Pith. sign in

REVIEW 5 major objections 5 minor 56 references

Distributionally Robust Graph Out-of-Distribution Recommendation via Diffusion Model

T0 review · 5 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that adding latent diffusion denoising and an entropy-regularized Sinkhorn DRO objective to a graph recommender stops worst-case training from latching onto noisy samples, and that the resulting DRGO model outperforms…

desk verdict A plausible engineering recipe for noise-robust DRO recommendation, but the advertised theoretical guarantee does not cover the test distribution and the empirical report has enough inconsistencies to require major revision. read the letter →

arxiv 2501.15555 v1 pith:R2SMIUWI submitted 2025-01-26 cs.LG cs.AIcs.GRstat.ML

Chu Zhao , Enneng Yang , Yuliang Liang , Jianzhe Zhao , Guibing Guo , Xingwei Wang This is my paper · ORCID
classification cs.LGcs.AIcs.GRstat.ML
keywords Out-of-distributionrecommendationDistributionallyrobustoptimizationGraphneuralnetworksDiffusiondenoisingEntropyregularizationSinkhorndistanceBetweennesscentralityPopularityshift
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that current DRO-based graph recommenders fail in an overlooked way: when they optimize for the worst-case distribution, they end up assigning the largest weights to noise, and the model spends its capacity learning features that do not transfer to out-of-distribution data. DRGO is proposed as a fix that combines three pieces: a VGAE plus diffusion module that denoises user-item embeddings before grouping, Sinkhorn DRO instead of KL-based DRO so the uncertainty set remains meaningful when training and test supports barely overlap, and an entropy regularization term that keeps worst-case group weights from collapsing onto outliers. The paper backs the fix with a generalization bound (Theorem 3.1) and a gradient-level analysis of why the denoising and entropy terms reduce noise influence. If the central claim is right, robust OOD recommenders can be made markedly more accurate on shifted and ordinary data without needing explicit environment labels.

What carries the argument

Two objects carry the argument. The first is the uncertainty-set construction: the nominal distribution $P_{train}$ is formed from embeddings of the top-$n\%$ of nodes ranked by betweenness centrality (the frequency with which a node lies on shortest paths between other nodes), and the uncertainty set $Q$ is the set of KMeans clusters over the denoised user embeddings. The second is the joint objective of Eq. (12), which couples BPR loss, a Sinkhorn DRO constraint $W_{c,\lambda}(P_{train},Q) \le \rho$, the entropy regularization term $-\beta \sum_i w_{q_i}\log w_{q_i}$, and the VGAE-plus-diffusion denoising loss. The machinery works by making $Q$ a coarse, denoised proxy of the latent user distribution, by replacing the KL divergence with an entropy-regularized transport cost that remains finite under non-overlapping support, and by penalizing extreme group weights. Theorem 3.1 then converts the Sinkhorn distance into a generalization bound, while the gradient analysis shows the denoising and entropy terms reduce the influence of noisy samples on parameter updates.

What would settle it

Construct a test split from the least-connected tail of the training graph on Yelp2018 or Douban, and compare DRGO's Recall@20 with DR-GNN's; if the advantage disappears or reverses, the centrality prior, not the diffusion or entropy modules, is carrying the result. A more direct check is to measure the Sinkhorn distance between the top-centrality embedding distribution and the actual test embedding distribution on each shifted dataset and see whether DRGO's margin shrinks as that distance grows.

Watch

Extended reading notes

Core claim

The central claim is that a DRO-based graph recommender can be made robust to both distribution shift and training noise by jointly optimizing the objective in Eq. (12). That objective is a BPR ranking loss computed over KMeans groups of denoised latent embeddings, with a Sinkhorn DRO ambiguity set centered on a nominal distribution built from the top-$n\%$ most central nodes and a negative-entropy penalty $-\beta \sum_i w_{q_i}\log w_{q_i}$ on the group weights. The paper argues that previous DRO recommenders are dominated by noise because the worst-case distribution concentrates weight there; the diffusion path corrects gradients of noisy samples and the entropy term bounds the concentration, while Sinkhorn distance keeps the optimization well-defined even when the test distribution has no overlap with training support. Theorem 3.1 bounds the OOD generalization risk by Rademacher complexity, the Sinkhorn distance $W(P_{train},Q)$, and a sample-size term. Experiments on Food, KuaiRec, Yelp2018, and Douban under popularity, temporal, and exposure shifts report that DRGO beats the compared baselines on both OOD and IID splits.

Load-bearing premise

The load-bearing premise is that the most-connected bridge users and items in the training graph are the ones that determine the unseen test distribution; if the shifted test set is driven by low-centrality, off-the-beaten-path users, the robust optimization is aimed at the wrong distribution.

Editorial extensions

If this is right

  • DRO-based graph recommenders that ignore noise can be upgraded by plugging in latent diffusion and an entropy penalty, since the paper's ablations show the method also improves when mounted on SimGCL and LightGCL backbones.
  • Sinkhorn DRO gives graph recommenders a usable uncertainty set even when training and test interaction distributions barely overlap, because the entropy-regularized transport cost stays finite where KL divergence diverges.
  • The betweenness-centrality prior provides an environment-free way to build the nominal distribution, so the method does not need explicit environment labels to prepare for distribution shift.
  • At 25% injected edge noise, DRGO degrades less than the baselines, suggesting it can tolerate interaction logs with a substantial share of corrupt or accidental edges.
  • The generalization bound centers OOD risk on the Sinkhorn distance from nominal to test groups, so improving the nominal distribution choice should be at least as valuable as adding model capacity.

Reading between the lines

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

  • If the centrality prior generalizes beyond these four datasets, the same recipe could apply to other graph learning tasks where noisy edges are common and shift is concentrated around hub nodes, such as fraud detection or social feed ranking.
  • The entropy coefficient $\beta$ likely controls a phase transition: very small values should recover standard DRO with noise over-weighting, and very large values should push group weights toward uniform; a sweep across $\beta$ might localize exactly where the noise regime begins.
  • Measuring the Sinkhorn distance between the top-centrality embedding distribution and the actual test embedding distribution on each shifted dataset would give a direct test of Theorem 3.1: datasets with larger measured distance should show smaller DRGO margins.
  • Replacing the betweenness ranking with a learnable group-importance score could reveal whether the topological prior is necessary or merely convenient, and whether the method can adapt to shifts that are not centrality-driven.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes DRGO, a graph recommendation method for out-of-distribution (OOD) generalization that combines a variational graph autoencoder with a latent diffusion denoising module, a Sinkhorn distributionally robust optimization (DRO) objective over cluster weights, and an entropy regularization term intended to prevent DRO from concentrating weight on noisy samples. The paper claims two theoretical contributions: a generalization bound (Theorem 3.1) for DRGO and a gradient-based analysis of how diffusion denoising and entropy regularization mitigate noise. The method is evaluated on four datasets under popularity, temporal, and exposure shift, as well as IID conditions, and reports large improvements over existing graph recommenders. The code is released. The central claims are that current DRO-based recommenders over-weight noise, that DRGO fixes this via diffusion and entropy regularization, and that Sinkhorn DRO provides an OOD generalization guarantee.

Significance. If the theoretical and experimental claims were sound, DRGO would be a meaningful contribution: it would be the first DRO-based graph recommender to explicitly target noise sensitivity, coupling a diffusion denoiser with a Sinkhorn-constrained worst-case loss, and it would provide a formal generalization bound with a topological prior. The paper also provides code and extensive experiments across three distribution-shift scenarios, which are useful resources. However, the theoretical results as written are not established: the key Wasserstein term in the bound is asserted without derivation, the boundedness constant relies on a false bound on BPR loss, and the entropy-gradient formula has a sign error. In addition, the claimed OOD guarantee is tied to a training-derived nominal distribution and uncertainty set, not to the test distribution, and the paper offers no validation that the topological prior places the test distribution inside the Sinkhorn ball. The experimental tables also contain implausible inconsistencies. Because these issues affect the paper's main claims, the current contribution is not reliable in its present form.

major comments (5)
  1. [Appendix A.3, Eq. (52)] Theorem 3.1's bound includes the term B W(P_train, Q), but the proof never derives it. The Rademacher bound in Eq. (51) is a standard i.i.d. result and contains no Wasserstein distance; the transition to Eq. (52) is stated as "Furthermore, we have" with no argument linking the error on Q to the error on P_train via a transportation cost. A Wasserstein term typically arises from a Lipschitz condition on the loss with respect to the transport cost or from a distribution-distance inequality, none of which is stated or proved. Consequently, the central generalization guarantee of the paper is unproven as written.
  2. [Appendix A.3, Eq. (41)] The claim that the BPR loss is bounded by log2 is false for negative margins. For z = f(x_u) - f(x_i), the BPR loss is -log sigma(z). Since sigma(0)=0.5, the inequality -log sigma(z) <= log2 holds only when z >= 0. In BPR training, negative items are sampled randomly, so z is often negative, and then sigma(z) < 0.5, which makes the loss strictly larger than log2. Therefore the constant B in Eq. (48) is not established, and the boundedness assumption used in Theorem 3.1 is unsupported.
  3. [Section 3.3.4, Eq. (15)] The gradient of the negative entropy term -sum_i w_i log w_i is written as -log w_i + 1, but the correct derivative is -log w_i - 1. This sign error invalidates the stated mechanism: the paper claims that when w_i is large the gradient is large and therefore the weight is reduced, which does not follow from the printed formula. Since this gradient analysis is the paper's theoretical explanation of how entropy regularization mitigates noise, the noise-mitigation argument is not supported.
  4. [Tables 1 and 2] On the KuaiRec dataset, the OOD results in Table 1 are implausibly large compared with the IID results in Table 2: for example, LightGCN has NDCG@10 = 0.5096 in Table 1 but 0.0174 in Table 2, and the same discrepancy appears across baselines and metrics. Several entries in Table 1 also report NDCG@20 lower than NDCG@10 (e.g., Ours: 0.6367 vs. 0.6837), which, although not mathematically impossible, is unusual and inconsistent with the IID table. These numbers require clarification or correction; otherwise the reported improvements are not trustworthy.
  5. [Section 3.3.2 and Theorem 3.1] The DRO guarantee is stated for a distribution Q that is built from KMeans of training embeddings (Eq. 10) and a nominal distribution P_train built from the top-n% betweenness-centrality training nodes (Eq. 9 and Section 3.3.2). The Wasserstein term W(P_train, Q) in Theorem 3.1 therefore measures a distance between two training-derived quantities, not between the training distribution and the unseen test distribution. The topological prior asserted in Section 3.3.2 — that high-centrality training groups strongly influence unseen distribution groups — is never validated, and no experiment reports W(P_train, P_test) or checks whether P_test lies within the Sinkhorn radius rho. Without such validation, the bound does not explain the reported OOD improvements, which may be due to the diffusion or regularization terms alone.
minor comments (5)
  1. [Appendix C] The implementation paragraph says "We implement our CausalDiffRec in Pytorch," which appears to be a leftover from a different paper; it should say DRGO.
  2. [Algorithm 1, lines 5] The algorithm states "Calculate the nominal distribution Q by Eq. (9) and the uncertainty sets P_train is calculated by Eq. (10)," but Eq. (9) defines betweenness centrality, not a nominal distribution, and Eq. (10) defines Q, not P_train. The labels should be corrected.
  3. [Theorem 3.1 and Eq. (14)] The statement says "For delta > 0, with probability at least 1-delta," but Eq. (14) contains sqrt(ln(1/sigma)/2n) with sigma, not delta. The relationship between sigma and delta is not defined.
  4. [Reference list] Reference [1] is listed as an arXiv preprint, but Table 1 and the text cite LightGCL as ICLR 2023; the reference should be updated to the published version.
  5. [Eq. (3)] The expectation is written as E_{P_train(x,y)∼Q}, which is malformed; it should be E_{(x,y)∼Q}.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the empirical claims are tested against external baselines, and the theoretical OOD bound rests on an explicitly stated centrality assumption rather than a hidden reduction.

full rationale

The paper's main empirical claim (DRGO improves accuracy under popularity, temporal, and exposure shifts and on IID data) is supported by comparisons to external baselines on four datasets, so it does not reduce to the model's own definitions. The theoretical result, Theorem 3.1 (Eq. 14), is a standard DRO/Rademacher-type bound whose uncertainty term W(P_train,Q) is computed between two training-constructed objects: P_train from top-n% betweenness-centrality training nodes (Eq. 9) and Q from KMeans of denoised training embeddings (Eq. 10). The paper explicitly acknowledges that the test distribution is unknown and bridges the gap with the topological-prior assumption: 'groups with high centrality during training strongly influence unseen distribution groups [4,20]' (Section 3.3.2). That assumption is unvalidated, and no experiment measures W(P_train,P_test) or checks whether test embeddings fall inside the Sinkhorn ball, but it is stated as an assumption rather than smuggled in as a derived consequence. The bound itself is not equivalent to its input: it still relates empirical and true risk on Q. Hyperparameters rho, K, and top-n are tuned and reported as tuned (Section 4.4), not fitted and then relabeled as predictions. Several non-circular concerns deserve flags: the proof of Theorem 3.1 jumps from Eq. (51) to Eq. (52) by appending B W(P_train,Q) without a derivation, leaving the theorem incomplete as a rigorous OOD guarantee; Algorithm 1 swaps the labels 'nominal distribution Q by Eq. (9)' and 'uncertainty sets P_train by Eq. (10)'; and Appendices C and D refer to 'CausalDiffRec' ('We implement our CausalDiffRec in Pytorch', 'We evaluate CausalDiffRec against these leading models'), indicating text reuse from the authors' prior work [44]. These affect correctness, clarity, and provenance, but they are not cases of a claimed prediction reducing by construction to its own inputs. Hence the derivation chain is not circular in the sense this pass is asked to detect.

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

The central claim rests on several unproven modeling choices: the topological prior, cluster-as-environment identification, Lipschitz and boundedness assumptions for the loss, and hyperparameter-dependent uncertainty geometry. The free-parameter count is dominated by standard tuning but also includes rho, K, top-n, and beta, which jointly define the DRO problem.

free parameters (7)
  • Sinkhorn radius rho = 0.05 (chosen from [0.001, 0.01, 0.05, 0.1, 0.5])
    Controls the size of the DRO uncertainty set; tuned per dataset in Section 4.4.
  • Number of KMeans clusters K = 5 (chosen from [1, 3, 5, 8, 10])
    Defines the groups forming the uncertainty set; tuned per dataset.
  • Betweenness centrality top-n% = 10% (chosen from [1%, 5%, 10%, 15%, 25%])
    Selects nodes that form the nominal distribution; tuned per dataset.
  • Entropy penalty coefficient beta = Range [0.1, 0.001, 0.0001, 0.00001]
    Appendix C labels it weight decay, but in Eq. (12) it multiplies the entropy regularization term; fitted by search.
  • Maximum diffusion steps T = Range [20, 50, 100, 200, 500]
    Diffusion schedule length; tuned per dataset.
  • Embedding size d = Range [16, 32, 64, 128]
    Standard model capacity hyperparameter; tuned per dataset.
  • Number of GNN layers L = Range [1, 2, 3, 4, 5]
    Standard model depth hyperparameter; tuned per dataset.
assumptions (7)
  • ad hoc to paper The BPR loss is bounded by log2 for all inputs.
    Used to define the constant B in Theorem 3.1; Eq. (41) only holds when the positive-minus-negative score is non-negative, which is not guaranteed during training.
  • ad hoc to paper The rating function f is K-Lipschitz, and the BPR loss is G-Lipschitz and L-smooth.
    Assumed in Appendix A.3 to justify the Rademacher and Wasserstein bound; not proven for LightGCN embeddings.
  • domain assumption High-betweenness-centrality training groups strongly influence unseen distribution groups.
    Central to nominal distribution construction in Section 3.3.2, citing [4,20]; not theoretically established.
  • domain assumption KMeans clusters of denoised user embeddings correspond to environments or groups relevant to distribution shift.
    Used to define the uncertainty set Q in Eq. (10); no evidence is given that clusters align with actual shift factors.
  • domain assumption The topological prior guarantees the test distribution lies inside the Sinkhorn ball around P_train.
    Needed for the DRO guarantee to transfer to real OOD data; it is asserted, not stated as a theorem.
  • standard math Sinkhorn distance with entropy regularization is a valid replacement for KL divergence for non-overlapping distributions.
    This is a legitimate entropic optimal transport definition, but the paper never shows how the constraint is enforced during optimization.
  • ad hoc to paper Noise sample weights are related to noise variance in the way assumed in the variance formula.
    Appendix A.2 assumes w_o(j) relates to sigma_o^2 and clean weights relate to sigma_c^2 without a derivation of that relationship.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Distributionally Robust Graph Out-of-Distribution Recommendation via Diffusion Model." pith.science (2026). https://pith.science/paper/R2SMIUWI

@misc{pith2026250115555,
  author       = {Pith},
  title        = {Pith review of: Distributionally Robust Graph Out-of-Distribution Recommendation via Diffusion Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R2SMIUWI}},
  note         = {Machine review of arXiv:2501.15555}
}
read the original abstract

The distributionally robust optimization (DRO)-based graph neural network methods improve recommendation systems' out-of-distribution (OOD) generalization by optimizing the model's worst-case performance. However, these studies fail to consider the impact of noisy samples in the training data, which results in diminished generalization capabilities and lower accuracy. Through experimental and theoretical analysis, this paper reveals that current DRO-based graph recommendation methods assign greater weight to noise distribution, leading to model parameter learning being dominated by it. When the model overly focuses on fitting noise samples in the training data, it may learn irrelevant or meaningless features that cannot be generalized to OOD data. To address this challenge, we design a Distributionally Robust Graph model for OOD recommendation (DRGO). Specifically, our method first employs a simple and effective diffusion paradigm to alleviate the noisy effect in the latent space. Additionally, an entropy regularization term is introduced in the DRO objective function to avoid extreme sample weights in the worst-case distribution. Finally, we provide a theoretical proof of the generalization error bound of DRGO as well as a theoretical analysis of how our approach mitigates noisy sample effects, which helps to better understand the proposed framework from a theoretical perspective. We conduct extensive experiments on four datasets to evaluate the effectiveness of our framework against three typical distribution shifts, and the results demonstrate its superiority in both independently and identically distributed distributions (IID) and OOD.

Figures

Figures reproduced from arXiv: 2501.15555 by the authors.

Figure 1
Figure 1. (a) Divide the user-item interaction data in the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The proposed DRGO model schematic. An bipar [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Analysis of the impact of various hyperparameters [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 40 canonical work pages

  1. [1]

    Xuheng Cai, Chao Huang, Lianghao Xia, and Xubin Ren. 2023. LightGCL: Simple yet effective graph contrastive learning for recommendation. arXiv preprint arXiv:2302.08191 (2023)

    arXiv 2023

  2. [2]

    Wei Chen, Yuxuan Liu, Zhao Zhang, Fuzhen Zhuang, and Jiang Zhong. 2024. Mod- eling Adaptive Inter-Task Feature Interactions via Sentiment-Aware Contrastive Learning for Joint Aspect-Sentiment Prediction. In AAAI, Vol. 38. 17781–17789

    work page 2024

  3. [3]

    Wei Chen, Yiqing Wu, Zhao Zhang, Fuzhen Zhuang, Zhongshi He, Ruobing Xie, and Feng Xia. 2024. FairGap: Fairness-aware Recommendation via Generating Counterfactual Graph. TOIS 42, 4 (2024), 1–25

    work page 2024

  4. [4]

    LC Freeman. 1977. A set of measures of centrality based on betweenness. So- ciometry (1977)

    work page 1977

  5. [5]

    Chen Gao, Yu Zheng, Nian Li, Yinfeng Li, Yingrong Qin, Jinghua Piao, Yuhan Quan, Jianxin Chang, Depeng Jin, Xiangnan He, et al. 2023. A survey of graph neural networks for recommender systems: Challenges, methods, and directions. TORS 1, 1 (2023), 1–51

    work page 2023

  6. [6]

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. In SIGIR. 639–648

    2020

  7. [7]

    Yue He, Zimu Wang, Peng Cui, Hao Zou, Yafeng Zhang, Qiang Cui, and Yong Jiang. 2022. Causpref: Causal preference learning for out-of-distribution recom- mendation. In WWW. 410–421

    work page 2022

  8. [8]

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. 2020. Denoising diffusion probabilistic models. NIPS 33 (2020), 6840–6851

    work page 2020

Show all 56 references
  1. [9]

    Yangqin Jiang, Yuhao Yang, Lianghao Xia, and Chao Huang. 2024. Diffkg: Knowl- edge graph diffusion model for recommendation. In WWW. 313–321

    2024

  2. [10]

    Thomas N Kipf and Max Welling. 2016. Variational graph auto-encoders. arXiv preprint arXiv:1611.07308 (2016)

    2016 arXiv

  3. [11]

    Xiao Li, Li Sun, Mengjie Ling, and Yan Peng. 2023. A survey of graph neural network based recommendation in social networks. Neurocomputing 549 (2023), 126441

    2023

  4. [12]

    Zongwei Li, Lianghao Xia, and Chao Huang. 2024. RecDiff: Diffusion Model for Social Recommendation. arXiv preprint arXiv:2406.01629 (2024)

    2024 arXiv

  5. [13]

    Xinyu Lin, Wenjie Wang, Jujia Zhao, Yongqi Li, Fuli Feng, and Tat-Seng Chua

  6. [14]

    Chan Liu, Lun Li, Xiaolu Yao, and Lin Tang. 2019. A survey of recommendation algorithms based on knowledge graph embedding. In CSEI. IEEE, 168–171

    2019

  7. [15]

    Jiashuo Liu, Tianyu Wang, Peng Cui, and Hongseok Namkoong. 2024. On the need for a language describing distribution shifts: Illustrations on tabular datasets. NIPS 36 (2024)

    2024

  8. [16]

    Jiashuo Liu, Jiayun Wu, Tianyu Wang, Hao Zou, Bo Li, and Peng Cui. 2023. Geometry-Calibrated DRO: Combating Over-Pessimism with Free Energy Impli- cations. arXiv preprint arXiv:2311.05054 (2023)

    2023 arXiv

  9. [17]

    Yong Liu, Susen Yang, Yonghui Xu, Chunyan Miao, Min Wu, and Juyong Zhang

  10. [18]

    Haokai Ma, Yimeng Yang, Lei Meng, Ruobing Xie, and Xiangxu Meng. 2024. Multimodal Conditioned Diffusion Model for Recommendation. In WWW. 1733– 1740

    2024

  11. [19]

    Hyelin Nam, Gihyun Kwon, Geon Yeong Park, and Jong Chul Ye. 2024. Contrastive denoising score for text-guided latent diffusion image editing. In CVPR. 9192– 9201

    2024

  12. [20]

    Fengchun Qiao and Xi Peng. 2023. Topology-aware robust optimization for out-of-distribution generalization. arXiv preprint arXiv:2307.13943 (2023)

    2023 arXiv

  13. [21]

    Yifang Qin, Hongjun Wu, Wei Ju, Xiao Luo, and Ming Zhang. 2023. A diffusion model for poi recommendation. TOIS 42, 2 (2023), 1–27

    2023

  14. [22]

    Hamed Rahimian and Sanjay Mehrotra. 2019. Distributionally robust optimiza- tion: A review. arXiv preprint arXiv:1908.05659 (2019)

    2019 arXiv

  15. [23]

    Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme

  16. [24]

    Tianhao Shi, Yang Zhang, Jizhi Zhang, Fuli Feng, and Xiangnan He. 2024. Fair Recommendations with Limited Sensitive Attributes: A Distributionally Robust Optimization Approach. arXiv preprint arXiv:2405.01063 (2024)

    2024 arXiv

  17. [25]

    Bohao Wang, Jiawei Chen, Changdong Li, Sheng Zhou, Qihao Shi, Yang Gao, Yan Feng, Chun Chen, and Can Wang. 2024. Distributionally Robust Graph-based Recommendation System. In WWW. 3777–3788

    2024

  18. [26]

    Wenjie Wang, Xinyu Lin, Fuli Feng, Xiangnan He, Min Lin, and Tat-Seng Chua

  19. [27]

    Wenjie Wang, Xinyu Lin, Liuhui Wang, Fuli Feng, Yunshan Ma, and Tat-Seng Chua. 2023. Causal disentangled recommendation against user preference shifts. TOIS 42, 1 (2023), 1–27

    2023

  20. [28]

    Wenjie Wang, Yiyan Xu, Fuli Feng, Xinyu Lin, Xiangnan He, and Tat-Seng Chua

  21. [29]

    Xiang Wang, Xiangnan He, Yixin Cao, Meng Liu, and Tat-Seng Chua. 2019. Kgat: Knowledge graph attention network for recommendation. In KDD. 950–958

    2019

  22. [30]

    Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019. Neural graph collaborative filtering. In SIGIR. 165–174

    2019

  23. [31]

    Yunke Wang, Xiyu Wang, Anh-Dung Dinh, Bo Du, and Charles Xu. 2023. Learning to schedule in diffusion probabilistic models. In KDD. 2478–2488

    2023

  24. [32]

    Zimu Wang, Yue He, Jiashuo Liu, Wenchao Zou, Philip S Yu, and Peng Cui. 2022. Invariant preference learning for general debiasing in recommendation. In KDD. 1969–1978

    2022

  25. [33]

    Jiancan Wu, Xiang Wang, Fuli Feng, Xiangnan He, Liang Chen, Jianxun Lian, and Xing Xie. 2021. Self-supervised graph learning for recommendation. In SIGIR. 726–735

    2021

  26. [34]

    Le Wu, Junwei Li, Peijie Sun, Richang Hong, Yong Ge, and Meng Wang. 2020. Diffnet++: A neural influence and interest diffusion network for social recom- mendation. TKDE 34, 10 (2020), 4753–4766

    2020

  27. [35]

    Yiyan Xu, Wenjie Wang, Fuli Feng, Yunshan Ma, Jizhi Zhang, and Xiangnan He. 2024. Diffusion Models for Generative Outfit Recommendation. In SIGIR. 1350–1359

    2024

  28. [36]

    Jiyuan Yang, Yue Ding, Yidan Wang, Pengjie Ren, Zhumin Chen, Fei Cai, Jun Ma, Rui Zhang, Zhaochun Ren, and Xin Xin. 2024. Debiasing Sequential Recom- menders through Distributionally Robust Optimization over System Exposure. In WSDM. 882–890

    2024

  29. [37]

    Zhengyi Yang, Xiangnan He, Jizhi Zhang, Jiancan Wu, Xin Xin, Jiawei Chen, and Xiang Wang. 2023. A generic learning framework for sequential recommendation with distribution shifts. In SIGIR. 331–340

    2023

  30. [38]

    Zixuan Yi, Xi Wang, and Iadh Ounis. 2024. A Directional Diffusion Graph Transformer for Recommendation. arXiv preprint arXiv:2404.03326 (2024)

    2024 arXiv

  31. [39]

    Junliang Yu, Hongzhi Yin, Xin Xia, Tong Chen, Lizhen Cui, and Quoc Viet Hung Nguyen. 2022. Are graph augmentations necessary? simple graph contrastive learning for recommendation. In SIGIR. 1294–1303

    2022

  32. [40]

    An Zhang, Wenchang Ma, Pengbo Wei, Leheng Sheng, and Xiang Wang. 2024. General Debiasing for Graph-based Collaborative Filtering via Adversarial Graph Dropout. In WWW. 3864–3875

    2024

  33. [41]

    An Zhang, Leheng Sheng, Zhibo Cai, Xiang Wang, and Tat-Seng Chua. 2024. Empowering collaborative filtering with principled adversarial contrastive loss. NIPS 36 (2024)

    2024

  34. [42]

    An Zhang, Jingnan Zheng, Xiang Wang, Yancheng Yuan, and Tat-Seng Chua

  35. [43]

    Chuxu Zhang, Dongjin Song, Chao Huang, Ananthram Swami, and Nitesh V Chawla. 2019. Heterogeneous graph neural network. In KDD. 793–803

    2019

  36. [44]

    Chu Zhao, Enneng Yang, Yuliang Liang, Pengxiang Lan, Yuting Liu, Jianzhe Zhao, Guibing Guo, and Xingwei Wang. 2024. Graph Representation Learning via Causal Diffusion for Out-of-Distribution Recommendation. arXiv preprint arXiv:2408.00490 (2024)

    2024 arXiv

  37. [45]

    Jujia Zhao, Wenjie Wang, Xinyu Lin, Leigang Qu, Jizhi Zhang, and Tat-Seng Chua. 2023. Popularity-aware Distributionally Robust Optimization for Recom- mendation System. In CIKM. 4967–4973. WWW ’25, April 28-May 2, 2025, Sydney, NSW, Australia Trovato et al

    2023

  38. [46]

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

  39. [47]

    Rui Zhou, Xian Wu, Zhaopeng Qiu, Yefeng Zheng, and Xu Chen. 2023. Distribu- tionally Robust Sequential Recommnedation. In SIGIR. 279–288

    2023

  40. [48]

    Invariant collaborative filtering to popularity distribution shift. In WWW. 1240–1251

  41. [49]

    Yuanzhi Zhu, Kai Zhang, Jingyun Liang, Jiezhang Cao, Bihan Wen, Radu Timofte, and Luc Van Gool. 2023. Denoising diffusion models for plug-and-play image restoration. In CVPR. 1219–1229. A PROOFS AND DERIV ATIONS A.1 The proof of KL divergence Proof. Given the mathematical defi...

    2023

  42. [53]

    In SIGIR

    Denoising diffusion recommender model. In SIGIR. 1370–1379

  43. [55]

    Rong Zhu, Kun Zhao, Hongxia Yang, Wei Lin, Chang Zhou, Baole Ai, Yong Li, and Jingren Zhou. 2019. Aligraph: A comprehensive graph neural network platform. arXiv preprint arXiv:1902.08730 (2019)

    2019 arXiv

  44. [2009]

    BPR: Bayesian personalized ranking from implicit feedback. In UAI. 452– 461

  45. [2021]

    TKDE 35, 1 (2021), 181–195

    Contextualized graph attention network for recommendation with item knowledge graph. TKDE 35, 1 (2021), 181–195

    2021

  46. [2022]

    Causal representation learning for out-of-distribution recommendation. In WWW. 3562–3571

  47. [2023]

    In SIGIR

    Diffusion recommender model. In SIGIR. 832–841

  48. [2024]

    In AAAI, Vol

    Temporally and distributionally robust optimization for cold-start recom- mendation. In AAAI, Vol. 38. 8750–8758

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.