Pith. sign in

REVIEW 5 major objections 5 minor 47 references

Learning Robust Heterogeneous Graph Representations via Contrastive-Reconstruction under Sparse Semantics

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a single HAN encoder shared between masked reconstruction and contrastive learning, with contrastive views built from embeddings plus two positive-sample augmentation strategies, outperforms state-of-the-art…

desk verdict HetCRF is a promising hybrid SSL architecture for heterogeneous graphs that is not review-ready because the stated contrastive loss is internally inconsistent, though the issue looks like a fixable typo. read the letter →

arxiv 2506.06682 v1 pith:6HYUVKVU submitted 2025-06-07 cs.LG

Di Lin , Wanjing Ren , Xuanbin Li , Rui Zhang This is my paper
classification cs.LG
keywords self-supervisedlearningcontrastivemaskedautoencoderheterogeneousgraphneuralnetworkpositivesampleaugmentationmeta-pathsemanticsparsity
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 claims that on heterogeneous graphs, a single shared encoder can serve both generative reconstruction and contrastive learning, provided the contrastive branch is rebuilt on top of encoder embeddings instead of raw inputs. Its proposed framework, HetCRF, pairs masked feature and meta-path reconstruction with a GCN-augmented contrastive channel, and balances the contrastive loss by augmenting positive samples through meta-path connectivity and K-means clustering. The payoff would be node representations that survive semantically sparse settings, where raw features are missing and ordinary view augmentation destroys what little signal exists. On AMiner and Freebase at a 40% label rate, the paper reports Macro-F1 gains of 2.75 and 2.2 points over the second-best baseline.

What carries the argument

The framework's load-bearing object is the shared HAN encoder, a heterogeneous graph attention network that maps multiple node and edge types into common embeddings, reused by both channels. The generative channel masks node features and meta-path edges and reconstructs them under scaled cosine error; the contrastive channel takes the encoder's embeddings, applies PathSim-based top-K neighbor selection and GCN propagation to form schema and fusion views, and optimizes a contrastive loss with a learned positive-sample set. That set is built by counting connecting meta-paths across k hops and by K-means on the encoder's own embeddings, selecting 'key deviated nodes' per cluster as extra positives; the final loss is a weighted sum of the two reconstruction losses and the contrastive loss.

What would settle it

On AMiner at a 40% label rate, replace the K-means positive augmentation with randomly sampled positives of the same set size and keep everything else fixed; if Macro-F1 stays near 79.68, the semantic validity of the clusters is not doing the work, whereas a clear drop would confirm the assumption. A second observation: run the same experiment with the cluster count set to a value different from the known class count; if performance is unchanged, the method does not actually depend on knowing $S$.

Watch

Extended reading notes

Core claim

The central claim is that the apparent incompatibility between masked autoencoding and contrastive learning under a shared encoder disappears if the contrastive channel performs a second aggregation step on the encoder's output. HetCRF builds two contrastive views, a schema view from heterogeneous-neighbor aggregation and a fusion view from attention-weighted meta-path adjacency, both propagated through a GCN, so contrastive semantics are drawn from embeddings rather than from brittle raw-feature augmentations. In addition, the paper proves that in InfoNCE the positive sample's gradient equals the sum over all negatives, and argues that adding extra positives rebalances the gradient and protects global-structure learning. The experimental claim is that this design outperforms state-of-the-art generative, contrastive, and hybrid baselines across DBLP, ACM, AMiner, and Freebase, with its largest margins precisely on the two feature-missing datasets.

Load-bearing premise

For the clustering augmentation, the model assumes it knows how many true classes $S$ exist and that K-means on its own current embeddings produces semantically valid positive pairs, so noisy clusters would inject false supervision into contrastive training.

Editorial extensions

If this is right

  • On feature-missing heterogeneous graphs such as AMiner and Freebase, the reported gains (Macro-F1 +2.75 and +2.2 points at 40% labels) are larger than on feature-rich graphs, so embedding-level view construction appears to be the regime where the design pays off most.
  • The ablation shows that the hybrid beats generative-only and contrastive-only training, and removing the GCN/view-construction step costs up to 5.04 and 4.05 Macro-F1 points on the two sparse datasets.
  • Positive augmentation itself matters: with neither augmentation strategy, Macro-F1 drops by several points at 40% labels on Freebase and AMiner, and the two strategies combined give the best numbers.
  • When original node features are removed from DBLP and ACM, HetCRF still leads four strong baselines, which is the paper's evidence that the framework generalizes to semantically sparse conditions.
  • The gradient-balance theorem implies that any InfoNCE-based contrastive channel can rebalance positive and negative contributions by enlarging the positive set, not just this particular architecture.

Reading between the lines

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

  • A random-positive control of equal cardinality would separate the gradient-balancing effect from the semantic quality of K-means positives; the paper's ablations do not include one, so the reported gains could partly come from having more positives rather than from the specific chosen positives.
  • Because the clustering augmentation needs the true number of classes $S$ at pre-training time, applying HetCRF to a graph with an unknown class count is a risk the paper does not address; varying $S$ in the AMiner setting is a cheap test of how load-bearing this assumption is.
  • The gradient-balance theorem is stated for the standard InfoNCE objective, so the same positive-augmentation logic could be lifted into any InfoNCE-based graph contrastive learner without the rest of HetCRF.
  • The two attention-weighted view constructions give a template for handling sparse semantics; one could try building the same schema/fusion views on top of a homogeneous masked-autoencoder encoder, transferring the mechanism beyond heterogeneous graphs.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes HetCRF, a self-supervised dual-channel framework for heterogeneous graphs. The generative channel performs masked feature reconstruction and meta-path reconstruction with a shared HAN encoder; the contrastive channel builds views from encoder embeddings (schema view and fusion view) and passes them through a GCN for secondary aggregation. Two positive-sample augmentation strategies are introduced, one based on meta-path connection counts and one based on K-means clustering of encoder embeddings, with the stated goal of balancing positive and negative gradient contributions in the contrastive loss. Experiments on DBLP, Freebase, ACM, and AMiner report gains in node classification and clustering, with the largest claimed gains on feature-missing datasets such as AMiner and Freebase.

Significance. If the claims held, HetCRF would be a useful contribution to heterogeneous-graph self-supervision, particularly in sparse-feature regimes: the paper targets a real limitation of existing hybrid frameworks and provides ablations on four datasets. However, the theoretical gradient-balance argument is not correctly connected to the contrastive loss as written in Eq. (26), and the ablation results in Table 3 do not consistently support the claim that the two augmentation strategies are jointly beneficial. These are load-bearing issues for the paper's central contribution, so the current version cannot be accepted. I also note that no code or full hyperparameter settings are provided, which limits independent verification.

major comments (5)
  1. [§3.2.2, Eq. (26)] The contrastive loss in Eq. (26) is written with a denominator Σ_{j∈P_i∪N_i} exp(Z_j^fusion · Z_j^schema / τ), i.e., self-similarities rather than anchor-to-sample similarities. This is not a valid contrastive softmax over the anchor's neighbors, and it does not reduce to the InfoNCE form used in Theorem 1. Consequently, the gradient-balance theorem and its generalization in Eq. (21) do not apply to the objective as specified. Please correct the denominator to Σ_{j∈P_i∪N_i} exp(Z_i^fusion · Z_j^schema / τ), or state explicitly that a different objective was used in the experiments. This is not a cosmetic issue: the theoretical motivation for both positive-augmentation strategies depends on it.
  2. [§3.2.2, Eq. (21)] For a genuine multi-positive InfoNCE loss L_i = -log[Σ_{p∈P} exp(s_ip/τ) / (Σ_{p∈P} exp(s_ip/τ) + Σ_{n∈N} exp(s_in/τ))], the identity Σ_{p∈P} ∂L_i/∂f_p = -Σ_{n∈N} ∂L_i/∂f_n is actually correct, so I do not dispute the generalization itself. However, the paper does not show the derivation, and the identity fails for the loss as written in Eq. (26). Please provide the multi-positive derivation explicitly and state it for the corrected loss; the current 'analogous proof' is insufficient because the denominator terms in Eq. (26) change the relevant derivatives.
  3. [§4.3.2, Table 3] Table 3 does not support the claim that the two augmentation strategies are jointly beneficial. On DBLP, w/o_Cluster (MPC-only) achieves 92.75 Macro-F1 versus 92.55 for the full HetCRF; on ACM, w/o_MPC (cluster-only) achieves 92.51 versus 92.35 for the full model. Thus, adding the second strategy degrades performance on two of the four datasets. The paper should either explain this inconsistency or revise the claim that both strategies contribute to the final model.
  4. [§3.2.2, Positive Sample Augmentation via Clustering Algorithm] The clustering-based augmentation uses the true number of classes S in K-means during self-supervised pre-training. If S is derived from label information, the method is not label-free in the usual self-supervised sense, and the comparison to baselines is unfair unless all baselines also use S. Moreover, noisy cluster assignments can inject false positives; the paper provides no experiment varying S or measuring cluster quality. Please justify the use of S or show robustness to misspecified S.
  5. [§4.2, Table 2] The statement that HetCRF demonstrates superior performance over the existing baselines in clustering is not accurate for all datasets and metrics. On ACM, HGMS-C achieves NMI 71.97 while HetCRF achieves 69.48, so HetCRF is not the best on that dataset. The clustering claim should be qualified to reflect the actual wins and losses in Table 2.
minor comments (5)
  1. [§4.3.1] The text says the three training strategies were evaluated under a 20% label rate, while Figure 3's caption says 'Macro-F1, 40%'; please make the label rate consistent between text and figure.
  2. [§3.1 and §3.2.1] Notation is inconsistent: Eq. (5) uses σ for the activation, while Eq. (7) uses δ; Eq. (9) has |Z_i| in the softmax denominator, which should presumably be |Φ|; and the text near Eq. (5) says 'Hermitian space', which appears to mean the node-type feature space.
  3. [Figures 3 and 4] Figures 3 and 4 are difficult to read: the t-SNE labels in Figure 4 are illegible, and Figure 3 renders axis labels as encoded tokens rather than readable text. Please provide vector graphics with readable fonts.
  4. [§4] No code, hyperparameter table, or training details are provided, so the numerical results in Tables 1-4 cannot be independently reproduced. Please include a reproducibility appendix with exact settings and random-seed handling.
  5. [References and baselines] The reference list says 'HERO ICRL'24', which should be 'ICLR'24'; also, the baseline enumeration in Section 4 omits DGI and DMGI, which appear in Table 2, so the list of compared methods in the text does not match the tables.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: HetCRF's empirical benchmark claims stand on external comparisons, and the positive-sample augmentation rules are heuristics rather than fitted inputs or self-satisfying predictions.

full rationale

I find no circular step under the requested definition. The central claim of the paper is an empirical comparison on four heterogeneous graph datasets (Tables 1, 2, and the ablations in Tables 3 and 4); the reported Macro-F1 gains are benchmark results, not quantities predicted from fitted parameters. The two positive-sample augmentation strategies are explicit heuristics: meta-path connection counting selects nodes with many meta-path links, and the clustering strategy selects cluster outliers using K-means on current embeddings with the dataset's known number of classes. Neither strategy takes the downstream label performance as input, so there is no 'fitted input renamed as prediction' pattern. Theorem 1 is a direct derivative computation for the standard one-positive InfoNCE loss, and Eq. (21) is asserted as a generalization; even if Eq. (21) is mathematically incorrect for multiple positives and Eq. (26) appears to have a denominator that is not a proper anchor-to-sample softmax, these are correctness or typographical concerns, not circular reductions. The paper does not rest its core results on a self-citation chain: references such as GCMAE, GraphMAE, HGMAE, and HeCo are external prior work cited as baselines or design sources, and no 'uniqueness theorem' from the authors' own prior work is imported to force a modeling choice. The ablation studies directly compare HetCRF against its own ablations, which is an internal empirical check rather than a circular justification. The noted limitation that HGMS-C results are taken from the original paper because source code is inaccessible affects reproducibility, but does not make the comparison self-referential by construction. I therefore assign a circularity score of 0.

Assumptions & free parameters 9 free parameters · 5 assumptions · 0 invented entities

HetCRF depends on a large set of unreported hyperparameters and on domain assumptions about label availability and cluster semantics; these are not provided with values or sensitivity analysis, so the empirical claim is under-specified.

free parameters (9)
  • mask rate rho = not reported
    Fraction of nodes whose features are masked in the generative channel (Section 3.1).
  • edge masking rate p_e = not reported
    Bernoulli rate for meta-path edge masking (Section 3.1).
  • scaling factors gamma1, gamma2 = not reported
    Exponents in scaled cosine error terms in Eq (1) and Eq (2).
  • temperature tau = not reported
    Temperature in the contrastive loss, Section 3.2.2.
  • threshold T_pos = not reported
    Maximum number of meta-path positive samples selected per node in the first augmentation strategy, Section 3.2.2.
  • positive sample hop count k = not reported
    Number of hops used to expand positive sample matrix in Eq (23); authors state different k per dataset but do not report values.
  • cluster count S = true number of classes
    Number of K-means clusters in the second augmentation strategy; this uses downstream label information during pre-training.
  • key deviated nodes count K = not reported
    Number of farthest nodes per cluster selected as augmented positives, Section 3.2.2.
  • loss weights lambda1, lambda2 = not reported
    Weights of the two reconstruction losses in Eq (27); tuned but not reported.
assumptions (5)
  • domain assumption Softmax-normalized dot-product contrastive loss (InfoNCE) is the appropriate objective for the contrastive channel.
    The entire gradient analysis in Theorem 1 builds on this loss form; the paper does not compare alternatives.
  • domain assumption The number of classes S is known during self-supervised pre-training.
    Used in K-means clustering for positive augmentation; not stated as a label-dependent input.
  • ad hoc to paper Meta-path connectivity indicates semantic relatedness, so heavily connected nodes are valid positive samples.
    First augmentation strategy in Section 3.2.2; no external validation that these are true positives.
  • ad hoc to paper K-means clusters in embedding space are semantically coherent, and the farthest nodes in a cluster provide useful hard positives.
    Second augmentation strategy; depends on the current embedding quality and can reinforce model bias.
  • domain assumption Augmenting encoder embeddings rather than raw features preserves semantic invariance under missing features.
    Core design choice in Section 3.2.1; presented without formal justification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Robust Heterogeneous Graph Representations via Contrastive-Reconstruction under Sparse Semantics." pith.science (2026). https://pith.science/paper/6HYUVKVU

@misc{pith2026250606682,
  author       = {Pith},
  title        = {Pith review of: Learning Robust Heterogeneous Graph Representations via Contrastive-Reconstruction under Sparse Semantics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6HYUVKVU}},
  note         = {Machine review of arXiv:2506.06682}
}
read the original abstract

In graph self-supervised learning, masked autoencoders (MAE) and contrastive learning (CL) are two prominent paradigms. MAE focuses on reconstructing masked elements, while CL maximizes similarity between augmented graph views. Recent studies highlight their complementarity: MAE excels at local feature capture, and CL at global information extraction. Hybrid frameworks for homogeneous graphs have been proposed, but face challenges in designing shared encoders to meet the semantic requirements of both tasks. In semantically sparse scenarios, CL struggles with view construction, and gradient imbalance between positive and negative samples persists. This paper introduces HetCRF, a novel dual-channel self-supervised learning framework for heterogeneous graphs. HetCRF uses a two-stage aggregation strategy to adapt embedding semantics, making it compatible with both MAE and CL. To address semantic sparsity, it enhances encoder output for view construction instead of relying on raw features, improving efficiency. Two positive sample augmentation strategies are also proposed to balance gradient contributions. Node classification experiments on four real-world heterogeneous graph datasets demonstrate that HetCRF outperforms state-of-the-art baselines. On datasets with missing node features, such as Aminer and Freebase, at a 40% label rate in node classification, HetCRF improves the Macro-F1 score by 2.75% and 2.2% respectively compared to the second-best baseline, validating its effectiveness and superiority.

Figures

Figures reproduced from arXiv: 2506.06682 by the authors.

Figure 1
Figure 1. Existing GCMAE model architecture certain limitations. Specifically, generative learning reconstructs graph structures or feature representations by aggregating infor￾mation from neighboring nodes[35]. This process tends to cause over-smoothing, making it more suitable for shallower encoder structures to preserve semantic details. In contrast, contrastive learning relies on specially designed objective functions to … view at source ↗
Figure 2
Figure 2. The overall HetCRF framework comprises two channels: the generative channel (b) and the contrastive channel (c). [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Ablation results comparing different training [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Comparison of t-SNE visualization effects using [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 6
Figure 6. Figure 6: The contrastive learning channel of 𝐻𝑒𝑡𝐶𝑅𝐹𝑤/𝑜−𝐺𝐶𝑁 We applied HetCRF and 𝐻𝑒𝑡𝐶𝑅𝐹𝑤/𝑜−𝐺𝐶𝑁 on four benchmark datasets and presented the model accuracy results as bar charts in [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 8
Figure 8. Figure 8: Heatmap of loss weight adjustment impact on [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 30 canonical work pages

  1. [1]

    Renato B Arantes, George Vogiatzis, and Diego R Faria. 2022. Learning an augmentation strategy for sparse datasets.Image and Vision Computing117 (2022), 104338

    work page 2022

  2. [2]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. InInterna- tional conference on machine learning. PmLR, 1597–1607

    2020

  3. [3]

    Haoran Duan, Cheng Xie, and Linyu Li. 2024. Reserving-masking-reconstruction model for self-supervised heterogeneous graph representation. InProceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 689–700

    2024

  4. [4]

    Xinyu Fu, Jiani Zhang, Ziqiao Meng, and Irwin King. 2020. Magnn: Metap- ath aggregated graph neural network for heterogeneous graph embedding. In Proceedings of the web conference 2020. 2331–2341

    2020

  5. [5]

    Lukas Gianinazzi, Maximilian Fries, Nikoli Dryden, Tal Ben-Nun, Maciej Besta, and Torsten Hoefler. 2021. Learning combinatorial node labeling algorithms. arXiv preprint arXiv:2106.03594(2021)

    work page Pith review arXiv 2021

  6. [6]

    Thomas Gottron, Malte Knauf, and Ansgar Scherp. 2015. Analysis of schema structures in the Linked Open Data graph based on unique subject URIs, pay-level domains, and vocabulary usage.Distributed and Parallel Databases33 (2015), 515–553

    work page 2015

  7. [7]

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick

  8. [8]

    Zhenyu Hou, Xiao Liu, Yukuo Cen, Yuxiao Dong, Hongxia Yang, Chunjie Wang, and Jie Tang. 2022. Graphmae: Self-supervised masked graph autoencoders. In Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining. 594–604

    2022

Show all 47 references
  1. [9]

    Haojun Jiang, Jiawei Sun, Jie Li, and Chentao Wu. 2024. LocalGCL: Local-aware Contrastive Learning for Graphs.arXiv preprint arXiv:2402.17345(2024)

    2024 arXiv

  2. [10]

    Ruobing Jiang, Yacong Li, Haobing Liu, and Yanwei Yu. 2025. Incorporating at- tributes and multi-scale structures for heterogeneous graph contrastive learning. Information Fusion(2025), 103220

    2025

  3. [11]

    Thomas N Kipf and Max Welling. 2016. Semi-supervised classification with graph convolutional networks.arXiv preprint arXiv:1609.02907(2016)

    2016 arXiv

  4. [12]

    Xuanfeng Li, Lichuan Zhang, Qiaoqiao Zhao, Zixiao Zhu, and Feihu Zhang. 2025. P2GCN: Pixel-patch mutual enhancement graph convolutional network for sonar image super-resolution.Expert Systems with Applications279 (2025), 127265

    2025

  5. [13]

    Zongwei Li, Lianghao Xia, Hua Hua, Shijie Zhang, Shuangyang Wang, and Chao Huang. 2025. DiffGraph: Heterogeneous Graph Diffusion Model. InProceedings of the Eighteenth ACM International Conference on Web Search and Data Mining. 40–49

    2025

  6. [14]

    Qiqiang Lin, Xiaonan Huang, Ning Bi, Ching Y Suen, and Jun Tan. 2022. Cclsl: Combination of contrastive learning and supervised learning for handwritten mathematical expression recognition. InProceedings of the Asian Conference on Computer Vision. 3724–3739

    2022

  7. [15]

    Nian Liu, Xiao Wang, Hui Han, and Chuan Shi. 2023. Hierarchical contrastive learning enhanced heterogeneous graph neural network.IEEE Transactions on Knowledge and Data Engineering35, 10 (2023), 10884–10896

    2023

  8. [16]

    Xiao Liu, Fanjin Zhang, Zhenyu Hou, Li Mian, Zhaoyu Wang, Jing Zhang, and Jie Tang. 2021. Self-supervised learning: Generative or contrastive.IEEE transactions on knowledge and data engineering35, 1 (2021), 857–876

    2021

  9. [17]

    Stuart Lloyd. 1982. Least squares quantization in PCM.IEEE transactions on information theory28, 2 (1982), 129–137

    1982

  10. [18]

    Yujie Mo, Feiping Nie, Ping Hu, Heng Tao Shen, Zheng Zhang, Xinchao Wang, and Xiaofeng Zhu. 2024. Self-supervised heterogeneous graph learning: a homophily and heterogeneity view. InThe Twelfth International Conference on Learning Representations

    2024

  11. [19]

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. 2018. Representation learning with contrastive predictive coding.arXiv preprint arXiv:1807.03748(2018)

    2018 arXiv

  12. [20]

    Xiao Qin, Nasrullah Sheikh, Berthold Reinwald, and Lingfei Wu. 2021. Relation- aware graph attention model with adaptive self-adversarial training. InProceed- ings of the AAAI conference on artificial intelligence, Vol. 35. 9368–9376

    2021

  13. [21]

    Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. 2008. The graph neural network model.IEEE transactions on neural networks20, 1 (2008), 61–80

    2008

  14. [22]

    Chuan Shi, Yitong Li, Jiawei Zhang, Yizhou Sun, and Philip S Yu. 2016. A survey of heterogeneous information network analysis.IEEE Transactions on Knowledge and Data Engineering29, 1 (2016), 17–37

    2016

  15. [23]

    2012.Mining heterogeneous information networks: principles and methodologies

    Yizhou Sun and Jiawei Han. 2012.Mining heterogeneous information networks: principles and methodologies. Morgan & Claypool Publishers

    2012

  16. [24]

    Yizhou Sun, Jiawei Han, Xifeng Yan, Philip S Yu, and Tianyi Wu. 2011. Pathsim: Meta path-based top-k similarity search in heterogeneous information networks. Proceedings of the VLDB Endowment4, 11 (2011), 992–1003

    2011

  17. [25]

    Ying Tang, Yining Yang, and Guodao Sun. 2025. Generative and contrastive graph representation learning with message passing.Neural Networks(2025), 107224

    2025

  18. [26]

    Yijun Tian, Kaiwen Dong, Chunhui Zhang, Chuxu Zhang, and Nitesh V Chawla

  19. [27]

    Haosen Wang, Chenglong Shi, Can Xu, Surong Yan, and Pan Tang. 2025. Homophily-aware Heterogeneous Graph Contrastive Learning.arXiv preprint arXiv:2501.08538(2025)

    2025 arXiv

  20. [28]

    Jing Wang, Jiangyun Li, Wei Li, Lingfei Xuan, Tianxiang Zhang, and Wenxuan Wang. 2023. Positive–negative equal contrastive loss for semantic segmentation. Neurocomputing535 (2023), 13–24

    2023

  21. [29]

    Liang Wang, Xiang Tao, Qiang Liu, and Shu Wu. 2024. Rethinking graph masked autoencoders through alignment and uniformity. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 38. 15528–15536

    2024

  22. [30]

    Quan Wang, Zhendong Mao, Bin Wang, and Li Guo. 2017. Knowledge graph embedding: A survey of approaches and applications.IEEE transactions on knowledge and data engineering29, 12 (2017), 2724–2743

    2017

  23. [31]

    Xiao Wang, Houye Ji, Chuan Shi, Bai Wang, Yanfang Ye, Peng Cui, and Philip S Yu

  24. [32]

    Xiao Wang, Nian Liu, Hui Han, and Chuan Shi. 2021. Self-supervised heteroge- neous graph neural network with co-contrastive learning. InProceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. 1726–1736

    2021

  25. [33]

    Yu Wang, Lei Sang, Yi Zhang, Yiwen Zhang, and Xindong Wu. 2025. Generative- contrastive heterogeneous graph neural network.IEEE Transactions on Big Data (2025)

    2025

  26. [34]

    Yuxiang Wang, Xiao Yan, Chuang Hu, Quanqing Xu, Chuanhui Yang, Fangcheng Fu, Wentao Zhang, Hao Wang, Bo Du, and Jiawei Jiang. 2024. Generative and contrastive paradigms are complementary for graph self-supervised learning. In2024 IEEE 40th International Conference on Data Engi...

    2024

  27. [35]

    Shenzhi Yang, Jun Xia, Jingbo Zhou, Xingkai Yao, and Xiaofang Zhang

  28. [36]

    Xufeng Yao, Yang Bai, Xinyun Zhang, Yuechen Zhang, Qi Sun, Ran Chen, Ruiyu Li, and Bei Yu. 2022. Pcl: Proxy-based contrastive learning for domain general- ization. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 7097–7107

    2022

  29. [37]

    Yuning You, Tianlong Chen, Zhangyang Wang, and Yang Shen. 2020. When does self-supervision help graph convolutional networks?. Ininternational conference on machine learning. PMLR, 10871–10880

    2020

  30. [38]

    Jianxiang Yu, Qingqing Ge, Xiang Li, and Aoying Zhou. 2024. Heterogeneous graph contrastive learning with meta-path contexts and adaptively weighted negative samples.IEEE Transactions on Knowledge and Data Engineering(2024)

    2024

  31. [39]

    Chuxu Zhang, Dongjin Song, Chao Huang, Ananthram Swami, and Nitesh V Chawla. 2019. Heterogeneous graph neural network. InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. 793–803

    2019

  32. [40]

    Li Zhang, Yan Ge, and Haiping Lu. 2020. Hop-hop relation-aware graph neural networks.arXiv preprint arXiv:2012.11147(2020)

    2020 arXiv

  33. [41]

    Xin Zhang, Qiaoyu Tan, Xiao Huang, and Bo Li. 2024. Graph contrastive learning with personalized augmentation.IEEE Transactions on Knowledge and Data Engineering(2024)

    2024

  34. [42]

    Xin Zheng, Yi Wang, Yixin Liu, Ming Li, Miao Zhang, Di Jin, Philip S Yu, and Shirui Pan. 2022. Graph neural networks for graphs with heterophily: A survey. arXiv preprint arXiv:2202.07082(2022). Learning Robust Heterogeneous Graph Representations via Contrastive-Reconstruction...

    2022 arXiv

  35. [43]

    Yanqiao Zhu, Yichen Xu, Feng Yu, Qiang Liu, Shu Wu, and Liang Wang. 2021. Graph contrastive learning with adaptive augmentation. InProceedings of the web conference 2021. 2069–2080

    2021

  36. [2019]

    InThe world wide web conference

    Heterogeneous graph attention network. InThe world wide web conference. 2022–2032

    2022

  37. [2022]

    InProceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Masked autoencoders are scalable vision learners. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition. 16000–16009

  38. [2023]

    InProceedings of the AAAI conference on artificial intelligence, Vol

    Heterogeneous graph masked autoencoders. InProceedings of the AAAI conference on artificial intelligence, Vol. 37. 9997–10005

  39. [2025]

    NodeReg: Mitigating the Imbalance and Distribution Shift Effects in Semi-Supervised Node Classification via Norm Consistency.arXiv preprint arXiv:2503.03211(2025)

    2025 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.