Pith. sign in

REVIEW 3 major objections 4 minor 36 references

Metapath-based Hyperbolic Contrastive Learning for Heterogeneous Graph Embedding

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · 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 assigning each metapath of a heterogeneous graph its own hyperbolic space with a learned curvature, and then applying a contrastive loss that separates embeddings of different metapaths, yields better node…

desk verdict A modest extension of MSGAT whose central contrastive loss (Eq. 10) is mis-specified to repel positives and attract negatives — with no code, the reported gains cannot be credited to the proposed method. read the letter →

arxiv 2506.16754 v1 pith:KJZTUTV3 submitted 2025-06-20 cs.LG cs.AIcs.SI

Jongmin Park , Seunghoon Han , Won-Yong Shin , Sungsu Lim This is my paper · ORCID
classification cs.LGcs.AIcs.SI
keywords heterogeneousgraphembeddinghyperbolicspacemetapathcontrastivelearningneuralnetworkpower-lawstructurePoincaréballnodeclassification
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

MHCL is designed around a simple premise: a heterogeneous graph contains several different power-law structures, one per metapath, and a single hyperbolic space cannot faithfully encode all of them at once. The paper proposes to give each metapath its own Poincaré ball with a learnable negative curvature, aggregate metapath instances inside each ball with attention, and then align all metapath embeddings in a unified hyperbolic space where a contrastive loss keeps different metapaths apart. On the IMDB, DBLP, ACM, and LastFM datasets, the paper reports that MHCL outperforms state-of-the-art Euclidean and hyperbolic heterogeneous GNNs on node classification, clustering, and link prediction, and its ablations show that both the multi-space geometry and the contrastive term contribute to the gains. The significance, if the results hold, is that geometry should be matched to each semantic relation rather than shared across the whole graph.

What carries the argument

The load-bearing object is a family of metapath-specific Poincaré balls $\mathbb{D}^{d,c_\phi}$, one per metapath $\phi$, each with its own learnable negative curvature $-c_\phi$ ($c_\phi > 0$). Inside each ball, metapath instances are aggregated by intra-hyperbolic attention (Eqs. 4-6) with multi-head extension (Eq. 7); embeddings are then aligned into a unified hyperbolic space via the shared tangent space at the origin (Eq. 8) and combined by inter-hyperbolic attention (Eqs. 12-14). The contrastive loss $L_{\mathrm{hyp}}$ (Eq. 10), using hyperbolic distance $d_c$ (Eq. 11), is what the paper claims enforces discriminability of metapath embeddings.

What would settle it

Training MHCL with the exact loss of Eq. (10) and observing that same-metapath embeddings grow more distant and different-metapath embeddings more similar would show the described mechanism is not what the formula implements; alternatively, zeroing the contrastive term and finding no significant drop on the reported datasets would falsify the claimed benefit of contrastive separation.

Watch

Extended reading notes

Core claim

The central discovery claimed is that the power-law 'shape' of a heterogeneous graph is not a single global property but is metapath-specific, so an embedding model should use multiple hyperbolic geometries whose curvatures adapt per metapath. MHCL samples metapath instances via BFS up to a maximum length, embeds them in metapath-specific Poincaré balls, aggregates them with parameterized attention to form metapath embeddings, maps those into a unified hyperbolic space through the shared tangent space at the origin, and then applies a hyperbolic contrastive loss so that embeddings of different metapaths are pushed apart while each metapath stays consistent with its own embedding. The paper presents experiments on four real-world datasets and a synthetic graph, reporting consistent outperformance over 13 baselines, plus ablation and visualization evidence that the contrastive loss separates metapath clusters.

Load-bearing premise

The paper's described contrastive mechanism depends on the implemented loss being the negative logarithm of the ratio in Eq. (10); as written, minimizing that equation would push positive metapath pairs apart instead of together, and the paper does not supply the form actually implemented.

Editorial extensions

If this is right

  • If MHCL's claim is right, heterogeneous graph models should treat curvature as a per-metapath parameter rather than a global constant, since different metapaths induce measurably different power-law distributions.
  • The contrastive separation of metapath embeddings should transfer to any downstream task that relies on distinguishing semantic relations, including recommendation and knowledge-graph completion, not just the four datasets tested.
  • Because MHCL samples metapaths with BFS up to a maximum length instead of relying on hand-crafted metapath definitions, the approach removes the need for domain-specific metapath engineering while keeping linear time complexity in the maximum metapath length.
  • The paper's ablations indicate that geometric capacity and contrastive regularization contribute independently, since removing either component degrades performance across all tested datasets.

Reading between the lines

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

  • A natural next step the paper does not pursue: use the per-metapath Gromov $\delta$-hyperbolicity values (which the paper computes for motivating examples) to initialize or regularize the learned curvatures, potentially improving convergence and stability.
  • A testable claim implicit in the multi-space premise: on a heterogeneous graph whose metapaths all induce nearly identical power-law distributions (similar Gromov deltas), MHCL should converge to a single effective curvature; this could be checked on synthetic graphs with controlled metapath distributions.
  • The contrastive separation of metapath embeddings is measured only in the unified space; one could probe whether the learned per-metapath curvatures themselves encode semantic differences, for example by inspecting curvature values per metapath across datasets.
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 / 4 minor

Summary. The paper proposes MHCL, a heterogeneous graph embedding framework that uses multiple metapath-specific hyperbolic spaces with learnable curvatures to capture diverse power-law structures, plus a metapath-based hyperbolic contrastive loss intended to separate embeddings from different metapaths. The method is evaluated on node classification, clustering, and link prediction across IMDB, DBLP, ACM, LastFM, and a synthetic graph, with ablations, robustness tests, hyperparameter sensitivity, complexity experiments, and t-SNE visualizations. The central claimed contribution is the contrastive loss in Eq. (10), which the text states should minimize the distance between same-metapath embeddings and maximize the distance between different-metapath embeddings.

Significance. If the method works as described, the multi-hyperbolic-space architecture with per-metapath learnable curvatures is a meaningful extension of prior hyperbolic heterogeneous GNNs such as MSGAT, and the empirical study is reasonably broad: four real-world datasets plus a synthetic robustness study, multiple downstream tasks, ablations, and complexity measurements. The paper does not provide code, and the central contrastive-loss equation is not a valid contrastive objective as printed. The positive experimental results and the t-SNE visualizations are consistent with a corrected InfoNCE-style loss, but the exact implemented objective must be stated before the claimed improvement over MSGAT can be evaluated. No circularity issue arises from the use of the authors' earlier HHGAT and MSGAT as baselines, since they are trained and evaluated on held-out tasks.

major comments (3)
  1. [Section IV-D.3, Eq. (10); Algorithm 1 line 16; Eq. (18)] The printed contrastive loss L_hyp is not a valid contrastive objective. As written, L_hyp = sum_phi exp(-d_c(hat_h^phi_v, p(h^phi_v))/tau) / sum_{j(phi) in Phi \ {phi}} exp(-d_c(hat_h^phi_v, hat_h^{j(phi)}_v)/tau) has no negative logarithm. Minimizing this ratio decreases the numerator when the positive distance d_c(hat_h^phi_v, p(h^phi_v)) increases, and decreases the denominator when negative distances decrease, so gradient descent on Eq. (10) as printed would separate an anchor from its positive sample and pull it toward negative samples. This directly contradicts the stated goal in Section IV-D.3 and the abstract. Since Eq. (18) adds lambda * L_hyp to the task loss, the printed training objective is anti-contrastive. Please correct Eq. (10) to the negative log of the ratio (or an equivalent InfoNCE form), update Algorithm 1 and all references to it, and state whether the implemented loss matches the corrected equation. Because no code is provided, the current text does not allow the reader to verify which objective was actually optimized.
  2. [Section IV-D.3, Eq. (10); Algorithm 1] The formula in Eq. (10) also contains an inconsistent use of the temperature tau: in the numerator tau appears inside the distance as p(h^phi_v)/tau, while in the denominator it scales the distance, i.e., d_c(hat_h^phi_v, hat_h^{j(phi)}_v)/tau. Additionally, the loss is written for a single node v with no aggregation over the target node set V_t, although Algorithm 1 takes a single node v as input. Please specify how L_hyp is aggregated over nodes (e.g., summed or averaged over V_t) and correct the placement of tau. These details are needed to reproduce the reported loss values and the experimental results.
  3. [Section IV-G and Section V-K, Fig. 9] The claim that the overall time complexity increases linearly with respect to the maximum metapath length l is not supported by the stated complexity analysis. Metapath instance sampling via breadth-first search from a node with branching factor b and maximum length l enumerates O(b^l) instances in the worst case (up to type constraints), not O(l). The empirical near-linear growth in Fig. 9 may reflect the specific degree distributions of the ACM dataset, but the text should restate the worst-case complexity in terms of the branching factor and explain why the observed training time grows approximately linearly on this dataset. As written, the complexity analysis in Section IV-G is inaccurate.
minor comments (4)
  1. [Throughout] There are several typos and wording issues, including 'postive' and 'metpath-specific' in Section IV-D, 'Additonally' in Section V-K, and 'metapfath' in the Introduction. A careful proofread is needed.
  2. [Section III-B, Eq. (11)] The hyperbolic distance formula in Eq. (11) uses 1/sqrt(|c|) although c is already defined to be positive in Definition 4; this is needlessly confusing and can be written as 1/sqrt(c).
  3. [Section IV-D.1, Eq. (8)] The alignment step in Eq. (8) applies a linear transformation W2 in the tangent space at the origin, but the text does not specify whether this tangent-space coordinate representation is shared across different curvatures; a brief clarification of the coordinate convention would improve reproducibility.
  4. [Tables III-V] The main result tables are extremely wide and difficult to read in the provided format; splitting them by dataset or grouping competitors differently would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: MHCL's components are trained end-to-end and evaluated on held-out tasks; self-citations are transparent and not load-bearing.

full rationale

The paper is an empirical systems paper. No target quantity is derived and then refitted: metapath embeddings, curvatures, attention weights, and the contrastive objective are learned from data and assessed on held-out node classification, clustering, and link prediction. The multi-hyperbolic-space design is explicitly built on the authors' prior MSGAT/HHGAT work (Section V-G says 'MHCL_Single and MHCL_w/o cont correspond to our earlier works HHGAT [15] and MSGAT [27]'), but this is disclosed and used as a baseline/ablation, not as an unverified premise that forces the conclusion. The central novel component, the metapath hyperbolic contrastive loss (Eq. 10), is not obtained by fitting the downstream metric; its claimed benefit is tested empirically. The apparent missing negative logarithm in Eq. (10) is a correctness/verification issue (the printed objective would repel positives and attract negatives, contrary to Section IV-D.3), but it is not a circular reduction of the paper's conclusions to its inputs. Self-citations to HHGAT/MSGAT are normal prior-work references and are not load-bearing in the derivation. Therefore the paper warrants a low circularity score.

Assumptions & free parameters 7 free parameters · 4 assumptions · 2 invented entities

The paper rests on the assumption that a single hyperbolic space per metapath can encode that metapath's structure, on the alignment of tangent spaces across curvatures, and on a contrastive loss that is mis-specified in Eq. (10). These are not derived from first principles; they are modeling choices fitted and evaluated empirically.

free parameters (7)
  • Curvature c_phi of each metapath-specific Poincare ball = learned during training, not reported numerically
    The paper's premise is that each metapath needs its own curvature; these are fit to the task loss and are the central learnable geometric parameters.
  • Unified hyperbolic space curvature c = not specified in paper
    Used in the alignment and contrastive loss (Eqs. 8-11) but the value or whether it is fixed or learned is not stated; it affects the scale of the loss and geometry.
  • Contrastive temperature tau = not reported
    Temperature in Eq. (10) controls the sharpness of the contrastive distribution; no value or sensitivity analysis is provided.
  • Contrastive loss weight lambda = not reported, stated as ranging from 0 to 1
    The total loss L = L_task + lambda L_hyp; the relative weighting is set by hand and not analyzed in the sensitivity study.
  • Maximum metapath length l = 4, 5, 4, 3 for IMDB, DBLP, ACM, LastFM
    Set per dataset; the paper does sensitivity analysis on IMDB only.
  • Number of attention heads K = 8
    Chosen without reported tuning across datasets.
  • Embedding dimensions d and d' = d=128, d'=64
    Hand-chosen; sensitivity is shown on IMDB only.
assumptions (4)
  • domain assumption Each metapath-specific structure in a heterogeneous graph can be characterized by a single constant negative curvature.
    Stated in Section I and used to justify one Poincare ball per metapath; no theoretical bound or empirical validation that one curvature suffices for a metapath's degree distribution.
  • standard math Tangent spaces at the origin of Poincare balls with different curvatures can be identified, so log/exp maps allow meaningful alignment between metapath-specific spaces.
    Mathematically the tangent space at 0 is R^d for all curvatures, but the composition exp_0^c(log_0^{c_phi}(x)) is a nonlinear radial rescaling; treating this as semantic alignment is a modeling assumption used in Eq. (8).
  • domain assumption Breadth-first sampling of all metapath instances up to length l captures the semantic neighborhoods needed for embedding.
    Section IV-B assumes BFS over the full graph up to a bounded length yields a representative set P_v; the paper provides no convergence analysis.
  • ad hoc to paper The contrastive loss in Eq. (10) drives same-metapath embeddings together and different-metapath embeddings apart.
    As written, minimization of Eq. (10) would repel positives; the intended behavior requires a negative logarithm and is not specified correctly in the paper.
invented entities (2)
  • Metapath-specific hyperbolic space D^{n,c_phi}
    purpose: Assigned to each metapath to capture its hypothesized distinct power-law structure; used as the embedding space for all instances of that metapath.
    The paper introduces these as learnable-curvature per-metapath spaces but provides no independent evidence, such as linking learned curvature to measured Gromov delta of the subgraph, that they correspond to actual structural differences.
  • Unified hyperbolic space D^{d,c} for aggregating metapath embeddings
    purpose: Serves as the common space for attention-based aggregation and contrastive comparisons after alignment from metapath-specific spaces.
    A modeling convenience; the choice of curvature c is not justified or tied to data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Metapath-based Hyperbolic Contrastive Learning for Heterogeneous Graph Embedding." pith.science (2026). https://pith.science/paper/KJZTUTV3

@misc{pith2026250616754,
  author       = {Pith},
  title        = {Pith review of: Metapath-based Hyperbolic Contrastive Learning for Heterogeneous Graph Embedding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KJZTUTV3}},
  note         = {Machine review of arXiv:2506.16754}
}
read the original abstract

The hyperbolic space, characterized by a constant negative curvature and exponentially expanding space, aligns well with the structural properties of heterogeneous graphs. However, although heterogeneous graphs inherently possess diverse power-law structures, most hyperbolic heterogeneous graph embedding models rely on a single hyperbolic space. This approach may fail to effectively capture the diverse power-law structures within heterogeneous graphs. To address this limitation, we propose a Metapath-based Hyperbolic Contrastive Learning framework (MHCL), which uses multiple hyperbolic spaces to capture diverse complex structures within heterogeneous graphs. Specifically, by learning each hyperbolic space to describe the distribution of complex structures corresponding to each metapath, it is possible to capture semantic information effectively. Since metapath embeddings represent distinct semantic information, preserving their discriminability is important when aggregating them to obtain node representations. Therefore, we use a contrastive learning approach to optimize MHCL and improve the discriminability of metapath embeddings. In particular, our contrastive learning method minimizes the distance between embeddings of the same metapath and maximizes the distance between those of different metapaths in hyperbolic space, thereby improving the separability of metapath embeddings with distinct semantic information. We conduct comprehensive experiments to evaluate the effectiveness of MHCL. The experimental results demonstrate that MHCL outperforms state-of-the-art baselines in various graph machine learning tasks, effectively capturing the complex structures of heterogeneous graphs.

Figures

Figures reproduced from arXiv: 2506.16754 by the authors.

Figure 1
Figure 1. Examples of heterogeneous graph representations in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Metapath instance distributions of some metapaths on the DBLP dataset. Here, [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Basic concept of MHCL.(Here, the striped circles represent the embedding of each metapath.) [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: The framework of proposed MHCL. Here, node [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Example of the sample selection strategy. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Node classification accuracy for varying values of [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Parameter sensitivity of MHCL w.r.t Dimension of the metapath embedding [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Visualization of metapath embeddings on the ACM and DBLP datasets. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Training time of MHCL on the ACM dataset. [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 26 canonical work pages

  1. [1]

    Pathsim: Meta path- based top-k similarity search in heterogeneous information networks,

    Y . Sun, J. Han, X. Yan, P. S. Yu, and T. Wu, “Pathsim: Meta path- based top-k similarity search in heterogeneous information networks,” inVLDB, 2011, pp. 992–1003

    work page 2011

  2. [2]

    Heterogeneous graph attention network,

    X. Wang, H. Ji, C. Shi, B. Wang, Y . Ye, P. Cui, and P. S. Yu, “Heterogeneous graph attention network,” inWWW, 2019, pp. 2022– 2032

    2019

  3. [3]

    Graph transformer networks,

    S. Yun, M. Jeong, R. Kim, J. Kang, and H. J. Kim, “Graph transformer networks,” inNeurIPS, 2019, pp. 11 960–11 970

    work page 2019

  4. [4]

    Graphmse: Efficient meta-path selection in semantically aligned feature space for graph neural networks,

    Y . Li, Y . Jin, G. Song, Z. Zhu, C. Shi, and Y . Wang, “Graphmse: Efficient meta-path selection in semantically aligned feature space for graph neural networks,” inAAAI, 2021, pp. 4206–4214

    work page 2021

  5. [5]

    Heterogeneous graph trans- former,

    Z. Hu, Y . Dong, K. Wang, and Y . Sun, “Heterogeneous graph trans- former,” inWWW, 2020, pp. 2704–2710

    work page 2020

  6. [6]

    Are we really making much progress? revisiting, benchmarking and refining heterogeneous graph neural networks,

    Q. Lv, M. Ding, Q. Liu, Y . Chen, W. Feng, S. He, C. Zhou, J. Jiang, Y . Dong, and J. Tang, “Are we really making much progress? revisiting, benchmarking and refining heterogeneous graph neural networks,” in KDD, 2021, pp. 1150–1160

    2021

  7. [7]

    Curvature regu- larization to prevent distortion in graph embedding,

    H. Pei, B. Wei, K. Chang, C. Zhang, and B. Yang, “Curvature regu- larization to prevent distortion in graph embedding,” inNeurIPS, 2020, pp. 20 779–20 790

    work page 2020

  8. [8]

    Poincar ´e embeddings for learning hierarchical representations,

    M. Nickel and D. Kiela, “Poincar ´e embeddings for learning hierarchical representations,” inNIPS, 2017, pp. 6338–6348

    work page 2017

Show all 36 references
  1. [9]

    Multi-relational poincar ´e graph embeddings,

    I. Balazevic, C. Allen, and T. Hospedales, “Multi-relational poincar ´e graph embeddings,” inNeurIPS, 2019, pp. 4465–4475

    2019

  2. [10]

    Hyperbolic hierarchy-aware knowledge graph embedding for link prediction,

    Z. Pan and P. Wang, “Hyperbolic hierarchy-aware knowledge graph embedding for link prediction,” inFindings of EMNLP, 2021, pp. 2941– 2948

    2021

  3. [11]

    Hyperbolic deep neural networks: A survey,

    W. Peng, T. Varanka, A. Mostafa, H. Shi, and G. Zhao, “Hyperbolic deep neural networks: A survey,”IEEE Transactions on pattern analysis and machine intelligence, vol. 44, no. 12, pp. 10 023–10 044, 2021

    2021

  4. [12]

    Hy- perbolic graph neural networks: A review of methods and applications,

    M. Yang, M. Zhou, Z. Li, J. Liu, L. Pan, H. Xiong, and I. King, “Hy- perbolic graph neural networks: A review of methods and applications,” arXiv preprint arXiv:2202.13852, 2022

    2022 arXiv

  5. [13]

    Hyperbolic heterogeneous information network embedding,

    X. Wang, Y . Zhang, and C. Shi, “Hyperbolic heterogeneous information network embedding,” inAAAI, 2019, pp. 5337–5344

    2019

  6. [14]

    Multi-order relations hyperbolic fusion for heterogeneous graphs,

    J. Li, Y . Sun, and M. Shao, “Multi-order relations hyperbolic fusion for heterogeneous graphs,” inCIKM, 2023, pp. 1358–1367

    2023

  7. [15]

    Hyperbolic heterogeneous graph attention networks,

    J. Park, S. Han, S. Jeong, and S. Lim, “Hyperbolic heterogeneous graph attention networks,” inWWW, 2024, pp. 561–564

    2024

  8. [16]

    Tree-like structure in large social and information networks,

    A. B. Adcock, B. D. Sullivan, and M. W. Mahoney, “Tree-like structure in large social and information networks,” inICDM, 2013, pp. 1–10

    2013

  9. [17]

    Scaled gromov hyperbolic graphs,

    E. Jonckheere, P. Lohsoonthorn, and F. Bonahon, “Scaled gromov hyperbolic graphs,”Journal of Graph Theory, vol. 57, no. 2, pp. 157– 180, 2008

    2008

  10. [18]

    Large-scale curvature of networks,

    O. Narayan and I. Saniee, “Large-scale curvature of networks,”Physical Review E, vol. 84, no. 6, p. 066108, 2011

    2011

  11. [19]

    A comprehensive survey on graph neural networks,

    Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and S. Y . Philip, “A comprehensive survey on graph neural networks,”IEEE Trans. on Neural Networks and Learning Systems, vol. 32, no. 1, pp. 4–24, 2020

    2020

  12. [20]

    Graph neural networks: A review of methods and applications,

    J. Zhou, G. Cui, S. Hu, Z. Zhang, C. Yang, Z. Liu, L. Wang, C. Li, and M. Sun, “Graph neural networks: A review of methods and applications,” AI Open, vol. 1, pp. 57–81, 2020

    2020

  13. [21]

    A review of graph neural networks: concepts, architectures, techniques, challenges, datasets, applications, and future directions,

    B. Khemani, S. Patil, K. Kotecha, and S. Tanwar, “A review of graph neural networks: concepts, architectures, techniques, challenges, datasets, applications, and future directions,”Journal of Big Data, vol. 11, no. 1, p. 18, 2024

    2024

  14. [22]

    A survey on heterogeneous graph embedding: Methods, techniques, applications and sources,

    X. Wang, D. Bo, C. Shi, S. Fan, Y . Ye, and S. Y . Philip, “A survey on heterogeneous graph embedding: Methods, techniques, applications and sources,”IEEE Trans. on Big Data, vol. 9, no. 2, pp. 415–436, 2022

    2022

  15. [23]

    Heterogeneous graph neural networks analysis: A survey of techniques, evaluations and applications,

    R. Bing, G. Yuan, M. Zhu, F. Meng, H. Ma, and S. Qiao, “Heterogeneous graph neural networks analysis: A survey of techniques, evaluations and applications,”Artificial Intelligence Review, vol. 56, no. 8, pp. 8003– 8042, 2023

    2023

  16. [24]

    Magnn: Metapath aggregated graph neural network for heterogeneous graph embedding,

    X. Fu, J. Zhang, Z. Meng, and I. King, “Magnn: Metapath aggregated graph neural network for heterogeneous graph embedding,” inWWW, 2020, pp. 2331–2341

    2020

  17. [25]

    Multilayer feedforward networks are universal approximators,

    K. Hornik, M. Stinchcombe, and H. White, “Multilayer feedforward networks are universal approximators,”Neural networks, vol. 2, no. 5, pp. 359–366, 1989. 14

    1989

  18. [26]

    Mch-hgcn: Multi-curvature hyperbolic heteroge- neous graph convolutional network with type triplets,

    Y . Liu and B. Lang, “Mch-hgcn: Multi-curvature hyperbolic heteroge- neous graph convolutional network with type triplets,”Neural Comput- ing and Applications, vol. 35, no. 20, pp. 15 033–15 049, 2023

    2023

  19. [27]

    Multi-hyperbolic space-based heterogeneous graph attention network,

    J. Park, S. Han, J.-R. Lee, and S. Lim, “Multi-hyperbolic space-based heterogeneous graph attention network,” inICDM, 2024, pp. 815–820

    2024

  20. [28]

    Towards graph contrastive learning: A survey and beyond,

    W. Ju, Y . Wang, Y . Qin, Z. Mao, Z. Xiao, J. Luo, J. Yang, Y . Gu, D. Wang, Q. Longet al., “Towards graph contrastive learning: A survey and beyond,”arXiv preprint arXiv:2405.11868, 2024

    2024 arXiv

  21. [29]

    Enhancing hyperbolic graph embeddings via contrastive learning,

    J. Liu, M. Yang, M. Zhou, S. Feng, and P. Fournier-Viger, “Enhancing hyperbolic graph embeddings via contrastive learning,”arXiv preprint arXiv:2201.08554, 2022

    2022 arXiv

  22. [30]

    Hyperbolic contrastive learning for cross-domain recommen- dation,

    X. Yang, H. Chang, Z. Lai, J. Yang, X. Li, Y . Lu, S. Wang, D. Yin, and E. Min, “Hyperbolic contrastive learning for cross-domain recommen- dation,” inCIKM, 2024, pp. 2920–2929

    2024

  23. [31]

    Hyperbolic graph contrastive learning for collaborative filtering,

    Z. Qin, W. Cheng, W. Ding, and G. Ding, “Hyperbolic graph contrastive learning for collaborative filtering,”IEEE Transactions on Knowledge and Data Engineering, vol. 37, no. 3, pp. 1255–1267, 2025

    2025

  24. [32]

    Semi-supervised classification with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” inICLR, 2017

    2017

  25. [33]

    Graph attention networks,

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Li `o, and Y . Bengio, “Graph attention networks,” inICLR, 2018

    2018

  26. [34]

    Hyperbolic graph convolu- tional neural networks,

    I. Chami, Z. Ying, C. R ´e, and J. Leskovec, “Hyperbolic graph convolu- tional neural networks,” inNeurIPS, 2019, pp. 4869–4880

    2019

  27. [35]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” inICLR, 2019

    2019

  28. [36]

    Emergence of scaling in random net- works,

    A.-L. Barab ´asi and R. Albert, “Emergence of scaling in random net- works,”Science, vol. 286, no. 5439, pp. 509–512, 1999

    1999

Pith tools

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