Ng. 2019. SDM: Sequential Deep Matching Model for Online Large-scale Rec- ommender System. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management, CIKM 2019,...

    2019

  8. [17]

    arXiv preprint arXiv:2404.01165 (2024)

    LITE: Modeling Environmental Ecosystems with Multimodal Large Lan- guage Models. arXiv preprint arXiv:2404.01165 (2024)

    2024 arXiv

  9. [18]

    Yun Pang, Jiawei Mao, Libo He, Hong Lin, and Zhenping Qiang. 2024. An Improved Face Image Restoration Method Based on Denoising Diffusion Prob- abilistic Models. IEEE Access 12 (2024), 3581–3596. https://doi.org/10.1109/ ACCESS.2024.3349423

    2024

  10. [19]

    Haolei Pei, Yuanyuan Xu, Yangping Zhu, and Yuan Nie. 2024. RimiRec: Modeling Refined Multi-interest in Hierarchical Structure for Recommendation. InCompan- ion Proceedings of the ACM on Web Conference 2024, WWW 2024, Singapore, Sin- gapore, May 13-17, 2024. ACM, 746–749. https...

    2024

  11. [20]

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. 2022. High-Resolution Image Synthesis with Latent Diffusion Models. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022 . IEEE, 10...

    2022

  12. [21]

    Haokai Ma, Yimeng Yang, Lei Meng, Ruobing Xie, and Xiangxu Meng. 2024. Multimodal Conditioned Diffusion Model for Recommendation. In Companion Proceedings of the ACM on Web Conference 2024, WWW 2024, Singapore, Singapore, May 13-17, 2024. ACM, 1733–1740. https://doi.org/10.114...

    2024

  13. [22]

    Weiss, Niru Maheswaranathan, and Surya Ganguli

    Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, and Surya Ganguli

  14. [23]

    Jiaming Song, Chenlin Meng, and Stefano Ermon. 2021. Denoising Diffusion Implicit Models. In 9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021 . https://openreview.net/forum?id= St1giarCHLP

    2021

  15. [24]

    Chenyang Wang, Zhefan Wang, Yankai Liu, Yang Ge, Weizhi Ma, Min Zhang, Yiqun Liu, Junlan Feng, Chao Deng, and Shaoping Ma. 2022. Target Interest Distillation for Multi-Interest Recommendation. In Proceedings of the 31st ACM International Conference on Information & Knowledge M...

    2022 doi

  16. [25]

    Sara Sabour, Nicholas Frosst, and Geoffrey E. Hinton. 2017. Dynamic Routing Between Capsules. In Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, December 4-9, 2017, Long Beach, CA, USA . 3856–3866

    2017

  17. [26]

    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. In Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Proce...

    2023

  18. [27]

    Shengyu Zhang, Lingxiao Yang, Dong Yao, Yujie Lu, Fuli Feng, Zhou Zhao, Tat-Seng Chua, and Fei Wu. 2022. Re4: Learning to Re-contrast, Re-attend, Re-construct for Multi-interest Recommendation. In WWW ’22: The ACM Web Conference 2022, Virtual Event, Lyon, France, April 25 - 29...

    2022

  19. [28]

    Jujia Zhao, Wenjie Wang, Yiyan Xu, Teng Sun, and Fuli Feng. 2024. Plug- In Diffusion Model for Embedding Denoising in Recommendation System. CoRR abs/2401.06982 (2024). https://doi.org/10.48550/ARXIV.2401.06982 arXiv:2401.06982

    2024 doi arXiv

  20. [29]

    Jujia Zhao, Wenjie Wang, Yiyan Xu, Teng Sun, Fuli Feng, and Tat-Seng Chua. 2024. Denoising Diffusion Recommender Model. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2024, Washington DC, USA, July 14-18...

    2024

  21. [30]

    Yueqi Xie, Jingqi Gao, Peilin Zhou, Qichen Ye, Yining Hua, Jae Boum Kim, Fangzhao Wu, and Sunghun Kim. 2023. Rethinking Multi-Interest Learning for Candidate Matching in Recommender Systems. In Proceedings of the 17th ACM Conference on Recommender Systems, RecSys 2023, Singapo...

    2023

  22. [35]

    Jujia Zhao, Wenjie Wang, Yiyan Xu, Teng Sun, Fuli Feng, and Tat-Seng Chua. 2024. Denoising Diffusion Recommender Model. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2024, Washington DC, USA, July 14-18...

    2024

  23. [2015]

    In Proceedings of the 32nd International Conference on Machine Learning, ICML 2015, Lille, France, 6-11 July 2015 (JMLR Workshop and Conference Proceedings, Vol

    Deep Unsupervised Learning using Nonequilibrium Thermodynamics. In Proceedings of the 32nd International Conference on Machine Learning, ICML 2015, Lille, France, 6-11 July 2015 (JMLR Workshop and Conference Proceedings, Vol. 37) . JMLR.org, 2256–2265

    2015

  24. [2016]

    In 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings

    Session-based Recommendations with Recurrent Neural Networks. In 4th International Conference on Learning Representations, ICLR 2016, San Juan, Puerto Rico, May 2-4, 2016, Conference Track Proceedings

    2016

  25. [2020]

    In KDD ’20: The 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Virtual Event, CA, USA, August 23-27, 2020

    Controllable Multi-Interest Framework for Recommendation. In KDD ’20: The 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Virtual Event, CA, USA, August 23-27, 2020 . ACM, 2942–2951. https://doi.org/10. 1145/3394486.3403344

    2020

  26. [2024]

    Disentangled Multi-interest Representation Learning for Sequential Rec- ommendation. In Proceedings of the 30th ACM SIGKDD Conference on Knowl- edge Discovery and Data Mining, KDD 2024, Barcelona, Spain, August 25-29, 2024, Ricardo Baeza-Yates and Francesco Bonchi (Eds.). ACM,...

    2024

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.