Hypergraph Diffusion for High-Order Recommender Systems

Darnbi Sakong; Jun Jo; Thanh Trung Huynh

REVIEW 5 major objections 6 minor 1 cited by

Hypergraph Diffusion for High-Order Recommender Systems

T0 review · 5 major / 6 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 WaveHDNN, a wavelet-hypergraph diffusion recommender, claims consistent wins over six baselines on three datasets across all reported metrics.

desk verdict A plausible, incrementally novel hypergraph CF model with a sensible two-encoder design, but the empirical superiority claim is undercut by inconsistent tables and missing uncertainty estimates. read the letter →

arxiv 2501.16722 v1 pith:BJUE6WLD submitted 2025-01-28 cs.IR cs.AIcs.DBcs.LGcs.SI

Darnbi Sakong , Thanh Trung Huynh , Jun Jo This is my paper · ORCID

classification cs.IRcs.AIcs.DBcs.LGcs.SI

keywords recommendersystemscollaborativefilteringhypergraphdiffusionheterophilywavelettransformmulti-scalestructureencodingcontrastivelearningover-smoothing

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

This paper introduces WaveHDNN, a collaborative-filtering model that treats user-item interactions as a hypergraph and propagates information through wavelet-based diffusion. It claims that WaveHDNN outperforms six existing graph- and hypergraph-based recommender baselines on Amazon-Books, Steam, and Yelp across all six reported metrics, Recall@10/20/40 and NDCG@10/20/40. The design separates two failure modes the authors say limit prior GNN recommenders: heterophilic interactions, where a user's items span many categories, and over-smoothing, where deep layers make embeddings indistinguishable. A heterophily-aware encoder handles the first, a multi-scale wavelet structure encoder handles the second, and cross-view contrastive learning aligns them. If the comparisons are fair, the result is a practical recipe for higher-order group-wise recommendation with consistent gains over strong baselines.

What carries the argument

The load-bearing mechanism is a two-channel hypergraph diffusion architecture. Channel one, the Heterophily-aware Collaborative Encoder, transforms node embeddings with MLPs, passes them through a hypergraph convolution layer, applies layer normalization, and adds the result back to the original embeddings; this equivariant residual design lets messages to nodes sharing a hyperedge differ rather than converge. Channel two, the Multi-scale Group-wise Structure Encoder, is a wavelet hypergraph convolution $X^{(l+1)} = \Theta \Lambda \Theta' X^{(l)} W + X^{(l)}$, where $\Theta$ and $\Theta'$ are the wavelet basis and its inverse and $\Lambda$ is a diagonal frequency filter; this is what localizes message passing to specific regions and scales of the hypergraph. The two channels are tied together by an InfoNCE-style cross-view contrastive loss and trained with a BPR ranking loss.

What would settle it

Re-run WaveHDNN and the six baselines on the same three datasets with an equal hyperparameter budget and report per-seed standard deviations; if any baseline ties or beats WaveHDNN on any of the six metrics, the uniform superiority claim fails. A cheaper check is to reproduce Table 3's SHT NDCG@20 entry, because the printed value (0.02647) sits below that model's NDCG@10 value (0.07583), and to check Table 5's WaveHDNN Amazon Recall@40 (0.21845), which differs from Table 2's value (0.21812).

Watch

Extended reading notes

Core claim

The paper's central claim is that a single model can handle heterophilic user-item interactions and localized high-order structure at the same time, and that doing so yields consistently better recommendations than existing graph- and hypergraph-based collaborative filtering. WaveHDNN encodes users and items through two separate channels. The Heterophily-aware Collaborative Encoder uses an equivariant operator, built on hypergraph diffusion, to send different messages to different nodes within a hyperedge, then layer-normalizes and adds the input back to preserve identity. The Multi-scale Group-wise Structure Encoder applies wavelet hypergraph convolution, $X^{(l+1)} = \Theta \Lambda \Theta' X^{(l)} W + X^{(l)}$, so the spread of information can be tuned per scale instead of by stacking more layers. A cross-view contrastive loss keeps the two encoders' embeddings consistent, and ranking is optimized with a Bayesian personalized ranking loss. On the three datasets the model reports wins on all six metrics for every baseline, with the largest relative margin, 7.24%, on Steam NDCG@20; ablations show removing either encoder lowers Recall@40 and NDCG@40 on all three datasets.

Load-bearing premise

The claim that WaveHDNN consistently outperforms every baseline rests on the fairness and accuracy of the reported comparisons, which the paper supports only with point estimates and no tuning budget or code.

Editorial extensions

If this is right

Hypergraph collaborative filtering can capture both heterophily and multi-scale locality without very deep GNN stacks, because the wavelet channel tunes information spread by scale instead of by layer count.
On the three datasets tested, the reported gains are consistent across all six ranking metrics, not concentrated in one cutoff, suggesting the design improves overall ranking quality rather than only recall.
Ablations say both channels matter: removing either encoder drops Recall@40 and NDCG@40 on Amazon-Books, Steam, and Yelp, so the improvement is not carried by a single component.
The largest relative margin appears on Steam at 7.24% NDCG@20, so the method may be most valuable in denser interaction data where group-wise structure is more informative.

Reading between the lines

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

An independent re-run with matched hyperparameter budgets and per-seed variances would be the decisive test; the paper reports point estimates only, so the ordinal claim is exactly what needs reproduction.
The wavelet scale parameters could be adapted to temporal recommender settings, where local neighborhoods drift over time, but the paper only evaluates static datasets.
The equivariant heterophily encoder is not limited to recommendation; the same message-differentiation mechanism could apply to hypergraph node classification or link prediction in graphs with mixed homophily.

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.

Desk Editor's Note

read the letter

WaveHDNN is a reasonable engineering contribution: it takes the equivariant operator from ED-HNN, the wavelet hypergraph convolution from [14], and cross-view InfoNCE, and applies the combination to collaborative filtering. That is a real combination, not a trivial one, and the ablation in Table 5 does what an ablation should: removing either encoder clearly hurts. The paper is also honest about its building blocks, which I appreciate.

But the central claim, that WaveHDNN “consistently outperforms all baselines,” isn’t yet supported cleanly. Table 3 has at least one impossible entry: SHT’s NDCG@20 is 0.02647, far below its own NDCG@10 of 0.07583, and AutoCF’s NDCG@10 (0.07632) is higher than its NDCG@20 (0.07615). Those look like transposition or rounding errors, but they make the whole Steam table hard to trust. Table 5 also gives WaveHDNN’s Amazon Recall@40 as 0.21845 while Table 2 says 0.21812; small, but it shows the numbers weren’t double-checked. And there are no standard deviations or significance tests anywhere. Some of the reported margins, like the 1.59% gain on Steam NDCG@10, are exactly the size that run-to-run noise can explain. With no code, no data splits, and no per-baseline tuning budget, a reader can’t reproduce the setup.

The stress-test note is therefore on target, and I agree with the conditional verdict. Still, I’d hesitate to call these load-bearing flaws in the sense of a wrong method. The architecture is sensible, the ablation supports both components, and the direction of the gains is consistent across datasets. The problems are reporting and reproducibility problems, not evidence of a broken approach. What’s missing is a cleaner, statistically grounded comparison and artifacts. I also would have liked to see ED-HNN and the wavelet HGCN themselves as baselines, since the novelty claim rests on beating the source models, not just LightGCN and AutoCF.

The paper deserves a serious referee. It is not a desk reject; it is a revise-and-resubmit case. I’d send it out, require corrected tables, standard deviations, and released code or splits, and ask for a comparison against the two cited source components. If the numbers hold up after that, it’s a modest but useful result for hypergraph-based CF.

Referee Report

5 major / 6 minor

Summary. This paper proposes WaveHDNN, a hypergraph-based collaborative filtering model with two parallel encoders: a Heterophily-aware Collaborative Encoder inspired by ED-HNN, and a Multi-scale Group-wise Structure Encoder using wavelet hypergraph convolution. The two views are aligned with an InfoNCE-style cross-view contrastive loss, and the final score is computed from the concatenated embeddings. The authors report experiments on Amazon-Books, Steam, and Yelp, comparing against LightGCN, SGL, DHCF, HCCF, SHT, and AutoCF across Recall@10/20/40 and NDCG@10/20/40, and they report ablations removing each encoder.

Significance. If the empirical claims hold, WaveHDNN would be a state-of-the-art hypergraph collaborative filtering method, with relative gains over the second-best baseline of roughly 3–7% on the three datasets. The architectural combination of heterophily-aware message passing and multi-scale wavelet filtering is well motivated, and the ablations in Table 5 provide initial evidence that both components contribute. The paper also transparently builds on external works (ED-HNN and wavelet HGCN). However, the strength of the paper currently rests entirely on the reliability of the reported numbers, which are not statistically supported and contain internal inconsistencies. The paper ships no code, no per-baseline tuning budgets, and no variance estimates, all of which are needed to verify the central empirical superiority claim.

major comments (5)

[Table 3 (Steam)] The SHT row reports NDCG@20 = 0.02647, which is about one-third of its NDCG@10 = 0.07583, while Recall@20 is 0.12687. With non-negative relevance gains, NDCG@20 cannot drop that far below NDCG@10; this entry is internally inconsistent and suggests a data transcription error. Because Tables 2–4 carry the central claim of consistent superiority, this error must be corrected and the affected conclusions re-verified.
[Tables 2 and 5] The WaveHDNN row gives Amazon Recall@40 = 0.21812 in Table 2 but 0.21845 in Table 5, and NDCG@40 = 0.12484 in Table 2 but 0.12486 in Table 5, for what is described as the same model and the same 5-run average. The discrepancy implies that at least one of these tables does not reflect the final configuration, and it undermines confidence in the reproducibility of the headline numbers.
[§4.1 and Tables 2–4] The paper reports only the mean over 5 runs with no standard deviations, confidence intervals, or significance tests. The smallest reported margin in the main tables is 1.59% (Steam NDCG@10: WaveHDNN 0.07754 vs. AutoCF 0.07632), which is likely within run-to-run variation; the claim that WaveHDNN "consistently outperforms all baselines across six evaluation metrics" is therefore not statistically established. Please provide per-metric variance and paired significance tests over the 5 runs.
[§4.1–4.2] The experimental protocol is under-specified: the paper does not state the embedding dimension, learning rate, batch size, regularization, number of layers for each model, the temperature τ in Eq. (4), or the per-baseline hyperparameter tuning budget. Without these details and without code or data splits, a reader cannot assess whether the reported margins reflect a fair comparison or favorable configuration. This is a load-bearing gap for an empirical state-of-the-art claim.
[§3.2, Eq. (3)] The wavelet-based hypergraph convolution is not fully defined: the wavelet basis Θ, its inverse Θ′, and the filter matrix Λ are introduced without specifying how the hypergraph wavelet transform is computed or how Λ is parametrized and learned. The reference to [14] is not sufficient because the paper modifies the layer with a residual concatenation, and the reader cannot implement or reproduce the proposed encoder from the text. Please supply the complete formulation or an unambiguous pointer to the exact construction used.

minor comments (6)

[§4.5] There are typos: "datsets" should be "datasets", and the section opening "mong" should be "Among".
[Tables 2–4] The model name appears as "W aveHDNN" in several table rows; please remove the spurious space.
[Table captions] The phrase "The top mark performance is highlighted" is awkward; consider "The best performance is in bold" and similar idiomatic wording.
[§4.2] AutoCF is listed as a baseline but no citation number is provided in the text; please add the corresponding reference.
[Eq. (4)] The summation bounds IX and LX in Eq. (4) are not defined, and the layer index l is ambiguous; please specify the ranges and clarify how layer-wise contrastive losses are aggregated.
[Header] The line "Preprint submitted to Technical Report" is not appropriate for a journal submission and should be removed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is an empirical benchmark comparison, not a derivation that reduces to its inputs.

full rationale

The paper's central claim is empirical: WaveHDNN is reported to outperform all baselines on held-out test splits of Amazon-Books, Steam, and Yelp, with ablations showing each encoder contributes (Section 4.3, Tables 2-5). No equation in the methodology reduces to a fitted version of the conclusion. The Heterophily-aware Collaborative Encoder follows the equivariant operator of ED-HNN [13], and the Multi-scale Group-wise Structure Encoder follows the wavelet hypergraph convolution of Sun et al. [14]; both are external prior works, not self-authored uniqueness theorems. The contrastive loss (Eq. 4) and BPR loss (Eq. 5) are standard objective functions that do not encode the benchmark outcome. No parameter is fitted to a subset of the evaluation metrics and then reported as a prediction of those same metrics. The self-citations in the reference list (e.g., [67], [74]) are contextual and are not load-bearing for the superiority claim. The internal inconsistencies noted in the reader's take, such as the SHT NDCG@20 value in Table 3 (0.02647, below its NDCG@10 of 0.07583) and the differing WaveHDNN Amazon Recall@40 values in Tables 2 and 5, are reproducibility and reporting concerns about the empirical claim, not evidence that the derivation is circular. Therefore, the circularity score is 0.

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

The central empirical claim rests on standard recommender assumptions (implicit feedback, BPR ranking), on the transfer of two external architectures into the collaborative filtering setting, and on several unstated hyperparameter choices. No new particles, forces, or post hoc entities are introduced; the two encoders are architectural modules, not invented entities.

free parameters (4)

number of HGCN layers per encoder = 3 (best of 1-5)
Section 4.5 reports best performance at 3 layers; the final model configuration used for Tables 2-4 depends on this search.
embedding dimension = not stated in text
Section 4.5 scans [8,16,64,32,128] but never reports the selected dimension used for Tables 2-5.
InfoNCE temperature tau = not reported
Equation (4) includes tau but the text gives no value or tuning range.
wavelet filter matrix Lambda = learned (not specified)
Equation (3) defines Lambda as a diagonal weight filter, but the paper does not state initialization, constraints, or relation to the hypergraph spectrum.

assumptions (4)

domain assumption Wavelet basis and inverse wavelet are defined on the hypergraph as in [14], enabling localized convolution via Eq. (3).
Section 3.2 adopts ref [14] without derivation; if the wavelet construction does not carry over to the user-item hypergraph, the Multi-scale encoder's mechanism claim fails.
domain assumption Heterophily in user-item interactions can be captured by applying MLP then HConv with a residual equivariant operator (Eqs. 1-2), following ED-HNN [13].
Section 3.1 states this without proof; no experiment measures heterophily directly.
domain assumption The BPR pairwise ranking loss (Eq. 5) is the correct surrogate for recommendation quality on implicit feedback.
Section 3.3 uses BPR without discussion; standard in CF, but an assumption about user preference.
ad hoc to paper Cross-view contrastive loss (Eq. 4) aligns embeddings from the two encoders without discarding complementary information.
The paper asserts in §3.3 that large distances indicate redundancy or conflict, but this is not established; the objective could also collapse usefully distinct views.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Hypergraph Diffusion for High-Order Recommender Systems." pith.science (2026). https://pith.science/paper/BJUE6WLD

@misc{pith2026250116722,
  author       = {Pith},
  title        = {Pith review of: Hypergraph Diffusion for High-Order Recommender Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BJUE6WLD}},
  note         = {Machine review of arXiv:2501.16722}
}

read the original abstract

Recommender systems rely on Collaborative Filtering (CF) to predict user preferences by leveraging patterns in historical user-item interactions. While traditional CF methods primarily focus on learning compact vector embeddings for users and items, graph neural network (GNN)-based approaches have emerged as a powerful alternative, utilizing the structure of user-item interaction graphs to enhance recommendation accuracy. However, existing GNN-based models, such as LightGCN and UltraGCN, often struggle with two major limitations: an inability to fully account for heterophilic interactions, where users engage with diverse item categories, and the over-smoothing problem in multi-layer GNNs, which hinders their ability to model complex, high-order relationships. To address these gaps, we introduce WaveHDNN, an innovative wavelet-enhanced hypergraph diffusion framework. WaveHDNN integrates a Heterophily-aware Collaborative Encoder, designed to capture user-item interactions across diverse categories, with a Multi-scale Group-wise Structure Encoder, which leverages wavelet transforms to effectively model localized graph structures. Additionally, cross-view contrastive learning is employed to maintain robust and consistent representations. Experiments on benchmark datasets validate the efficacy of WaveHDNN, demonstrating its superior ability to capture both heterophilic and localized structural information, leading to improved recommendation performance.

Figures

Figures reproduced from arXiv: 2501.16722 by the authors.

**Figure 1.** An illustration of WaveHDNN. 3.1. Heterophily-aware Collaborative Encoding Building on the principles of ED-HNN [13], we begin by learning user and item representations through the modeling of heterophilic patterns within collaborative hypergraphs. It is important to note that different types or categories of items are often grouped by specific users, a phenomenon commonly observed in user-item collaborative hype… view at source ↗

**Figure 2.** Effects of no. HGCN layers [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Representing Higher-Order Networks: A Survey of Graph-Based Frameworks
cs.SI 2026-03 unverdicted novelty 4.0 of 10

A survey organizing higher-order network formalisms into four families with a master comparison table, plus ~17 new superhypergraph-style definitions whose only supporting theorems are well-definedness checks.

Reference graph

Works this paper leans on

75 extracted references · 67 canonical work pages · cited by 1 Pith paper

[14]

X. Sun, H. Yin, B. Liu, H. Chen, J. Cao, Y. Shao, N. Q. Viet Hung, Heterogeneous hypergraph embedding for graph classification, in: WSDM, 2021, pp. 725–733

work page 2021
[1]

Huang, H

C. Huang, H. Xu, Y. Xu, P. Dai, L. Xia, M. Lu, L. Bo, H. Xing, X. Lai, Y. Ye, Knowledge-aware coupled graph neural network for social recommendation, in: AAAI, 2021, pp. 4115–4122

work page 2021
[2]

Y. Liu, Q. Liu, Y. Tian, C. Wang, Y. Niu, Y. Song, C. Li, Concept-aware denoising graph neural network for micro-video recommendation, in: CIKM, 2021, pp. 1099–1108

work page 2021
[3]

Huang, X

C. Huang, X. Wu, X. Zhang, C. Zhang, J. Zhao, D. Yin, N. V. Chawla, Online purchase prediction via multi-scale modeling of behavior dynamics, in: KDD, 2019, pp. 2613–2622

work page 2019
[4]

H. Ma, H. Yang, M. R. Lyu, I. King, Sorec: social recommenda- tion using probabilistic matrix factorization, in: CIKM, 2008, pp. 931–940

work page 2008
[5]

X. He, K. Deng, X. Wang, Y. Li, Y. Zhang, M. Wang, Light- gcn: Simplifying and powering graph convolution network for recommendation, in: SIGIR, 2020, pp. 639–648

work page 2020
[6]

K. Mao, J. Zhu, X. Xiao, B. Lu, Z. Wang, X. He, Ultragcn: ultra simplification of graph convolutional networks for recom- mendation, in: CIKM, 2021, pp. 1253–1262

work page 2021
[7]

L. Xia, C. Huang, Y. Xu, J. Zhao, D. Yin, J. Huang, Hyper- graph contrastive collaborative filtering, in: SIGIR, 2022, pp. 70–79

work page 2022

Show all 75 references

[8]

T.T.Nguyen, T.C.Phan, Q.V.H.Nguyen, K.Aberer, B.Stan- tic, Maximal fusion of facts on the web with credibility guar- antee, Information Fusion 48 (2019) 55–66

2019
[9]

T. T. Nguyen, T. T. Nguyen, T. T. Nguyen, B. Vo, J. Jo, Q. V. H. Nguyen, Judo: Just-in-time rumour detection in streaming social platforms, Information Sciences 570 (2021) 70–93

2021
[10]

T. T. Nguyen, M. Weidlich, H. Yin, B. Zheng, Q. H. Nguyen, Q. V. H. Nguyen, Factcatch: Incremental pay-as-you-go fact checking with minimal user effort, in: Proceedings of the 43rd International ACM SIGIR Conference on Research and Devel- opment in Information Retrieval, 2020,...

2020
[11]

T. T. Nguyen, T. T. Huynh, P. L. Nguyen, A. W.-C. Liew, H. Yin, Q. V. H. Nguyen, A survey of machine unlearning, arXiv preprint arXiv:2209.02299 (2022)

2022 arXiv
[12]

T. T. Nguyen, N. Quoc Viet Hung, T. T. Nguyen, T. T. Huynh, T. T. Nguyen, M. Weidlich, H. Yin, Manipulating recommender systems: A survey of poisoning attacks and countermeasures, ACM Computing Surveys 57 (2024) 1–39

2024
[13]

P. Wang, S. Yang, Y. Liu, Z. Wang, P. Li, Equivari- ant hypergraph diffusion neural operators, arXiv preprint arXiv:2207.06680 (2022)

2022 arXiv
[15]

T. T. Nguyen, Q. V. Hung Nguyen, M. Weidlich, K. Aberer, Result selection and summarization for web table search, in: 2015 IEEE 31st International Conference on Data Engineering, 2015, pp. 231–242

2015
[16]

N. T. Tam, M. Weidlich, D. C. Thang, H. Yin, N. Q. V. Hung, Retaining data from streams of social platforms with mini- mal regret, in: Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence, IJCAI-17, 2017, pp. 2850–2856. URL: https://doi.org...

2017 doi
[17]

N. T. Tam, M. Weidlich, B. Zheng, H. Yin, N. Q. V. Hung, B. Stantic, From anomaly detection to rumour detection using data streams of social platforms, Proceedings of the VLDB Endowment 12 (2019) 1016–1029

2019
[18]

T. T. Nguyen, T. D. Hoang, M. T. Pham, T. T. Vu, T. H. Nguyen, Q.-T. Huynh, J. Jo, Monitoring agriculture areas with satellite images and deep learning, Applied Soft Computing 95 (2020) 106565

2020
[19]

T. T. Nguyen, M. Weidlich, H. Yin, B. Zheng, Q. V. H. Nguyen, B.Stantic, Userguidanceforefficientfactchecking, Proceedings of the VLDB Endowment 12 (2019) 850–863

2019
[20]

N. T. Tam, H. T. Trung, H. Yin, T. Van Vinh, D. Sakong, B. Zheng, N. Q. V. Hung, Entity alignment for knowledge graphs with multi-order convolutional networks, TKDE 34 (2022) 4201–4214

2022
[21]

T. T. Nguyen, M. T. Pham, T. T. Nguyen, T. T. Huynh, Q. V. H. Nguyen, T. T. Quan, et al., Structural representa- tion learning for network alignment with self-supervised anchor links, Expert Systems with Applications 165 (2021) 113857

2021
[22]

J. S. Breese, D. Heckerman, C. Kadie, Empirical analysis of predictive algorithms for collaborative filtering, arXiv preprint arXiv:1301.7363 (2013)

2013 arXiv
[23]

Sarwar, G

B. Sarwar, G. Karypis, J. Konstan, J. Riedl, Item-based collab- orative filtering recommendation algorithms, in: WWW, 2001, pp. 285–295

2001
[24]

Y. Shi, M. Larson, A. Hanjalic, Collaborative filtering beyond the user-item matrix: A survey of the state of the art and future challenges, CSUR 47 (2014) 1–45

2014
[25]

Koren, R

Y. Koren, R. Bell, C. Volinsky, Matrix factorization techniques for recommender systems, Computer 42 (2009) 30–37

2009
[26]

Y.Koren, Factorizationmeetstheneighborhood: amultifaceted collaborative filtering model, in: KDD, 2008, pp. 426–434

2008
[27]

Zhang, W

S. Zhang, W. Wang, J. Ford, F. Makedon, Learning from in- complete ratings using non-negative matrix factorization, in: SDM, 2006, pp. 549–553

2006
[28]

X. He, L. Liao, H. Zhang, L. Nie, X. Hu, T.-S. Chua, Neural collaborative filtering, in: WWW, 2017, pp. 173–182

2017
[29]

Hidasi, A

B. Hidasi, A. Karatzoglou, L. Baltrunas, D. Tikk, Session- based recommendations with recurrent neural networks, arXiv preprint arXiv:1511.06939 (2015)

2015 arXiv
[30]

W.-C. Kang, J. McAuley, Self-attentive sequential recommen- dation, in: ICDM, 2018, pp. 197–206

2018
[31]

D. C. Thang, H. T. Dat, N. T. Tam, J. Jo, N. Q. V. Hung, K. Aberer, Nature vs. nurture: Feature vs. structure for graph neural networks, PRL 159 (2022) 46–53

2022
[32]

C. T. Duong, T. T. Nguyen, H. Yin, M. Weidlich, T. S. Mai, K. Aberer, Q. V. H. Nguyen, Efficient and effective multi-modal queries through heterogeneous network embed- ding, IEEE Transactions on Knowledge and Data Engineering 34 (2022) 5307–5320

2022
[33]

N. Q. V. Hung, D. C. Thang, N. T. Tam, M. Weidlich, K. Aberer, H. Yin, X. Zhou, Answer validation for generic crowdsourcing tasks with minimal efforts, The VLDB Journal 26 (2017) 855–880

2017
[34]

Q. V. H. Nguyen, C. T. Duong, T. T. Nguyen, M. Weidlich, K. Aberer, H. Yin, X. Zhou, Argument discovery via crowd- sourcing, The VLDB Journal 26 (2017) 511–535

2017
[35]

T. T. Huynh, C. T. Duong, T. T. Nguyen, V. T. Van, A. Sattar, H. Yin, Q. V. H. Nguyen, Network alignment with holistic embeddings, TKDE 35 (2021) 1881–1894

2021
[36]

C. T. Duong, T. T. Nguyen, T.-D. Hoang, H. Yin, M. Wei- dlich, Q. V. H. Nguyen, Deep mincut: Learning node embed- dings from detecting communities, Pattern Recognition (2022) 109126

2022
[37]

T. T. Nguyen, T. C. Phan, H. T. Pham, T. T. Nguyen, J. Jo, 7 Q. V. H. Nguyen, Example-based explanations for streaming fraud detection on graphs, Information Sciences 621 (2023) 319–340

2023
[38]

H. T. Trung, T. Van Vinh, N. T. Tam, J. Jo, H. Yin, N. Q. V. Hung, Learning holistic interactions in lbsns with high-order, dynamic, and multi-role contexts, TKDE 35 (2022) 5002–5016

2022
[39]

2766–2771

M.Gori, A.Pucci, V.Roma, I.Siena, Itemrank: Arandom-walk based scoring algorithm for recommender engines., in: IJCAI, volume 7, 2007, pp. 2766–2771

2007
[40]

G. Jeh, J. Widom, Simrank: a measure of structural-context similarity, in: KDD, 2002, pp. 538–543

2002
[41]

X. Wang, X. He, M. Wang, F. Feng, T.-S. Chua, Neural graph collaborative filtering, in: SIGIR, 2019, pp. 165–174

2019
[42]

C. Feng, Z. Liu, S. Lin, T. Q. Quek, Attention-based graph convolutionalnetworkforrecommendationsystem, in: ICASSP, 2019, pp. 7560–7564

2019
[43]

R. Ying, R. He, K. Chen, P. Eksombatchai, W. L. Hamilton, J. Leskovec, Graph convolutional neural networks for web-scale recommender systems, in: KDD, 2018, pp. 974–983

2018
[44]

W. Fan, Y. Ma, Q. Li, Y. He, E. Zhao, J. Tang, D. Yin, Graph neural networks for social recommendation, in: WWWW, 2019, pp. 417–426

2019
[45]

Zhang, X

J. Zhang, X. Shi, S. Zhao, I. King, Star-gcn: Stacked and reconstructed graph convolutional networks for recommender systems, arXiv preprint arXiv:1905.13129 (2019)

2019 arXiv
[46]

L. Chen, L. Wu, R. Hong, K. Zhang, M. Wang, Revisiting graph based collaborative filtering: A linear residual graph con- volutional network approach, in: AAAI, volume 34, 2020, pp. 27–34

2020
[47]

B. Zhao, H. van der Aa, T. T. Nguyen, Q. V. H. Nguyen, M. Weidlich, Eires: Efficient integration of remote data in event stream processing, in: SIGMOD, 2021, pp. 2128–2141

2021
[48]

T. T. Nguyen, T. C. Phan, M. H. Nguyen, M. Weidlich, H. Yin, J. Jo, Q. V. H. Nguyen, Model-agnostic and diverse explana- tions for streaming rumour graphs, Knowledge-Based Systems 253 (2022) 109438

2022
[49]

T. T. Nguyen, T. T. Huynh, H. Yin, M. Weidlich, T. T. Nguyen, T. S. Mai, Q. V. H. Nguyen, Detecting rumours with latency guarantees using massive streaming data, The VLDB Journal (2022) 1–19

2022
[50]

T. T. Huynh, M. H. Nguyen, T. T. Nguyen, P. L. Nguyen, M. Weidlich, Q. V. H. Nguyen, K. Aberer, Efficient integration of multi-order dynamics and internal dynamics in stock move- ment prediction, in: Proceedings of the Sixteenth ACM Inter- national Conference on Web Search and...

2023
[51]

Y. Feng, H. You, Z. Zhang, R. Ji, Y. Gao, Hypergraph neural networks, in: AAAI, volume 33, 2019, pp. 3558–3565

2019
[52]

S. Ji, Y. Feng, R. Ji, X. Zhao, W. Tang, Y. Gao, Dual channel hypergraph collaborative filtering, in: KDD, 2020, pp. 2020– 2029

2020
[53]

L. Xia, C. Huang, C. Zhang, Self-supervised hypergraph trans- former for recommender systems, in: KDD, 2022, pp. 2100– 2109

2022
[54]

T. E. Jendal, T.-H. Le, H. W. Lauw, M. Lissandrini, P. Dolog, K. Hose, Hypergraphs with attention on reviews for explainable recommendation, in: ECIR, 2024, pp. 230–246

2024
[55]

M. Yang, Z. Liu, L. Yang, X. Liu, C. Wang, H. Peng, P. S. Yu, Unified pretraining for recommendation via task hypergraphs, in: WSDM, 2024, pp. 891–900

2024
[56]

J. Yu, H. Yin, J. Li, Q. Wang, N. Q. V. Hung, X. Zhang, Self- supervised multi-channel hypergraph convolutional network for social recommendation, in: WWW, 2021, pp. 413–424

2021
[57]

Nguyen Thanh, N

T. Nguyen Thanh, N. D. K. Quach, T. T. Nguyen, T. T. Huynh, V. H. Vu, P. L. Nguyen, J. Jo, Q. V. H. Nguyen, Poisoning gnn- basedrecommendersystemswithgenerativesurrogate-basedat- tacks, ACM Transactions on Information Systems 41 (2023) 1–24

2023
[58]

Q. V. H. Nguyen, T. Nguyen Thanh, Z. Miklós, K. Aberer, Reconciling schema matching networks through crowdsourc- ing, EAI Endorsed Transactions on Collaborative Computing 1 (2014) e2

2014
[59]

Q. V. H. Nguyen, T. T. Nguyen, V. T. Chau, T. K. Wijaya, Z. Miklós, K. Aberer, A. Gal, M. Weidlich, Smart: A tool for analyzingandreconcilingschemamatchingnetworks, in: ICDE, 2015, pp. 1488–1491

2015
[60]

D. C. Thang, N. T. Tam, N. Q. V. Hung, K. Aberer, An evalua- tion of diversification techniques, in: DEXA, 2015, pp. 215–231

2015
[61]

Q. V. H. Nguyen, S. T. Do, T. T. Nguyen, K. Aberer, Tag-based paper retrieval: minimizing user effort with diversity awareness, in: International Conference on Database Systems for Advanced Applications, 2015, pp. 510–528

2015
[62]

N. Q. V. Hung, M. Weidlich, N. T. Tam, Z. Miklós, K. Aberer, A. Gal, B. Stantic, Handling probabilistic integrity constraints in pay-as-you-go reconciliation of data models, Information Sys- tems 83 (2019) 166–180

2019
[63]

J. Xu, X. Sun, Z. Zhang, G. Zhao, J. Lin, Understanding and improving layer normalization, NeurIPS 32 (2019)

2019
[64]

A. v. d. Oord, Y. Li, O. Vinyals, Representation learning with contrastive predictive coding, arXiv preprint arXiv:1807.03748 (2018)

2018 arXiv
[65]

Rendle, C

S. Rendle, C. Freudenthaler, Z. Gantner, L. Schmidt-Thieme, Bpr: Bayesian personalized ranking from implicit feedback, in: UAI, 2009, pp. 452–461

2009
[66]

C. Yang, W. Yuan, L. Qu, T. T. Nguyen, Pdc-frs: Privacy- preservingdatacontributionforfederatedrecommendersystem, in: International Conference on Advanced Data Mining and Applications, Springer, 2024, pp. 65–79

2024
[67]

Sakong, V

D. Sakong, V. H. Vu, T. T. Huynh, P. Le Nguyen, H. Yin, Q. V. H. Nguyen, T. T. Nguyen, Higher-order knowledge- enhanced recommendation with heterogeneous hypergraph multi-attention, Information Sciences 680 (2024) 121165

2024
[68]

T. T. Huynh, T. B. Nguyen, P. L. Nguyen, T. T. Nguyen, M. Weidlich, Q. V. H. Nguyen, K. Aberer, Fast-fedul: A training-free federated unlearning with provable skew resilience, in: Joint European Conference on Machine Learning and Knowledge Discovery in Databases, Springer, 202...

2024
[69]

T. T. Huynh, T. B. Nguyen, T. T. Nguyen, P. L. Nguyen, H. Yin, Q. V. H. Nguyen, T. T. Nguyen, Certified unlearning for federated recommendation, ACM Transactions on Information Systems (2025)

2025
[70]

T. T. Nguyen, T. T. Nguyen, T. H. Nguyen, H. Yin, T. T. Nguyen, J. Jo, Q. V. H. Nguyen, Isomorphic graph embed- ding for progressive maximal frequent subgraph mining, ACM Transactions on Intelligent Systems and Technology 15 (2023) 1–26

2023
[71]

T. T. Nguyen, Z. Ren, T. T. Nguyen, J. Jo, Q. V. H. Nguyen, H.Yin, Portablegraph-basedrumourdetectionagainst multi-modal heterophily, Knowledge-Based Systems 284 (2024) 111310

2024
[72]

T. T. Nguyen, T. T. Huynh, Z. Ren, T. T. Nguyen, P. L. Nguyen, H. Yin, Q. V. H. Nguyen, Privacy-preserving explain- able ai: a survey, Science China Information Sciences 68 (2025) 111101

2025
[73]

M. T. Pham, T. T. Huynh, T. T. Nguyen, T. T. Nguyen, T. T. Nguyen, J. Jo, H. Yin, Q. V. Hung Nguyen, A dual benchmark- ing study of facial forgery and facial forensics, CAAI Transac- tions on Intelligence Technology (????)
[74]

D. D. A. Nguyen, M. H. Nguyen, P. L. Nguyen, J. Jo, H. Yin, T. T. Nguyen, Multi-task learning of heterogeneous hypergraph representations in lbsns, in: International Conference on Ad- vanced Data Mining and Applications, Springer, 2024, pp. 161– 177

2024
[75]

T. T. Nguyen, T. T. Nguyen, M. Weidlich, J. Jo, Q. V. H. Nguyen, H. Yin, A. W.-C. Liew, Handling low homophily in rec- ommender systems with partitioned graph transformer, IEEE Transactions on Knowledge and Data Engineering (2024). 8

2024

Pith tools

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

[1] [14]

X. Sun, H. Yin, B. Liu, H. Chen, J. Cao, Y. Shao, N. Q. Viet Hung, Heterogeneous hypergraph embedding for graph classification, in: WSDM, 2021, pp. 725–733

work page 2021

[2] [1]

Huang, H

C. Huang, H. Xu, Y. Xu, P. Dai, L. Xia, M. Lu, L. Bo, H. Xing, X. Lai, Y. Ye, Knowledge-aware coupled graph neural network for social recommendation, in: AAAI, 2021, pp. 4115–4122

work page 2021

[3] [2]

Y. Liu, Q. Liu, Y. Tian, C. Wang, Y. Niu, Y. Song, C. Li, Concept-aware denoising graph neural network for micro-video recommendation, in: CIKM, 2021, pp. 1099–1108

work page 2021

[4] [3]

Huang, X

C. Huang, X. Wu, X. Zhang, C. Zhang, J. Zhao, D. Yin, N. V. Chawla, Online purchase prediction via multi-scale modeling of behavior dynamics, in: KDD, 2019, pp. 2613–2622

work page 2019

[5] [4]

H. Ma, H. Yang, M. R. Lyu, I. King, Sorec: social recommenda- tion using probabilistic matrix factorization, in: CIKM, 2008, pp. 931–940

work page 2008

[6] [5]

X. He, K. Deng, X. Wang, Y. Li, Y. Zhang, M. Wang, Light- gcn: Simplifying and powering graph convolution network for recommendation, in: SIGIR, 2020, pp. 639–648

work page 2020

[7] [6]

K. Mao, J. Zhu, X. Xiao, B. Lu, Z. Wang, X. He, Ultragcn: ultra simplification of graph convolutional networks for recom- mendation, in: CIKM, 2021, pp. 1253–1262

work page 2021

[8] [7]

L. Xia, C. Huang, Y. Xu, J. Zhao, D. Yin, J. Huang, Hyper- graph contrastive collaborative filtering, in: SIGIR, 2022, pp. 70–79

work page 2022