Negative Metric Learning for Graphs

Chengpei Wu; Lilin Zhang; Ning Yang; Yiyang Zhao

REVIEW 4 major objections 4 minor 56 references

Negative Metric Learning for Graphs

T0 review · 4 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 a learnable negative metric network tightens the mutual-information lower bound for graph contrastive learning and consistently outperforms prior methods on six benchmark graphs.

desk verdict A learnable negative-weighting module gives consistent but modest gains in graph contrastive learning, yet the advertised theory (especially Theorem 2) does not hold up; treat it as an empirical paper with a broken theoretical veneer. read the letter →

arxiv 2505.10307 v1 pith:QBTTZWFC submitted 2025-05-15 cs.LG

Yiyang Zhao , Chengpei Wu , Lilin Zhang , Ning Yang This is my paper · ORCID

classification cs.LG

keywords graphcontrastivelearningfalsenegativesnegativemetricmutualinformationInfoNCEneuralnetworksbi-leveloptimizationself-supervised

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

Graph contrastive learning treats every non-anchor node as a negative sample, so same-label nodes are often wrongly pushed apart. This paper proposes replacing the fixed negative weights in the InfoNCE loss with a learnable Negative Metric Network that outputs a soft, normalized weight for every pair, indicating how likely one node is a true negative of another. Because these weights are learned from the contrastive views themselves through a bi-level objective, the method needs no hand-picked similarity thresholds, cluster counts, or neighborhood rules. The paper proves that optimizing this weighted loss maximizes a lower bound on mutual information that is at least as tight as InfoNCE, and that an optimal encoder keeps false negatives closer to the anchor than true negatives. On six graph benchmarks, the resulting embeddings improve downstream node classification and clustering over nine baselines.

What carries the argument

The load-bearing objects are the Negative Metric Network $M$ and the bi-level optimization that couples it with the encoder $E$. $M$ maps an embedding pair $(u_i, v_j)$ through an MLP and normalizes over $j$ so that each row of the negative metric matrix $M\in[0,1]^{N\times N}$ sums to one; cell $m_{ij}$ is the soft label for whether $j$ is a negative of $i$. The proof machinery is the variational mutual-information lower bound: representing the KL divergence as a maximization over test functions shows that minimizing $\mathcal{L}_{\mathrm{NML}}^{(i)}$ over $M$ maximizes a lower bound on $I(U;V)$ that dominates InfoNCE. Equation (11) is the feedback mechanism: the optimal metric weight $m_{ik}^{(t+1)}$ is inversely proportional to the encoder's current similarity $\theta(u_i^{(t)}, v_k^{(t)})$, so the encoder's push-out of $v_k$ makes the metric mark $v_k$ as a stronger negative, which then instructs the next encoder update to push it even farther, forming a self-training loop. Theorem 2 rests on a Markov-chain argument in which the augmented views are generated by the label variable plus independent random perturbations.

What would settle it

Run NML-GCL on a heterophilous graph where connected nodes usually have different labels, using the same augmentation recipe, and measure the median distance ratio between false negatives and true negatives as in the paper's Table 3. If the ratio no longer falls below the InfoNCE baseline on such graphs, the label-only Markov-chain assumption behind Theorem 2 is the cause; the paper's current evidence covers only homophilous citation and co-purchase networks.

Watch

Extended reading notes

Core claim

The central claim is that the false-negative problem in graph contrastive learning can be addressed by learning the geometry of negativity itself rather than assuming it. NML-GCL augments a GCN encoder with a Negative Metric Network, an MLP followed by a softmax over nodes, that assigns each candidate negative pair a weight $m_{ij}$, interpretable as the probability that node $j$ is a true negative of anchor $i$. Training alternates between updating the metric network with the encoder frozen and updating the encoder with the metric network frozen, in a bi-level objective; the self-supervised signal that the two views of the same node form a positive pair implicitly supervises the metric network, since the hinge-form loss forces large weights onto dissimilar candidates. The theoretical part proves $I(U;V) \ge I_{\mathrm{NML}}(U;V) \ge I_{\mathrm{NCE}}(U;V)$, so maximizing the proposed loss maximizes a tighter MI lower bound than InfoNCE, and proves that an MI-optimal encoder satisfies $\mathbb{E}_{j\in S_i^+}[d(z_i,z_j)] \le \mathbb{E}_{j\in S_i^-}[d(z_i,z_j)]$, meaning it ranks negatives the way the oracle label function would. The experiments report consistent accuracy and clustering gains over baselines that use fixed, prior-driven negative weights.

Load-bearing premise

The theory's guarantee that a trained encoder ranks false negatives closer than true negatives assumes the two augmented views are generated entirely from the node's class label plus independent random noise, so the original graph structure contributes nothing beyond the label; real graphs carry structure beyond labels, so that premise is typically false.

Editorial extensions

If this is right

Because the learned loss is a provably tighter MI lower bound than InfoNCE, NML-GCL can replace fixed negative weighting without sacrificing the theoretical grounding of contrastive learning.
At convergence the metric network assigns smaller weights to false negatives and larger weights to true negatives, so the encoder's loss no longer pushes same-label nodes apart; the reported weight curves in the appendix show this on all six datasets.
The bi-level training is self-reinforcing: a better encoder gives cleaner similarity supervision for the metric network, and a better metric network gives sharper push-away instructions for the next encoder update.
Downstream, the method reports higher node classification accuracy and clustering purity than hard-weight, soft-weight, and no-false-negative baselines on Cora, CiteSeer, PubMed, Photo, Computers, and Wiki-CS.
The method needs no human priors such as similarity thresholds or cluster counts to identify false negatives, which is the specific limitation the paper targets.

Reading between the lines

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

If the label-generation model behind Theorem 2 fails on graphs where structure carries information beyond class labels, NML-GCL's advantage on heterophilous graphs may shrink; testing distance ratios on heterophilous benchmarks would reveal how much of the gain is label collapse versus structure preservation.
The learned negative metric matrix could be exported as a soft edge-weight assignment for tasks beyond contrastive pretraining, such as label propagation or graph clustering; the paper does not explore this direction.
The method's $\mathcal{O}(N^2)$ per-iteration cost and full-batch softmax over all nodes limit it to moderate-size graphs, so a sampled or blockwise version of the metric network is a natural testable extension.
Because the hinge form resembles margin-based self-training, NML-GCL could inherit known instabilities such as confirmation bias, which the KL regularization only partially addresses; this is an inference, not a claim of the paper.

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.

Desk Editor's Note

read the letter

Quick take: the method is a plausible empirical contribution—learnable soft weighting of negative samples in graph contrastive learning, trained with alternating optimization. The experiments are decent: six benchmarks, consistent gains around 1–3 points, standard deviations, p-values, and ablations. The novelty is real but narrow: weighting negatives is established, but learning the weighting function with a small MLP is new.

What does not hold up is the theory the paper leans on. Theorem 2's proof assumes augmented views are generated from the label variable plus independent noise, so the original graph structure and features contribute nothing beyond the labels. That is false for the actual augmentations, which are DropEdge and feature masking applied to the observed graph. Even if the mutual information equality were true, it is an average statement and does not imply every false-negative pair gets identical embeddings. So the claimed guarantee that maximizing MI collapses false negatives to the anchor is not established. Theorem 1 is more defensible but close to trivial: since the learned weights include uniform as a special case, the bound is automatically at least as tight as InfoNCE. Calling that a theoretical result is overclaiming. And Equations (9) and (11), the closed-form update for the negative metric network, have exponent errors and a misspecified denominator; as printed they do not follow from the loss.

The empirical section is the real strength. The method consistently beats the baselines, and the ablations (w/o M, replace with cosine similarity) show the learned metric matters. But there is no code release, which hurts reproducibility, and the hyperparameter search for baselines is not fully detailed. These are minor relative to the theory problem, but they matter.

Who is this for? Someone working on false negatives in graph contrastive learning might get ideas from the loss design and the alternating training. It is not a paper to cite for theoretical guarantees. My recommendation: this deserves a serious referee, but not acceptance in the current form. Ask the authors to either fix the proofs with real assumptions or, preferably, resubmit as a purely empirical method paper with the theory section cut down to a simple inequality. The method is worth having; the packaging is not.

Referee Report

4 major / 4 minor

Summary. The paper proposes NML-GCL, a graph contrastive learning method that augments InfoNCE with a learnable Negative Metric Network (NMN). The NMN assigns soft weights to negative samples, and the encoder and NMN are trained jointly through a bi-level optimization objective. The manuscript claims two theoretical results: Theorem 1 states that the NML loss is a tighter lower bound on view mutual information than InfoNCE, and Theorem 2 states that maximizing this bound makes the optimal encoder place false negatives closer to an anchor than true negatives. The authors report consistent improvements over several GCL baselines on six benchmark datasets, with ablations and hyper-parameter studies.

Significance. The empirical component is carefully presented: results are averaged over 10 runs, standard deviations and p-values are reported, ablation variants are compared, and hyper-parameter search spaces are documented. If the theoretical claims were sound, the work would offer a principled, prior-free treatment of false negatives in graph contrastive learning. However, the central theoretical results are not established: the proof of Theorem 1 contains an algebraic error in the variational lower-bound argument, and Theorem 2 relies on an augmentation model that does not match the actual algorithm. Because the advertised false-negative discrimination guarantee is the main conceptual contribution, the significance of the paper as submitted is substantially weakened.

major comments (4)

[Appendix B.1, proof of Theorem 1] The variational proof drops the penalty term. After substituting f(u_i,v_i)=log(N s_ii/(s_ii+(N-1) sum_j E_{P_U P_V}[m_ij s_ij])), the displayed chain contains the term '- E_{(u_i,v_i)~P_U P_V}[N s_ii/(s_ii+(N-1) sum_j E_{P_U P_V}[m_ij s_ij])] + 1'. In the next line this term is removed without justification; it is not identically 1. Without it, the inequality I(U;V) >= I_NML(U;V) does not follow from the variational formula. The theorem statement is also internally inconsistent with the proof: it defines I_NCE(U;V)=L_InfoNCE+C, while the proof uses I_NCE=-E[L_InfoNCE]+C. Since Theorem 1 is the stated basis for the bi-level objective, this is a load-bearing error.
[Appendix B.2, proof of Theorem 2] The proof models the contrastive views as generated from the label variable T and independent perturbations e1,e2, leading to I(G_U;G_V)=I(T;T). This contradicts the actual method: Section 3.1 and Algorithm 1 generate G_U and G_V from the original graph G via DropEdge and FeatureMasking. In real graphs, attributes and topology contain information beyond labels, so I(G_U;G_V) generally exceeds I(T;T). Consequently, the claim that the optimal lossless encoder satisfies E[I(U*;V*)]=I(T;T) is not justified, and the conclusion z_i=z_j for all same-label pairs does not follow. Moreover, even if the mutual information equality held, it is an average statement and does not imply pointwise collapse of embeddings for every false-negative pair. Theorem 2's advertised guarantee that NML-GCL can distinguish false negatives from true negatives is therefore unsupported.
[Section 3.3, Eqs. (9)-(11)] The stationary-condition derivation for the negative metric network is not correct. Differentiating L_NML^(i) with respect to m_ik gives (N-1)s_ik/(s_ii+(N-1) sum_j m_ij s_ij), not the expression in Eq. (9), which uses s_ii in every denominator term and omits the factor N-1. As a result, Eq. (11) is not a valid solution of the stated first-order condition. This matters because the inverse proportionality between m_ik and theta(u_i,v_k) is used to argue that the encoder and NMN reinforce each other in bi-level training.
[Section 4.1, Theorem 1] The claim that I_NML(U;V) >= I_NCE(U;V) is essentially by construction: the uniform weighting m_ij=1/(N-1) with m_ii=0 is a feasible choice in Eq. (5), so the optimum over M is at least the InfoNCE objective. This only shows a comparison between two training objectives, not that the resulting value is a valid tighter lower bound on the true mutual information. Establishing the latter requires the missing penalty-term argument from Appendix B.1, which is not supplied.

minor comments (4)

[Abstract and Section 1] The phrase 'solid theoretical analysis' overstates the current state of the proofs, given the issues in Theorem 1 and Theorem 2.
[Section 3.3, text near Eq. (7)] Line 8 of Algorithm 1 refers to updating M according to Eq. (7), but Eq. (7) is the joint bi-level objective for both E and M; the intended update rule appears to be the inner minimization of Eq. (4)-(5). Please clarify the reference.
[Section 4.1, Theorem 1 statement] The definition I_NCE(U;V)=L_InfoNCE+C conflicts with the proof's use of -E[L_InfoNCE]+C. This sign inconsistency should be corrected.
[Appendix E, Table 4] The table reports hyper-parameter ranges but does not specify how alpha, T_M, and T_E were selected; for reproducibility, a short statement on the selection criterion would be helpful.

Circularity Check

2 steps flagged · score 5.0 of 10

Theoretical core is partly construction-based: the tighter-bound claim is built into the loss definition, and Theorem 2's premise already contains the false-negative collapse it claims to prove.

self definitional [Appendix B.1, proof of Theorem 1 ('Tighter than InfoNCE') and Section 4.1, Theorem 1]
"the inequality holds because InfoNCE simply sets mij = 0 if i = j otherwise mij = 1/(N−1) while INML seeks the optimum mij ∈ [0,1]."

The paper defines I_NML as -min_M E_i[L_NML] + C, and L_NML with m_ij=0 for i=j and m_ij=1/(N-1) otherwise is exactly L_InfoNCE. Therefore I_NML >= I_NCE is an algebraic consequence of taking a minimum over a family of weight matrices that includes the uniform InfoNCE weights; the inequality is forced by the definition of the objective itself, not by a property of graph encoders or of mutual information. The only substantive part is whether I(U;V) >= I_NML holds for the data-dependent learned m, and the proof's step replacing an expectation over P_U P_V with the empirical sum over j∈V is not justified as an inequality, so the 'tighter lower bound' claim reduces, as written, to 'the optimized weighted loss cannot be worse than the uniform-weight loss on the training objective.'
self definitional [Appendix B.2, proof of Theorem 2; compared with Algorithm 1 / Section 3.1 view generation]
"the optimal E∗ satisfies Ee1,e2[I(U∗;V∗)] = I(T;T), which indicates that E∗ can simulate to the oracle label function Y in Definition 1 and thus ∀i∈V, j∈S+i, zi = zj holds."

The proof constructs the view-generation process as {T,e1}->G_U->U and {T,e2}->G_V->V, so that E[I(G_U;G_V)] = I(T;T) follows from the chain rule plus independence of e1,e2. This premise already assumes that the original graph G contributes no information to the views beyond the label variable T, which is exactly the condition under which same-label nodes carry no distinguishing shared signal. The conclusion that an optimal lossless encoder maps every false negative to the anchor's embedding is therefore the content of the premise restated after an encoder step, not a property derived from the NML-GCL objective.

full rationale

The paper contains no load-bearing self-citation chain: the references are standard external works, and the experimental comparisons are independent. The circularity is confined to the theoretical claims. Theorem 1's assertion that NML gives a tighter bound than InfoNCE is partly definitional, because the NML objective's feasible set contains the uniform InfoNCE weighting, so I_NML >= I_NCE is built into the min over M; the genuinely nontrivial lower-bound step I(U;V) >= I_NML is asserted through a Donsker-Varadhan substitution whose empirical replacement is not shown to preserve the inequality. Theorem 2 is stronger but rests on a generative premise that already assumes the views share only label information, making the false-negative collapse it announces true by construction of the premise rather than by the proposed bi-level training; the real augmentation pipeline generates views from G, so the equality I(G_U;G_V)=I(T;T) does not hold for the method as implemented. Because the experiments are external and the method's practical gains could in principle be real, the overall circularity is partial rather than total: score 5.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central methodological novelty is a learned negative-weighting network. The method's free parameters are standard hyperparameters tuned per dataset. The theoretical proof of Theorem 2 rests on two ad hoc assumptions in Appendix B.2 that are not stated in the theorem and do not hold for real graphs.

free parameters (4)

regularization weight alpha = 0.05 to 0.2 per dataset (Table 4)
Controls the KL regularization that prevents the negative weight distribution from collapsing; tuned per dataset.
temperature tau = 0.4 to 0.8 per dataset
Temperature in the InfoNCE-style losses; tuned per dataset.
inner iterations T_M = 2 to 8 per dataset
Number of update steps for the negative metric network in the alternating scheme; tuned per dataset.
outer epochs T_E = 50 to 200 per dataset
Training length; tuned per dataset.

assumptions (3)

standard math Variational representation of mutual information (Nguyen et al., 2010, Lemma 1)
Used to express I(U;V) as a supremum over functions f; standard, but the subsequent substitution of a data-dependent m_ij into f is not justified.
ad hoc to paper The augmented views G_U and G_V are generated from the label T and independent perturbations e1 and e2, with the original graph carrying no information beyond T
Appendix B.2 uses the Markov chain {T,e1}->G_U->U, which implies I(G_U;G_V)=I(T;T). This assumption is false for real graphs and is not stated in the theorem.
ad hoc to paper An encoder that achieves I(U*;V*)=I(G_U;G_V) maps all false negatives to the same embedding (z_i = z_j for same-label nodes)
The proof leaps from lossless encoding to equality of embeddings for false negatives; maximizing MI does not require same-class nodes to collapse to a point.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Negative Metric Learning for Graphs." pith.science (2026). https://pith.science/paper/QBTTZWFC

@misc{pith2026250510307,
  author       = {Pith},
  title        = {Pith review of: Negative Metric Learning for Graphs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QBTTZWFC}},
  note         = {Machine review of arXiv:2505.10307}
}

read the original abstract

Graph contrastive learning (GCL) often suffers from false negatives, which degrades the performance on downstream tasks. The existing methods addressing the false negative issue usually rely on human prior knowledge, still leading GCL to suboptimal results. In this paper, we propose a novel Negative Metric Learning (NML) enhanced GCL (NML-GCL). NML-GCL employs a learnable Negative Metric Network (NMN) to build a negative metric space, in which false negatives can be distinguished better from true negatives based on their distance to anchor node. To overcome the lack of explicit supervision signals for NML, we propose a joint training scheme with bi-level optimization objective, which implicitly utilizes the self-supervision signals to iteratively optimize the encoder and the negative metric network. The solid theoretical analysis and the extensive experiments conducted on widely used benchmarks verify the superiority of the proposed method.

Figures

Figures reproduced from arXiv: 2505.10307 by the authors.

**Figure 2.** Overview of NML-GCL. First, NML-GCL generates two contrastive views [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 4.** Hyper-parameter analysis. 5.4 Hyper-parameter Analysis (RQ3) Now we investigate the effectiveness of the two most important hyper-parameters α in Equation (7) and the iteration number TM of inner minimization in Algorithm 1. We take the Photo dataset as an example. Fig. 4a shows that as α increases, the classification accuracy first rises and then falls. Since α is the weight of the regularization term in Equation… view at source ↗

Figures from the paper (3 more)

**Figure 5.** Figure 5: Predicted weights of False Negatives (FN) and True Negatives (TN) during training of NML-GCL. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

**Figure 6.** Figure 6: Similarity distributions of negatives with anchors. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]

**Figure 7.** Figure 7: Mutual information estimation. with the conclusion of Theorem 1. We also present in Table 3 the ratio of the median distance of false negatives and true negatives to anchor nodes. From [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 44 canonical work pages

[1]

An introduction to single-user information theory

[Alajaji et al., 2018] Fady Alajaji, Po-Ning Chen, et al. An introduction to single-user information theory . Springer,

work page 2018
[5]

Enhancing contrastive learning on graphs with node similarity

[Chi and Ma, 2024] Hongliang Chi and Yao Ma. Enhancing contrastive learning on graphs with node similarity. InPro- ceedings of the 30th ACM SIGKDD Conference on Knowl- edge Discovery and Data Mining, pages 456–465,

work page 2024
[7]

Neighborhood-based hard negative min- ing for sequential recommendation

[Fan et al., 2023] Lu Fan, Jiashu Pu, Rongsheng Zhang, and Xiao-Ming Wu. Neighborhood-based hard negative min- ing for sequential recommendation. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages 2042– 2046,

work page 2023
[11]

Graph representa- tion learning

[Hamilton, 2020] William L Hamilton. Graph representa- tion learning. Morgan & Claypool Publishers,

work page 2020
[12]

Topology-aware debiased self-supervised graph learning for recommendation

[Han et al., 2023] Lei Han, Hui Yan, and Zhicheng Qiao. Topology-aware debiased self-supervised graph learning for recommendation. In 2023 IEEE International Con- ference on Systems, Man, and Cybernetics (SMC) , pages 2339–2344. IEEE,

work page 2023
[13]

Towards expansive and adaptive hard negative mining: Graph con- trastive learning via subspace preserving

[Hao et al., 2024] Zhezheng Hao, Haonan Xin, Long Wei, Liaoyuan Tang, Rong Wang, and Feiping Nie. Towards expansive and adaptive hard negative mining: Graph con- trastive learning via subspace preserving. In Proceedings of the ACM on Web Conference 2024,

work page 2024
[14]

Contrastive multi-view repre- sentation learning on graphs

[Hassani and Khasahmadi, 2020] Kaveh Hassani and Amir Hosein Khasahmadi. Contrastive multi-view repre- sentation learning on graphs. In International conference on machine learning, pages 4116–4126. PMLR,

work page 2020
[15]

Graph-mlp: Node classification without message passing in graph

[Hu et al., 2021] Yang Hu, Haoxuan You, Zhecan Wang, Zhicheng Wang, Erjin Zhou, and Yue Gao. Graph-mlp: Node classification without message passing in graph. arXiv preprint arXiv:2106.04051,

arXiv 2021

Show all 56 references

[17]

Variational graph auto-encoders

[Kipf and Welling, 2016] Thomas N Kipf and Max Welling. Variational graph auto-encoders. arXiv preprint arXiv:1611.07308,

2016 arXiv
[19]

Homogcl: Rethinking ho- mophily in graph contrastive learning

[Li et al., 2023] Wen-Zhi Li, Chang-Dong Wang, Hui Xiong, and Jian-Huang Lai. Homogcl: Rethinking ho- mophily in graph contrastive learning. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Dis- covery and Data Mining, pages 1341–1352,

2023
[20]

Spectral augmentation for self-supervised learning on graphs

[Lin and Chen, 2023] Lu Lin and Jinghui Chen. Spectral augmentation for self-supervised learning on graphs. In The Eleventh International Conference on Learning Rep- resentations,

2023
[21]

Prototypical graph contrastive learning

[Lin et al., 2022] Shuai Lin, Chen Liu, Pan Zhou, Zi-Yuan Hu, Shuojia Wang, Ruihui Zhao, Yefeng Zheng, Liang Lin, Eric Xing, and Xiaodan Liang. Prototypical graph contrastive learning. IEEE transactions on neural net- works and learning systems, 35(2):2747–2758,

2022
[22]

B2-sampling: Fusing balanced and biased sampling for graph contrastive learning

[Liu et al., 2023] Mengyue Liu, Yun Lin, Jun Liu, Bohao Liu, Qinghua Zheng, and Jin Song Dong. B2-sampling: Fusing balanced and biased sampling for graph contrastive learning. In Proceedings of the 29th ACM SIGKDD Con- ference on Knowledge Discovery and Data Mining, pages 1489–1500,

2023
[23]

Seeking false hard negatives for graph contrastive learning

[Liu et al., 2024] Xin Liu, Biao Qian, Haipeng Liu, Dan Guo, Yang Wang, and Meng Wang. Seeking false hard negatives for graph contrastive learning. IEEE Transac- tions on Circuits and Systems for Video Technology,

2024
[24]

Wiki-cs: A wikipedia-based benchmark for graph neural networks

[Mernyei and Cangea, 2020] P´eter Mernyei and C ˘at˘alina Cangea. Wiki-cs: A wikipedia-based benchmark for graph neural networks. arXiv preprint arXiv:2007.02901,

2020 arXiv
[25]

Estimating divergence functionals and the likelihood ratio by convex risk min- imization

[Nguyen et al., 2010] XuanLong Nguyen, Martin J Wain- wright, and Michael I Jordan. Estimating divergence functionals and the likelihood ratio by convex risk min- imization. IEEE Transactions on Information Theory , 56(11):5847–5861,

2010
[27]

Graph representation learning via graphical mutual information maximization

[Peng et al., 2020] Zhen Peng, Wenbing Huang, Minnan Luo, Qinghua Zheng, Yu Rong, Tingyang Xu, and Jun- zhou Huang. Graph representation learning via graphical mutual information maximization. In Proceedings of The Web Conference 2020, pages 259–270,

2020
[28]

Deepwalk: Online learning of social repre- sentations

[Perozzi et al., 2014] Bryan Perozzi, Rami Al-Rfou, and Steven Skiena. Deepwalk: Online learning of social repre- sentations. In Proceedings of the 20th ACM SIGKDD in- ternational conference on Knowledge discovery and data mining, pages 701–710,

2014
[31]

Dropedge: Towards deep graph con- volutional networks on node classification

[Rong et al., 2020] Yu Rong, Wenbing Huang, Tingyang Xu, and Junzhou Huang. Dropedge: Towards deep graph con- volutional networks on node classification. In Interna- tional Conference on Learning Representations,

2020
[32]

Pitfalls of graph neural network evalu- ation

[Shchur et al., 2018] Oleksandr Shchur, Maximilian Mumme, Aleksandar Bojchevski, and Stephan G¨unnemann. Pitfalls of graph neural network evalu- ation. Relational Representation Learning Workshop, NeurIPS 2018,

2018
[33]

Properties of the hubert- arable adjusted rand index

[Steinley, 2004] Douglas Steinley. Properties of the hubert- arable adjusted rand index. Psychological methods , 9(3):386,

2004
[35]

Adversarial graph augmentation to im- prove graph contrastive learning

[Suresh et al., 2021] Susheel Suresh, Pan Li, Cong Hao, and Jennifer Neville. Adversarial graph augmentation to im- prove graph contrastive learning. Advances in Neural In- formation Processing Systems, 34:15920–15933,

2021
[36]

Bootstrapped representa- tion learning on graphs

[Thakoor et al., 2021] Shantanu Thakoor, Corentin Tallec, Mohammad Gheshlaghi Azar, Remi Munos, Petar Veliˇckovi´c, and Michal Valko. Bootstrapped representa- tion learning on graphs. In ICLR 2021 Workshop on Geo- metrical and Topological Representation Learning,

2021
[37]

Large- scale representation learning on graphs via bootstrapping

[Thakoor et al., 2022] Shantanu Thakoor, Corentin Tallec, Mohammad Gheshlaghi Azar, Mehdi Azabou, Eva L Dyer, Remi Munos, Petar Veliˇckovi´c, and Michal Valko. Large- scale representation learning on graphs via bootstrapping. In International Conference on Learning Representations,

2022
[38]

Visualizing data using t-sne.Journal of machine learning research, 9(11),

[Van der Maaten and Hinton, 2008] Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne.Journal of machine learning research, 9(11),

2008
[40]

Deep graph infomax

[Veliˇckovi´c et al., 2018] Petar Veli ˇckovi´c, William Fedus, William L Hamilton, Pietro Li `o, Yoshua Bengio, and R Devon Hjelm. Deep graph infomax. arXiv preprint arXiv:1809.10341,

2018 arXiv
[41]

Boosting graph contrastive learning via adaptive sampling

[Wan et al., 2023] Sheng Wan, Yibing Zhan, Shuo Chen, Shirui Pan, Jian Yang, Dacheng Tao, and Chen Gong. Boosting graph contrastive learning via adaptive sampling. IEEE Transactions on Neural Networks and Learning Sys- tems,

2023
[42]

Theoretical analysis of self-training with deep networks on unlabeled data

[Wei et al., 2020] Colin Wei, Kendrick Shen, Yining Chen, and Tengyu Ma. Theoretical analysis of self-training with deep networks on unlabeled data. arXiv preprint arXiv:2010.03622,

2020 arXiv
[43]

Graph contrastive learning via interventional view generation

[Wo et al., 2024] Zengyi Wo, Minglai Shao, Wenjun Wang, Xuan Guo, and Lu Lin. Graph contrastive learning via interventional view generation. InProceedings of the ACM on Web Conference 2024, pages 1024–1034,

2024
[44]

Progcl: Rethinking hard nega- tive mining in graph contrastive learning

[Xia et al., 2022] Jun Xia, Lirong Wu, Ge Wang, Jintao Chen, and Stan Z Li. Progcl: Rethinking hard nega- tive mining in graph contrastive learning. In International Conference on Machine Learning,

2022
[45]

Revisiting semi-supervised learning with graph embeddings

[Yang et al., 2016] Zhilin Yang, William Cohen, and Ruslan Salakhudinov. Revisiting semi-supervised learning with graph embeddings. In International conference on ma- chine learning, pages 40–48. PMLR,

2016
[46]

Region or global? a principle for negative sampling in graph-based recommendation

[Yang et al., 2022] Zhen Yang, Ming Ding, Xu Zou, Jie Tang, Bin Xu, Chang Zhou, and Hongxia Yang. Region or global? a principle for negative sampling in graph-based recommendation. IEEE Transactions on Knowledge and Data Engineering, 35(6):6264–6277,

2022
[47]

Graph con- trastive learning with augmentations

[You et al., 2020] Yuning You, Tianlong Chen, Yongduo Sui, Ting Chen, Zhangyang Wang, and Yang Shen. Graph con- trastive learning with augmentations. In Advances in neu- ral information processing systems,

2020
[48]

Topology reorganized graph contrastive learning with mit- igating semantic drift

[Zhang and Chen, 2024] Jiaqiang Zhang and Songcan Chen. Topology reorganized graph contrastive learning with mit- igating semantic drift. arXiv preprint arXiv:2407.16726,

2024 arXiv
[49]

From canonical correlation analysis to self-supervised graph neural networks

[Zhang et al., 2021] Hengrui Zhang, Qitian Wu, Junchi Yan, David Wipf, and Philip S Yu. From canonical correlation analysis to self-supervised graph neural networks. In Ad- vances in Neural Information Processing Systems,

2021
[50]

Lo- calized contrastive learning on graphs

[Zhang et al., 2022] Hengrui Zhang, Qitian Wu, Yu Wang, Shaofeng Zhang, Junchi Yan, and Philip S Yu. Lo- calized contrastive learning on graphs. arXiv preprint arXiv:2212.04604,

2022 arXiv
[51]

Empowering collabo- rative filtering with principled adversarial contrastive loss

[Zhang et al., 2024] An Zhang, Leheng Sheng, Zhibo Cai, Xiang Wang, and Tat-Seng Chua. Empowering collabo- rative filtering with principled adversarial contrastive loss. Advances in Neural Information Processing Systems , 36,

2024
[52]

Deep graph contrastive representation learning

[Zhu et al., 2020] Yanqiao Zhu, Yichen Xu, Feng Yu, Qiang Liu, Shu Wu, and Liang Wang. Deep graph contrastive representation learning. In ICML Workshop on Graph Representation Learning and Beyond,

2020
[53]

Graph contrastive learning with adaptive augmentation

[Zhu et al., 2021] Yanqiao Zhu, Yichen Xu, Feng Yu, Qiang Liu, Shu Wu, and Liang Wang. Graph contrastive learning with adaptive augmentation. In Proceedings of the web conference 2021, pages 2069–2080,

2021
[54]

Graph con- trastive learning reimagined: Exploring universality

[Zhuo et al., 2024a] Jiaming Zhuo, Can Cui, Kun Fu, Bingxin Niu, Dongxiao He, Chuan Wang, Yuanfang Guo, Zhen Wang, Xiaochun Cao, and Liang Yang. Graph con- trastive learning reimagined: Exploring universality. In Proceedings of the ACM on Web Conference 2024,

2024
[55]

, 2020a; Huynh et al

A RELATED WORK In the GCL domain, the existing methods typically address the issue of false negatives by weighting negative samples, following two technique lines: hard-weight based [Zhang et al., 2022; Wanget al., 2024b; Huet al., 2021; Liuet al., 2023; Wu et al. , 2020a; Huy...

2022
[56]

The best result on each dataset is highlighted in boldface and the runner-up results are underlined

Method Cora CiteSeer PubMed Photo Computers Wiki-CS BGRL 92.25 96.49 91.20 62.23 71.98 95.30 GRACE 89.50 95.04 91.40 60.30 63.26 74.20 MVGRL 82.05 90.48 85.71 100.28 100.26 99.37 LOCAL-GCL 92.30 96.70 92.22 77.30 71.95 80.58 PHASES 72.10 91.21 87.56 70.99 72.12 78.75 HomoGCL 8...

2000
[2004]

Progressive hard negative masking: From global uniformity to local tolerance.IEEE Transactions on Knowledge and Data Engineering , 35(12):12932–12943,

[Sun et al., 2023] Qingqiang Sun, Wenjie Zhang, and Xuemin Lin. Progressive hard negative masking: From global uniformity to local tolerance.IEEE Transactions on Knowledge and Data Engineering , 35(12):12932–12943,

2023
[2007]

A simple framework for contrastive learning of visual representations

[Chen et al., 2020] Ting Chen, Simon Kornblith, Moham- mad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. In In- ternational conference on machine learning, pages 1597–

2020
[2008]

Graph attention networks

[Veliˇckovi´c et al., 2017] Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. Graph attention networks. arXiv preprint arXiv:1710.10903,

2017 arXiv
[2010]

Affinity uncertainty-based hard negative mining in graph contrastive learning

[Niu et al., 2024] Chaoxi Niu, Guansong Pang, and Ling Chen. Affinity uncertainty-based hard negative mining in graph contrastive learning. IEEE Transactions on Neural Networks and Learning Systems,

2024
[2014]

On variational bounds of mutual information

[Poole et al., 2019] Ben Poole, Sherjil Ozair, Aaron Van Den Oord, Alex Alemi, and George Tucker. On variational bounds of mutual information. In International Confer- ence on Machine Learning , pages 5171–5180. PMLR,

2019
[2016]

Kipf and Max Welling

[Kipf and Welling, 2017] Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. In International Conference on Learning Rep- resentations,

2017
[2017]

Inductive representation learning on large graphs

[Hamilton et al., 2017] Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs. Advances in neural information processing sys- tems,

2017
[2018]

Learning from positive and unlabeled data: A survey

[Bekker and Davis, 2020] Jessa Bekker and Jesse Davis. Learning from positive and unlabeled data: A survey. Ma- chine Learning, 109(4):719–760,

2020
[2019]

Contrastive learning with hard negative samples

[Robinson et al., 2021] Joshua David Robinson, Ching-Yao Chuang, Suvrit Sra, and Stefanie Jegelka. Contrastive learning with hard negative samples. InInternational Con- ference on Learning Representations,

2021
[2020]

A fuzzy extension of the rand index and other related indexes for clustering and classification assessment

[Campello, 2007] Ricardo JGB Campello. A fuzzy extension of the rand index and other related indexes for clustering and classification assessment. Pattern Recognition Letters, 28(7):833–841,

2007
[2021]

Boosting contrastive self-supervised learning with false negative cancellation

[Huynh et al., 2022] Tri Huynh, Simon Kornblith, Matthew R Walter, Michael Maire, and Maryam Khademi. Boosting contrastive self-supervised learning with false negative cancellation. In Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages 2785–2795,

2022
[2022]

Neu- ral message passing for quantum chemistry

[Gilmer et al., 2017] Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neu- ral message passing for quantum chemistry. In Interna- tional conference on machine learning, pages 1263–1272. PMLR,

2017
[2023]

Adversarial graph contrastive learn- ing with information regularization

[Feng et al., 2022] Shengyu Feng, Baoyu Jing, Yada Zhu, and Hanghang Tong. Adversarial graph contrastive learn- ing with information regularization. In Proceedings of the ACM Web Conference 2022, pages 1362–1371,

2022
[2024]

Debiased contrastive learning

[Chuang et al., 2020] Ching-Yao Chuang, Joshua Robinson, Yen-Chen Lin, Antonio Torralba, and Stefanie Jegelka. Debiased contrastive learning. Advances in neural infor- mation processing systems, 33:8765–8775,

2020

Pith tools

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

[1] [1]

An introduction to single-user information theory

[Alajaji et al., 2018] Fady Alajaji, Po-Ning Chen, et al. An introduction to single-user information theory . Springer,

work page 2018

[2] [5]

Enhancing contrastive learning on graphs with node similarity

[Chi and Ma, 2024] Hongliang Chi and Yao Ma. Enhancing contrastive learning on graphs with node similarity. InPro- ceedings of the 30th ACM SIGKDD Conference on Knowl- edge Discovery and Data Mining, pages 456–465,

work page 2024

[3] [7]

Neighborhood-based hard negative min- ing for sequential recommendation

[Fan et al., 2023] Lu Fan, Jiashu Pu, Rongsheng Zhang, and Xiao-Ming Wu. Neighborhood-based hard negative min- ing for sequential recommendation. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages 2042– 2046,

work page 2023

[4] [11]

Graph representa- tion learning

[Hamilton, 2020] William L Hamilton. Graph representa- tion learning. Morgan & Claypool Publishers,

work page 2020

[5] [12]

Topology-aware debiased self-supervised graph learning for recommendation

[Han et al., 2023] Lei Han, Hui Yan, and Zhicheng Qiao. Topology-aware debiased self-supervised graph learning for recommendation. In 2023 IEEE International Con- ference on Systems, Man, and Cybernetics (SMC) , pages 2339–2344. IEEE,

work page 2023

[6] [13]

Towards expansive and adaptive hard negative mining: Graph con- trastive learning via subspace preserving

[Hao et al., 2024] Zhezheng Hao, Haonan Xin, Long Wei, Liaoyuan Tang, Rong Wang, and Feiping Nie. Towards expansive and adaptive hard negative mining: Graph con- trastive learning via subspace preserving. In Proceedings of the ACM on Web Conference 2024,

work page 2024

[7] [14]

Contrastive multi-view repre- sentation learning on graphs

[Hassani and Khasahmadi, 2020] Kaveh Hassani and Amir Hosein Khasahmadi. Contrastive multi-view repre- sentation learning on graphs. In International conference on machine learning, pages 4116–4126. PMLR,

work page 2020

[8] [15]

Graph-mlp: Node classification without message passing in graph

[Hu et al., 2021] Yang Hu, Haoxuan You, Zhecan Wang, Zhicheng Wang, Erjin Zhou, and Yue Gao. Graph-mlp: Node classification without message passing in graph. arXiv preprint arXiv:2106.04051,

arXiv 2021