Pith. sign in

REVIEW 3 major objections 5 minor 38 references

Distance-aware Self-adaptive Graph Convolution for Fine-grained Hierarchical Recommendation

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read SAGCN claims distance-weighted layer mixing improves GCN recommendations by over 5% on Yelp and 5.58% on ML_1M's Recall@10.

desk verdict A simple LightGCN variant with distance-modulated layer weights shows consistent but small gains on four datasets; the core mechanism is not isolated because no ablation separates distance weighting from final-layer-only readout and the residual gate. read the letter →

arxiv 2505.09590 v1 pith:LTUFDG5U submitted 2025-05-14 cs.IR

classification cs.IR
keywords RecommenderSystemsCollaborativeFilteringGraphNeuralNetworkInformationAggregationDistance-basedMulti-layerembeddingRepresentationspaceuniformityConvolutional
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 proposes SAGCN, a graph-convolutional collaborative-filtering model that changes how multiple convolution layers are combined. Instead of averaging the embeddings from all layers as LightGCN does, it adaptively weights each new layer's embedding by the distance between that embedding and the previous layer's embedding: the farther the freshly aggregated embedding has moved, the more it is allowed to update the representation. The paper claims this finer-grained aggregation captures hierarchical information more precisely, relieves over-similarity between neighboring nodes, and makes the representation space more uniform, which translates into better top-K recommendations. On four public datasets the model beats the strongest baselines on almost every Recall and NDCG metric, including relative gains of more than 5% on Yelp and 5.58% on ML_1M's Recall@10. If the claim holds, a small, parameter-light change to layer mixing yields consistent accuracy gains across datasets.

What carries the argument

The central mechanism is the distance-weighted interpolation of consecutive layer embeddings. Concretely, $w_{\text{new}} = \alpha\cdot\log(1 + \beta\cdot\text{dist}) / (1 + \alpha\cdot\log(1 + \beta\cdot\text{dist}))$ and $w_{\text{old}} = 1 / (1 + \alpha\cdot\log(1 + \beta\cdot\text{dist}))$, where $\text{dist}$ measures the distance between the node's embedding before and after neighborhood aggregation using a chosen metric. This identity carries the argument because it makes the update size self-regulating: nodes whose aggregated neighbors are very different from their current embedding get updated more, while nodes whose neighbors are already similar stay closer to where they are. Repeated over layers and used with the final embedding only, the mechanism is claimed to keep the representation space from collapsing toward near-identical node vectors while still propagating collaborative signal.

What would settle it

Run SAGCN with the distance weight replaced by a fixed constant (say $w_{\text{new}} = 0.5$) while keeping the final-layer-only representation and the same $\alpha$/ $\beta$ tuning procedure; if Recall@10 and NDCG@10 stay within the reported margins, the distance computation is not the causal factor. Complementarily, run LightGCN with a final-layer-only representation and a tuned fixed blend weight: matching SAGCN's numbers would show the gains come from the layer-selection change, not the distance signal.

Watch

Extended reading notes

Core claim

In the paper's own terms, the discovery is that layer-to-layer distance can serve as a self-adaptive weight for hierarchical aggregation in GCN-based recommendation. At each convolution step the model keeps the old embedding $e^{(k)}_{u,\text{old}}$ and computes the freshly aggregated neighbor embedding $e^{(k+1)}_{u,\text{new}}$; it scores the old embedding as 1 and the new one as $\alpha\cdot\log(1 + \beta\cdot\text{dist}(e^{(k)}_{u,\text{old}}, e^{(k+1)}_{u,\text{new}}))$, normalizes the two scores into weights $w_{\text{old}}$ and $w_{\text{new}}$, and forms $e^{(k+1)}_u = w_{\text{old}}\cdot e^{(k)}_{u,\text{old}} + w_{\text{new}}\cdot e^{(k+1)}_{u,\text{new}}$. Larger distances therefore push the representation further toward the aggregated information, while smaller distances preserve the old representation, which the paper interprets as preventing nodes from becoming too similar. The final representation is the embedding produced by the last aggregation step, not an average over all layers. Tested with Euclidean, cosine, and KL-divergence distances, the model achieves consistent gains over the compared baselines, with Euclidean distance the most stable choice.

Load-bearing premise

The assumption that the distance-based weights in equations (8)–(13) are what cause the reported accuracy gains, rather than the model's other differences from LightGCN—using only the final layer's embedding and per-dataset tuning of $\alpha$ and $\beta$—since no ablation isolates the distance signal.

Editorial extensions

If this is right

  • Across Yelp, ML_1M, Gowalla, and CiteULike, SAGCN beats the best baseline on all but a handful of the Recall@K and NDCG@K metrics, with the largest gains on Yelp (Recall@10 +5.29%, NDCG@10 +5.96%) and ML_1M (Recall@10 +5.58%).
  • The distance metric is decisive: Euclidean and KL-divergence distances beat LightGCN on the four datasets, while cosine distance improves only CiteULike and degrades the other three—so the choice of metric is part of the model's effectiveness.
  • Performance is stable across a wide range of the $\alpha$ hyperparameter (0.5 to 5, peaking near 1.5 on three datasets), suggesting the method does not hinge on a fragile parameter setting.
  • Using the final layer's embedding as the node representation—rather than averaging all layers—is part of the design, and the authors argue the approach alleviates over-smoothing and over-similarity common in multi-layer GCNs.
  • The same aggregation principle is proposed as transferable to other graph-based tasks such as social-network analysis and knowledge-graph completion.

Reading between the lines

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

  • One consequence the paper leaves untested is that the gains might come as much from discarding the early-layer average (LightGCN's final embedding) as from the distance signal itself; a variant that keeps the all-layer average but applies the distance weights would separate the two effects.
  • Because the weight $w_{\text{new}}$ is a monotone, bounded function of distance, the mechanism behaves like a learnable interpolator; a natural extension would let $\alpha$ and $\beta$ be trained parameters instead of per-dataset hyperparameters, making the method fully self-tuning.
  • The cosine-distance failure suggests the informative quantity may be the magnitude of embedding change rather than its direction; a test using angular distance with the same magnitude scaling as Euclidean would clarify the role of each component.
  • If the improvement is genuinely driven by uniformity of the representation space, the distance-weighted aggregation should be measurable directly: comparing the pairwise distance distribution of user and item embeddings against LightGCN's would offer a structural check on the claimed mechanism.
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

3 major / 5 minor

Summary. The manuscript introduces SAGCN, a graph convolutional recommendation model that adaptively weights the combination of a node's pre-aggregation embedding and its neighbor-aggregated embedding at each convolution layer, where the weights are computed from a distance (Euclidean, cosine, or KL) between hierarchical layer representations. The final user and item embeddings are taken from the last aggregation layer only, and the model is trained with a BPR loss. Experiments on Yelp2018, ML-1M, Gowalla, and CiteULike compare SAGCN against eight baselines and report relative gains, including more than 5% on Yelp and 5.58% in Recall@10 on ML-1M. The paper also studies the effect of the three distance metrics and the sensitivity of the alpha hyperparameter, and it releases code.

Significance. If the central claim were established, SAGCN would provide a simple, interpretable improvement over LightGCN's fixed layer averaging, which is a widely used baseline; the release of code and the systematic comparison across four datasets are strengths. The distance-metric comparison (RQ2) is a useful exploration, and the parameter sensitivity analysis shows robustness. However, the paper does not isolate the distance-weighting mechanism from two other architectural changes (final-layer-only readout and a large self-residual), and the 'adaptive' weights operate in a nearly constant regime because beta is calibrated to scale distances to about 1e-2. The empirical claims also lack error bars and significance tests, with several reported improvements below 1%. The contribution is therefore conditional on additional ablations and statistical validation.

major comments (3)
  1. [Section 3.3, Eqs. (8)-(13); Tables 3-4] The central claim that distance-based adaptive weighting improves accuracy is not isolated from other architectural changes. Relative to LightGCN, SAGCN changes (i) the readout from averaging all layer embeddings to using only the final layer's embedding (Section 3.3: 'the model in this study uses the embedding derived from the final aggregation as the final embedding'), (ii) the per-layer update by adding a self-residual term w_old * e^(k)_old with w_old approximately 0.985 at the reported operating point (alpha=1.5, beta=1, dist approximately 1e-2, from Eqs. (8)-(11)), and (iii) the distance-dependent weighting itself. Tables 3 and 4 compare only the full SAGCN with the default LightGCN; no ablation tests a LightGCN variant with final-layer-only readout, a SAGCN variant with a fixed w_new approximately 0.015, or a SAGCN variant without the self-residual. Consequently, the reported gains cannot be attributed specifically to the distance-based adaptivity, which is the paper's stated contribution.
  2. [Section 4.4 and Eqs. (8)-(11)] The adaptive mechanism is self-referential and operates in a near-constant regime. In Eqs. (9)-(11), score_new is computed from a distance defined on the embeddings that the weights then update, creating a feedback loop whose convergence and gradient behavior are not analyzed. More importantly, beta is explicitly chosen to rescale all distances to about 1e-2 (Sections 4.3 and 4.4), and with score_old fixed at 1 (Eq. (8)), score_new is at most about 0.05 for the tested alpha range; the resulting w_new is always below about 0.05, so the gate is nearly constant across the entire training trajectory. The robustness shown in Figure 6 across alpha in [0.5, 5] is therefore consistent with the model behaving like a fixed residual connection rather than with adaptivity being essential. The authors should report the empirical distribution of dist and w_new during training and include a fixed-weight control.
  3. [Section 4.1.4 and Table 4] There is no statistical support for the 'significant improvements' claim. The evaluation protocol (Section 4.1) does not mention multiple seeds, error bars, confidence intervals, or significance tests; Tables 3 and 4 report single point estimates. Several reported improvements are very small (Table 4, CiteULike recall@20 +0.23%, recall@50 +0.08%, Gowalla ndcg@10 +0.49%). In addition, alpha and beta are selected per dataset on the evaluation datasets (Sections 4.1.4 and 4.3), so the comparison is not a fixed-model evaluation. Without variance estimates or a separate validation protocol for hyperparameters, the phrase 'significant improvements' is not supported.
minor comments (5)
  1. [Table 2 and Section 4.2.1] The column header says 'density' but the reported values are sparsity (e.g., ML-1M has 1,000,209 / (6,041 * 3,707) approximately 4.47% interaction density, not 95.53%); please correct the header and reconcile the discussion of 'lowest sparsity' with the table values.
  2. [Section 3.3] The vectors P and Q are never precisely defined; specify whether the distance in Eq. (9) is computed between e^(k) and e^(k+1) for each node or between layer embedding matrices, and for which node subsets.
  3. [Figure 6] The x-axis labels in Figure 6 render as garbled tokens (e.g., '/s945/s61/s48/s46/s53'), making the parameter sensitivity results unreadable; please regenerate the figure with proper alpha-value labels.
  4. [Section 4.1.4] The early stopping description is ambiguous ('after 5 epochs of performance degradation'); state whether training stops when Recall@20 has not improved for 5 consecutive epochs.
  5. [Section 3.2 and Section 5] Section 3.2 says '64-bit vector' but should say '64-dimensional vector'; Section 5's claim of an 'average improvement of 5%' is not supported by all metrics in Table 4 (e.g., CiteULike Recall@20 +0.23%), so please qualify the average or specify the metrics included.

Circularity Check

0 steps flagged · score 2.0 of 10

No load-bearing circularity: the distance-gated aggregation is a defined architecture and the reported gains are measured against external baselines; the only self-citation is non-load-bearing.

full rationale

The central mechanism (Eqs. 8-13) defines the aggregation weights as w_new = alpha*log(1+beta*dist) and w_old = 1/(1+w_new), so the stated property that larger distances produce larger updates is an analytic consequence of the definition rather than a derived prediction. The performance improvements are empirical results from comparisons with external baselines such as LightGCN, SGL, NCL, and GDE; they are not constructed by equating the model's output with its input. Hyperparameters alpha and beta are tuned per dataset, with beta scaled to keep distances near 1e-2, but this is ordinary hyperparameter selection rather than fitting a parameter to a subset and then predicting a closely related quantity. The absence of an ablation that isolates the distance gate from the final-layer-only readout and the residual connection is a correctness or experimental-design concern, not a circularity. The only self-citation, reference [17] in the introduction, is used merely as an example of an application domain for social recommendation and carries no weight in the derivation. Therefore no circular step is present; the score of 2 reflects the minor non-load-bearing self-citation, while the central claim has independent empirical content.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The method rests on a small set of heuristics: distance between layer embeddings is a valid information signal; the log-scaled score with tuned alpha and beta is the right transformation; BPR is the right objective. Alpha and beta are fitted per dataset and per distance metric, and several standard hyperparameters (layer count, lambda) are not reported. No new entities are introduced.

free parameters (4)
  • alpha = 1.5 for Yelp, ML_1M, Gowalla; 1.2 for CiteULike
    Tuned over {0.5,0.8,1,1.2,1.5,2,5}; scales the effect of distance in score_new (Eq. 9).
  • beta = 1 (Euclidean), 0.001 (cosine), 100 (KL)
    Empirically chosen to force distances to about 1e-2 so weights take a desired magnitude (Section 4.4).
  • lambda (BPR L2 coefficient) = not reported
    Regularization weight in Eq. (15); no value or tuning procedure is given, so replication depends on unstated assumptions.
  • number of GCN layers = not reported
    Layer depth affects which distances are computed and how much smoothing occurs; implementation details omit it.
assumptions (5)
  • standard math Symmetric normalized adjacency matrix D^{-1/2} A D^{-1/2} is an appropriate convolution operator (Eq. 3).
    Adopted from standard GCN and LightGCN; not derived or validated here.
  • ad hoc to paper Nodes that are more similar exchange less information, so inter-layer distance can weight aggregation (Section 3.3).
    This information-entropy-inspired premise is the conceptual basis for Eqs. 8-11 but is not formalized or tested.
  • domain assumption Alignment and uniformity of representation space determine recommendation quality (from [18]).
    Used to motivate representation-space optimization; no uniformity or alignment metric is measured in the experiments.
  • domain assumption BPR pairwise loss with sampled negatives is the correct objective (Eq. 15).
    Standard in recommender systems; adopted from [29], not questioned in the paper.
  • domain assumption KL divergence can be applied directly to raw embedding vectors (Eqs. 6-7).
    KL is defined for positive distributions; the paper does not state how negative or zero embedding entries are handled, so the KL variant may be ill-posed for arbitrary embeddings.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Distance-aware Self-adaptive Graph Convolution for Fine-grained Hierarchical Recommendation." pith.science (2026). https://pith.science/paper/LTUFDG5U

@misc{pith2026250509590,
  author       = {Pith},
  title        = {Pith review of: Distance-aware Self-adaptive Graph Convolution for Fine-grained Hierarchical Recommendation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LTUFDG5U}},
  note         = {Machine review of arXiv:2505.09590}
}
read the original abstract

Graph Convolutional Networks (GCNs) are widely used to improve recommendation accuracy and performance by effectively learning the representations of user and item nodes. However, two major challenges remain: (1) the lack of further optimization in the graph representation structure and (2) insufficient attention given to the varying contributions of different convolutional layers.This paper proposes SAGCN, a distance-based adaptive hierarchical aggregation method that refines the aggregation process through differentiated representation metrics. SAGCN introduces a detailed approach to multilayer information aggregation and representation space optimization, enabling the model to learn hierarchical embedding weights based on the distance between hierarchical representations. This innovation allows for more precise cross-layer information aggregation, improves the model's ability to capture hierarchical embeddings, and optimizes the representation space structure. Additionally, the objective loss function is refined to better align with recommendation tasks.Extensive experiments conducted on four real-world datasets demonstrate significant improvements, including over a 5% increase on Yelp and a 5.58% increase in Recall@10 on the ML_1M dataset.

Figures

Figures reproduced from arXiv: 2505.09590 by the authors.

Figure 1
Figure 1. Overall framework of SAGCN recall@ 1 0 recall@ 20 recall@ 50 0. 00 0. 02 0. 04 0. 06 0. 08 0.1 0 0.1 2 0.1 4 0.1 6 0.1 8 0. 20 Yelp LightGCN SAGCN recall@ 1 0 recall@ 20 recall@ 50 0. 00 0. 05 0.1 0 0.1 5 0. 20 0. 25 0.30 0.3 5 0. 40 0. 45 0. 50 ML_1 M LightGCN SAGCN recall@ 1 0 recall@ 20 recall@ 50 0. 00 0. 05 0.1 0 0.1 5 0. 20 0. 25 0.30 0.3 5 0. 40 CiteULike LightGCN SAGCN recall@ 1 0 recall@ 20 recall@ 50 0. 00… view at source ↗
Figure 2
Figure 2. Comparison of Recall with benchmark model performance ndcg@ 1 0 ndcg@ 20 ndcg@ 50 0. 00 0. 02 0. 04 0. 06 0. 08 0.1 0 Yelp LightGCN SAGCN ndcg@ 1 0 ndcg@ 20 ndcg@ 50 0. 00 0. 05 0.1 0 0.1 5 0. 20 0. 25 0.30 ML_1 M LightGCN SAGCN ndcg@ 1 0 ndcg@ 20 ndcg@ 50 0. 00 0. 02 0. 04 0. 06 0. 08 0.1 0 0.1 2 0.1 4 0.1 6 Gowalla LightGCN SAGCN ndcg@ 1 0 ndcg@ 20 ndcg@ 50 0. 00 0. 02 0. 04 0. 06 0. 08 0.1 0 0.1 2 0.1 4 0.1 6 0.1… view at source ↗
Figure 3
Figure 3. Comparison of NDCG with benchmark model performance Tao Huang et al.: Preprint submitted to Elsevier Page 15 of 14 [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Comparison of recall Performance of Different Distance Metrics with Benchmark Model ndcg@ 1 0 ndcg@ 20 ndcg@ 50 0. 04 0. 06 0. 08 0.1 0 Yelp LightGCN_test SAGCN-O_test SAGCN-C_test SAGCN-K_test ndcg@ 1 0 ndcg@ 20 ndcg@ 50 0. 26 0. 28 0.3 0 0.3 2 0.3 4 ML_1 M LightGCN_t…
Figure 5
Figure 5. Figure 5: Comparison of NDCG Performance of Different Distance Metrics with Benchmark Model Yelp ML_1 M Gowalla 0. 00 0. 05 0.1 0 0.1 5 0. 20 0. 25 0.3 0 0.3 5 (a) NDCG@ 1 0                         Yelp ML_1 M Gowalla 0. 00 0. 05 0.1 0 0.1 5 0. 20 0. 25 0…
Figure 6
Figure 6. Figure 6: Hyperparametric analysis of 𝛼 on Yelp, ML_1M, Gowalla dataset Tao Huang et al.: Preprint submitted to Elsevier Page 16 of 14 [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 35 canonical work pages

  1. [1]

    Recommendation systems: Algorithms, challenges, metrics, and business opportunities.Applied Sciences, 10(21):7748, 2020

    Zeshan Fayyaz, Mahsa Ebrahimian, Dina Nawara, Ahmed Ibrahim, and Rasha Kashef. Recommendation systems: Algorithms, challenges, metrics, and business opportunities.Applied Sciences, 10(21):7748, 2020. Tao Huang et al.:Preprint submitted to ElsevierPage 12 of 14 SAGCN

  2. [2]

    Issues and solutions in deep learning-enabled recommendation systems within the e-commerce field.Applied Sciences, 12(21):11256, 2022

    Rand Jawad Kadhim Almahmood and Adem Tekerek. Issues and solutions in deep learning-enabled recommendation systems within the e-commerce field.Applied Sciences, 12(21):11256, 2022

  3. [3]

    Contemporary Recommendation Systems on Big Data and Their Applications: A Survey

    Anonymous. A survey on modern recommendation system based on big data.arXiv preprint arXiv:2206.02631, 2021

  4. [4]

    Neuralcollaborativefiltering

    XiangnanHe,LiziLiao,HanwangZhang,LiqiangNie,XiaHu,andTat-SengChua. Neuralcollaborativefiltering. InProceedingsofthe26th international conference on world wide web, pages 173–182, 2017

  5. [5]

    Attentive collaborative filtering: Multimedia recommendationwithitem-andcomponent-levelattention

    Jingyuan Chen, Hanwang Zhang, Xiangnan He, Liqiang Nie, Wei Liu, and Tat-Seng Chua. Attentive collaborative filtering: Multimedia recommendationwithitem-andcomponent-levelattention. InProceedingsofthe40thInternationalACMSIGIRconferenceonResearchand Development in Information Retrieval, pages 335–344, 2017

  6. [6]

    Lightgcn: Simplifying and powering graph convolution network for recommendation

    Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. Lightgcn: Simplifying and powering graph convolution network for recommendation. InProceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval, pages 639–648, 2020

  7. [7]

    Disentangled graph collaborative filtering

    Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. Disentangled graph collaborative filtering. InProceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, 2020

  8. [8]

    How powerful are k-hop message passing graph neural networks

    Jiarui Feng, Yixin Chen, Fuhai Li, Anindya Sarkar, and Muhan Zhang. How powerful are k-hop message passing graph neural networks. Advances in Neural Information Processing Systems, 35:4776–4790, 2022

Show all 38 references
  1. [9]

    Localaugmentation for graph neural networks

    SongtaoLiu,RexYing,HanzeDong,LanqingLi,TingyangXu,YuRong,PeilinZhao,JunzhouHuang,andDinghaoWu. Localaugmentation for graph neural networks. InInternational conference on machine learning, pages 14054–14072. PMLR, 2022

  2. [10]

    Robust mid-pass filtering graph convolutional networks

    Jincheng Huang, Lun Du, Xu Chen, Qiang Fu, Shi Han, and Dongmei Zhang. Robust mid-pass filtering graph convolutional networks. In Proceedings of the ACM Web Conference 2023, pages 328–338, 2023

  3. [11]

    Mixed graph contrastive network for semi-supervised node classification.ACM Transactions on Knowledge Discovery from Data, 2024

    Xihong Yang, Yiqi Wang, Yue Liu, Yi Wen, Lingyuan Meng, Sihang Zhou, Xinwang Liu, and En Zhu. Mixed graph contrastive network for semi-supervised node classification.ACM Transactions on Knowledge Discovery from Data, 2024

  4. [12]

    Graphneuralnetworksfornaturallanguage processing: A survey.Foundations and Trends®in Machine Learning, 16(2):119–328, 2023

    LingfeiWu,YuChen,KaiShen,XiaojieGuo,HanningGao,ShuchengLi,JianPei,BoLong,etal. Graphneuralnetworksfornaturallanguage processing: A survey.Foundations and Trends®in Machine Learning, 16(2):119–328, 2023

  5. [13]

    OnwhichnodesdoesGCNfail?enhancingGCNfromthenodeperspective

    JinchengHuang,JialieShen,XiaoshuangShi,andXiaofengZhu. OnwhichnodesdoesGCNfail?enhancingGCNfromthenodeperspective. InForty-first International Conference on Machine Learning, 2024

  6. [14]

    Nie-gcn: Neighbor item embedding-aware graph convolutional network for recommendation.IEEE Transactions on Systems, Man, and Cybernetics: Systems, 54(5):2810–2821, 2024

    Yi Zhang, Yiwen Zhang, Dengcheng Yan, Qiang He, and Yun Yang. Nie-gcn: Neighbor item embedding-aware graph convolutional network for recommendation.IEEE Transactions on Systems, Man, and Cybernetics: Systems, 54(5):2810–2821, 2024

  7. [15]

    Multi-behavior recommendation with graph convolutional networks

    Bo wen Jin, Chen Gao, Xiang nan He, De peng Jin, and Yong Li. Multi-behavior recommendation with graph convolutional networks. In Proceedingsofthe43rdInternationalACMSIGIRConferenceonResearchandDevelopmentinInformationRetrieval,pages659–668,2020

  8. [16]

    Alexander Felfernig and Bartosz Gula. An empirical study on consumer behavior in the interaction with knowledge-based recommender applications.InThe8thIEEEInternationalConferenceonE-CommerceTechnologyandThe3rdIEEEInternationalConferenceonEnterprise Computing, E-Commerce, and E...

  9. [17]

    Meta-relation assisted knowledge-aware coupled graph neural network for recommendation.Information Processing & Management, 60(3):103353, 2023

    Yao Chang, Wei Zhou, Hai ni Cai, Wei Fan, Lin feng Hu, and Jun hao Wen. Meta-relation assisted knowledge-aware coupled graph neural network for recommendation.Information Processing & Management, 60(3):103353, 2023

  10. [18]

    Understanding contrastive representation learning through alignment and uniformity on the hypersphere

    Tong zhou Wang and Phillip Isola. Understanding contrastive representation learning through alignment and uniformity on the hypersphere. InInternational Conference on Machine Learning, pages 9929–9939. PMLR, 2020

  11. [19]

    Lightgcn:Simplifyingandpoweringgraphconvolution networkforrecommendation

    XiangnanHe,KuanDeng,XiangWang,YanLi,YongdongZhang,andMengWang. Lightgcn:Simplifyingandpoweringgraphconvolution networkforrecommendation. InProceedingsofthe43rdInternationalACMSIGIRConferenceonResearchandDevelopmentinInformation Retrieval, pages 639–648, 2020

  12. [20]

    InProceedings of the Web Conference 2021, pages 593–601, 2021

    JianingSun,ZhaoyueCheng,SabaZuberi,FelipePérez,andMaksimsVolkovs.Hgcf:Hyperbolicgraphconvolutionnetworksforcollaborative filtering. InProceedings of the Web Conference 2021, pages 593–601, 2021

  13. [21]

    Hrcf: Enhancing collaborative filtering via hyperbolic geometric regularization

    Menglin Yang, Min Zhou, Jiahong Liu, Defu Lian, and Irwin King. Hrcf: Enhancing collaborative filtering via hyperbolic geometric regularization. InProceedings of the ACM web conference 2022, pages 2462–2471, 2022

  14. [22]

    Adaptive popularity debiasing aggregator for graph collaborative filtering.(2023).A APPENDICES A, 1, 2023

    Huachi Zhou, Hao Chen, Junnan Dong, Daochen Zha, Chuang Zhou, and Xiao Huang. Adaptive popularity debiasing aggregator for graph collaborative filtering.(2023).A APPENDICES A, 1, 2023

  15. [23]

    Model-agnostic counterfactual reasoning for eliminating popularity bias in recommender system

    Tianxin Wei, Fuli Feng, Jiawei Chen, Ziwei Wu, Jinfeng Yi, and Xiangnan He. Model-agnostic counterfactual reasoning for eliminating popularity bias in recommender system. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, pages 1791–1800, 2021

  16. [24]

    Investigating accuracy-novelty performance for graph-based collaborative filtering

    M Zhao, L Wu, Y Liang, L Chen, J Zhang, Q Deng, K Wang, X Shen, T Lv, and R Wu. Investigating accuracy-novelty performance for graph-based collaborative filtering. arxiv 2022.arXiv preprint arXiv:2204.12326

  17. [25]

    Self-guided learning to denoise for robust recommendation

    Yunjun Gao, Yuntao Du, Yujia Hu, Lu Chen, Xinjun Zhu, Ziquan Fang, and Baihua Zheng. Self-guided learning to denoise for robust recommendation. InProceedings of the 45th international ACM SIGIR conference on research and development in information retrieval, pages 1412–1422, 2022

  18. [26]

    Learning robust recommenders through cross-model agreement

    Yu Wang, Xin Xin, Zaiqiao Meng, Joemon M Jose, Fuli Feng, and Xiangnan He. Learning robust recommenders through cross-model agreement. InProceedings of the ACM web conference 2022, pages 2015–2025, 2022

  19. [27]

    Springer, 2002

    Ian T Jolliffe.Principal component analysis. Springer, 2002

  20. [28]

    Graphmae: Self-supervised masked graph autoencoders

    Zhenyu Hou, Xiao Liu, Yukuo Cen, Yuxiao Dong, Hongxia Yang, Chunjie Wang, and Jie Tang. Graphmae: Self-supervised masked graph autoencoders. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD ’22, page 594–604, New York, NY, USA, 2022. ...

  21. [29]

    Bayesian personalized ranking from implicit feedback

    Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and LB Schmidt-Thieme. Bayesian personalized ranking from implicit feedback. In Proc. of Uncertainty in Artificial Intelligence, pages 452–461, 2014

  22. [30]

    Field-aware factorization machines for ctr prediction

    Yuchin Juan, Yong Zhuang, Wei-Sheng Chin, and Chih-Jen Lin. Field-aware factorization machines for ctr prediction. InProceedings of the 10th ACM Conference on Recommender Systems, pages 43–50, 2016. Tao Huang et al.:Preprint submitted to ElsevierPage 13 of 14 SAGCN

  23. [31]

    A neural collaborative filtering model with interaction-based neighborhood

    Ting Bai, Ji-Rong Wen, Jun Zhang, and Wayne Xin Zhao. A neural collaborative filtering model with interaction-based neighborhood. In Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, pages 1979–1982, 2017

  24. [32]

    Neural graph collaborative filtering

    Xiang Wang, Xiang nan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. Neural graph collaborative filtering. InProceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 165–174, 2019

  25. [33]

    Disentangledgraphcollaborativefiltering

    XiangWang,HongyeJin,AnZhang,XiangnanHe,TongXu,andTat-SengChua. Disentangledgraphcollaborativefiltering. InProceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 1001–1010, 2020

  26. [34]

    Less is more: Reweighting important spectral graph features for recommendation

    Shao wen Peng, Kazunari Sugiyama, and Tsunenori Mine. Less is more: Reweighting important spectral graph features for recommendation. InProceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 1273–1282, 2022

  27. [35]

    Self-supervised graph learning for recommendation

    Jian can Wu, Xiang Wang, Fu li Feng, Xiang nan He, Liang Chen, Jian xun Lian, and Xing Xie. Self-supervised graph learning for recommendation. InProceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 726–735, 2021

  28. [36]

    Improving graph collaborative filtering with neighborhood-enriched contrastive learning

    Zi han Lin, Chang xin Tian, Yu peng Hou, and Wayne Xin Zhao. Improving graph collaborative filtering with neighborhood-enriched contrastive learning. InProceedings of the ACM Web Conference 2022, pages 2320–2329, 2022

  29. [37]

    Understandingthedifficultyoftrainingdeepfeedforwardneuralnetworks

    XavierGlorotandYoshuaBengio. Understandingthedifficultyoftrainingdeepfeedforwardneuralnetworks. InProceedingsoftheThirteenth International Conference on Artificial intelligence and statistics, pages 249–256. JMLR Workshop and Conference Proceedings, 2010

  30. [38]

    DiederikP.KingmaandJimmyBa.Adam:Amethodforstochasticoptimization.InYoshuaBengioandYannLeCun,editors,3rdInternational Conference on Learning Representations, 2015. Tao Huang et al.:Preprint submitted to ElsevierPage 14 of 14 SAGCN Figure 1:Overall framework of SAGCN r ecall@10r...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.