Pith. sign in

REVIEW 5 major objections 6 minor 107 references

Directed Link Prediction using GNN with Local and Global Feature Fusion

T0 review · 5 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper proves that fusing community-based labels with path and embedding features raises directed link prediction accuracy under a stochastic blockmodel, with gains over baselines on six directed networks.

desk verdict The empirical package is workmanlike and the low-training results are worth a look, but the theoretical centerpiece is invalid and the paper overstates what it proves. read the letter →

arxiv 2506.20235 v1 pith:WHJP3LNV submitted 2025-06-25 cs.LG cs.AI

Yuyang Zhang , Xu Shen , Yu Xie , Ka-Chun Wong , Weidun Xie , Chengbin Peng This is my paper
classification cs.LGcs.AI
keywords directedlinkpredictiongraphneuralnetworkcommunitydetectionlinenodeembeddingfeaturefusionstochasticblockmodelcontrastivelearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that directed link prediction improves when a graph neural network fuses three kinds of node features: path-based labels capturing local structure, community labels capturing mid-range structure, and contrastive embeddings capturing global structure. The theoretical centerpiece is a theorem stating that, for graphs generated by a stochastic blockmodel, a learnable weighted combination of a community-based predictor and a non-community-based predictor has higher expected accuracy than the non-community predictor alone, provided a stated condition on link probabilities and community-detection error rates holds. To feed these fused features into a GNN, the paper converts each directed link into a node of a directed line graph, turning link prediction into node classification. On six benchmark networks, the resulting FFD model attains the highest or second-highest AUC and AP against six GNN autoencoder baselines at 30%, 40%, 50%, and 60% training-link fractions, and converges in fewer epochs. The paper concludes that fusing community information with local and global features is a viable route to stronger directed link prediction.

What carries the argument

The load-bearing object is the hybrid node fusion: path-based labels OP, community-based labels OC, and contrastive embeddings OE are concatenated into a per-node feature vector OH, which is then assigned to each directed link represented as a line-graph node. Theorem 1 is the theoretical mechanism, stating under the SBM assumptions that the expected correct-prediction probability of the weighted fusion is the sum of the two predictors' expected accuracies, and the inequality in Eq. (7) guarantees the fusion beats the non-community predictor alone. The directed line graph transformation is the operational machinery: a subgraph around the target link is converted so that each original link becomes a node and links connect consecutive directed edges, giving each convolution a wider receptive field at the price of a graph roughly d times larger, where d is the average degree. These two components together support the claim that fusing local, community, and global signals helps directed link prediction.

What would settle it

Generate stochastic blockmodel graphs with p and q satisfying Eq. (7), train the community-based and non-community-based predictors, and compare the empirical accuracy of the weighted fusion with the weighted sum of the two individual accuracies; if the fused accuracy is consistently below the sum, the no-overlap premise of Proof 1 is violated.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that community-based node labels carry information that path-based labels and contrastive embeddings do not, and that fusing all three improves directed link prediction. The proof rests on a stochastic blockmodel with K equal-size communities, intra-community link probability p and inter-community probability q. With a community detector of error rates ε0 and ε1, the paper derives the expected correct-prediction probability of the weighted fusion and shows it exceeds that of the non-community predictor whenever the condition in Eq. (7) holds; the condition is easier to satisfy as p grows and q shrinks, matching the intuition that clearer community structure makes the fusion more helpful. The companion engineering contribution is a directed line graph transformation: each directed link (vi, vj) becomes a node whose features concatenate the fused features of its source and destination, so graph convolution runs on link-to-link adjacency and the prediction of a link is the prediction of the corresponding line-graph node. Experiments compare FFD with six GNN autoencoder baselines and more than twenty additional methods on six directed networks, reporting the best or second-best AUC/AP in most configurations and larger average gains when training samples are scarce.

Load-bearing premise

The guarantee stands on the premise that the community detector recovers the true block structure and that the two predictors never err on the same node pair, so their correct predictions simply add up—if either fails, the theorem's bound does not follow.

Editorial extensions

If this is right

  • If Theorem 1 holds, then on graphs with detectable community structure any directed link predictor should benefit from augmenting its features with community-affiliation labels, not just path and embedding features.
  • The directed line graph formulation implies that link prediction can be solved as node classification, so improvements to node-level GNN architectures transfer directly to link prediction.
  • At 30%-40% training data, the reported gains over autoencoder baselines are larger than at 60%, so the hybrid fusion appears most valuable when local supervision is scarce.
  • On weakly-structured networks such as Bitcoin and p2p-Gnutella04, the gains are smaller, consistent with the theorem's condition that clearer community structure (larger p, smaller q) yields a larger improvement.

Reading between the lines

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

  • A testable extension is that Eq. (7) defines a threshold: on a graph where p is close to q, fusing community labels could actually reduce accuracy, so practitioners should measure community clarity before adopting the fusion.
  • The line graph blow-up factor of average degree suggests a memory and runtime cost that grows with density; the paper reports convergence speed but not this scaling, so a complexity benchmark on dense directed graphs would clarify when the method is practical.
  • Because the proof adds the two predictors' accuracies, any real overlap in their errors will shrink the fusion gain below the theoretical prediction; an experiment that measures this overlap directly would probe the mechanism.
  • The same fusion recipe—community labels plus local path labels plus global embeddings—could be applied to other relational tasks such as directed community-aware node classification or edge-level anomaly detection.
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 / 6 minor

Summary. The paper proposes FFD, a graph neural network framework for directed link prediction that concatenates path-based node labels, community-based node labels, and contrastive node embeddings, transforms the resulting subgraph into a directed line graph, and applies graph convolutions. It claims a theoretical guarantee (Theorem 1) that fusing community-based labels with non-community features improves the probability of correct prediction, and it reports experiments on six datasets showing AUC and AP improvements over baseline methods at 30%, 40%, 50%, and 60% training-link fractions. The central theoretical claim is unsupported because the proof of Theorem 1 contains fundamental errors, including the addition of conditional probabilities from disjoint events and a sign error.

Significance. If Theorem 1 were valid, it would provide a nontrivial provable improvement result for community-aware directed link prediction, and the directed line-graph transformation is a reasonable engineering contribution. The paper also includes ablation studies and evaluates multiple training fractions, which are useful empirical practices. However, the proof of Theorem 1 is invalid for reasons detailed below, so the advertised theoretical support collapses. The empirical evaluation may still be of interest, but without Theorem 1 the paper is a purely empirical study, and the claim in the abstract that hybrid features are 'theoretically demonstrated' to improve directed link prediction is not supported.

major comments (5)
  1. [§3.3.1, Proof 1, Eqs. (13)–(17)] The quantity in Eq. (14) is not a probability. The four cases in Eq. (13) are conditional on disjoint events (A_xy = 0 or 1 and the sign of the fusion output), so their unweighted sum is not the expected accuracy. Since the benchmark set contains linked and unlinked pairs in equal proportion, the correct accuracy is (P(correct|A=1) + P(correct|A=0))/2; the missing factor of 1/2 means the left side can exceed 1. Consequently, the inequality in Eq. (17) compares an inflated sum against E_C,U + E_C,L rather than the accuracy of f(ζ_C) alone, and Theorem 1 is not established by this argument.
  2. [§3.3.1, Eqs. (14)–(15)] There is a sign error in the algebra: θ_L(p−q) + θ_U(q−p) equals (p−q)(θ_L − θ_U), not (p−q)(θ_L + θ_U) as written in Eq. (15). Replacing −θ_U by +θ_U changes the derived condition in Eq. (7), so the subsequent monotonicity analysis of g(p,q,K) in Eqs. (18)–(21) applies to a different expression from the one that actually follows from Eq. (14).
  3. [§3.3.1, Eq. (13) and Eq. (16)] The proof assumes, without stating, that the errors of the two fused predictors never overlap, so that the fused prediction is correct whenever either predictor is correct, and that the weights are normalized, W_c + W_c = 1. In general, a weighted sum of conditional expectation terms is not the probability that the weighted fusion is correct; if the two predictors make errors on the same pairs, the fused correct-prediction rate can be lower than the weighted sum. The condition in Eq. (7) does not include any non-overlap or independence assumption, so it is insufficient for the claimed conclusion.
  4. [§3.3.1, Eq. (7)] The theorem's condition is stated in terms of the learned fusion weight W_c and the baseline quantities E_C,U and E_C,L. This makes the purported guarantee conditional on fitted parameters, and the paper gives no argument that Eq. (7) actually holds for the networks and training setups used in the experiments. As a result, the abstract's claim that hybrid features are theoretically demonstrated to improve directed link prediction is not supported by the proof.
  5. [§3.3.1, before Eq. (12)] The proof assumes that ζ_C identifies the block structure correctly, so that predicted link probabilities can be approximated by the generative SBM parameters p and q. This is an unverified assumption on real datasets, especially those with weak community structure such as Bitcoin and p2p-Gnutella04 (Table 1), and the theorem therefore does not cover the experimental settings where community structure is weak. No alternative argument is provided for those cases.
minor comments (6)
  1. [§3.3.1, Eq. (5)] The two terms in the hybrid fusion expression are written identically as W_c f(ζ_C) + W_c f(ζ_C); presumably the second term should denote the non-community-based predictor with its own weight, rather than the same symbol ζ_C.
  2. [§3.3.1, Definition 2] Definition 2 contains a typo ('is defined as as ζ_C') and creates notation confusion: ζ_C was already introduced in Eq. (2) as the community detection algorithm, but Definition 2 says a non-community-based link prediction algorithm is ζ_C.
  3. [§4.3, Fig. 4] The text says that g(p,q,K)+1 is plotted, while the caption of Fig. 4 says g(p,q,K)−1; these should be reconciled.
  4. [§4.3, Table 10] The header 'AUC APP' contains an extra 'P', and the table is difficult to parse because the column headers are not aligned with the row entries.
  5. [§3.3.3, Eq. (27)] The loss function uses the symbol '⊑k' for a node in the transformed graph; this appears to be a rendering error and should be a standard node index.
  6. [§4.2] No statement is made about code availability, which limits the reproducibility of the empirical comparisons.

Circularity Check

2 steps flagged · score 6.0 of 10

Theorem 1's promised improvement is conditional on Eq. (7), which is the target inequality restated, and the proof's sign flip manufactures the gain rather than deriving it.

  1. self definitional [Section 3.3.1, Theorem 1 / Proof 1, Eqs. (7), (14)-(17)]
    "if the following condition is satisfied (p−q)( p/(p+(K−1)q) + (1−p)/(K−[p+(K−1)q]) ) + 1−ε0−ε1 ≥ (1−Wc)/Wc (EC,U +E C,L ). The probability of making correct predictions using hybrid node fusion Wcf(ζC)+Wcf(ζC) is expected to be higher than that of using non-community-based labeling f(ζC) only."

    The proof's Eq. (14) writes the hybrid expectation as Wc(E_C,U+E_C,L)+Wc[G]; Eq. (16) replaces the second bracket with (1−Wc)(E_C,U+E_C,L) explicitly “given that the condition of Eq.(7) is satisfied”, and Eq. (17) concludes the hybrid is at least E_C,U+E_C,L. Thus Eq. (7) is not derived from the SBM; it is exactly the inequality needed to reach the theorem's conclusion after cancelling the common baseline term. The abstract's claim that hybrid features are “theoretically demonstrated” to improve directed link prediction therefore reduces to a conditional tautology: if the fused expression is large enough relative to the baseline, the fused expression is at least the baseline, with the learned weight Wc as the free parameter.

  2. other [Proof 1, Eqs. (12)-(15)]
    "As ζ_C can identify the block structure correctly, the predicted link probability can be approximated by the generative parameters p and q ... (12) ... =Wc(EC,U+E_C,L)+Wc[(p−q)( p/(p+(K−1)q) + (1−p)/(K−[p+(K−1)q]) ) + 1−ε0−ε1](15)"

    Under Eq. (12), E_C,U = θ_U(1−p)+(1−θ_U)(1−q)−ε0 and E_C,L = θ_L p+(1−θ_L)q−ε1. Substituted into Eq. (14), the bracket is θ_L(p−q)+θ_U(q−p)+1−ε0−ε1 = (p−q)(θ_L−θ_U)+1−ε0−ε1, whereas Eq. (15) writes (p−q)(θ_L+θ_U)+1−ε0−ε1. This sign flip of the θ_U term is what creates the apparent gain; without it the bracket collapses to E_C,U+E_C,L and the hybrid equals the baseline exactly. The claimed theoretical improvement is therefore an artifact of the proof's algebra, not a consequence of the SBM or of community features.

full rationale

The central theoretical claim is circular/vacuous: Theorem 1's condition Eq. (7) is the inequality the proof needs to conclude, and the proof's only bridge from Eq. (14) to Eq. (16) is to invoke that condition as given. No independent SBM argument establishes Eq. (7); the subsequent monotonicity analysis of g(p,q,K) only shows one component grows with p and shrinks with q, without comparing it to the baseline E_C,U+E_C,L. In addition, the derivation contains a sign error between Eq. (14) and Eq. (15) that manufactures the claimed improvement: the correct bracket equals the baseline, so the hybrid is no better than the baseline under the paper's own assumptions. The experimental benchmark comparisons are external and not circular, and there is no load-bearing self-citation chain; references [106] and [107] appear only as future-work suggestions. The score is 6 because the paper's promised theoretical demonstration reduces to its own condition and to a proof artifact, while the empirical component remains independent.

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

The central claim rests on a set of assumptions about the data-generating process (SBM), the accuracy of modularity-based community detection, the non-overlap of errors between the fused classifiers, the normalization of fusion weights, and the informativeness of the directed line graph. The first two are domain assumptions with limited empirical support; the middle two are unstated and mathematically false in general; the last is asserted without proof. These assumptions do the heavy lifting of the theoretical section.

free parameters (4)
  • W_c (community fusion weight) = learned, not reported
    Theorem 1's condition Eq. (7) depends on W_c; the promised improvement is conditional on this learned weight. Appears in Eq. (5) and Eq. (13).
  • W_c (non-community fusion weight) = learned, implicitly 1 - W_c
    Eq. (16) substitutes 1 - W_c for W_c, so the proof assumes this weight is determined by the community weight. Not stated in the paper.
  • GNN and FCN network parameters = learned, unreported
    The classifier B = FCN(H_L) and GNN in Eqs. (25)-(26) are trained; all predictions depend on these fitted weights.
  • FFD and DiGCL training hyperparameters = lr 0.005, batch 50; DiGCL lr 0.001, dropout 0.3, weight decay 1e-5
    Chosen by hand in Section 4.2; no sensitivity analysis is reported, and the central results depend on these choices.
assumptions (5)
  • domain assumption The input network is generated by a stochastic blockmodel with K equal-size communities, intra-community connection probability p, and inter-community probability q.
    Definition 1, Section 3.3.1. All theoretical results are derived for SBM graphs, but the experiments use citation, transaction, and peer-to-peer networks that are not known to follow this model.
  • ad hoc to paper The community detection algorithm zeta_C identifies block structure correctly, so predicted link probability can be approximated by the generative parameters p and q.
    Proof 1, sentence before Eq. (12): 'As zeta_C can identify the block structure correctly...'. Modularity maximization is not guaranteed to recover ground truth on real graphs.
  • ad hoc to paper Errors of the two fused predictors are non-overlapping, so the correct-prediction probability of the weighted sum equals the weighted sum of the individual correct-prediction probabilities.
    Used in Eqs. (13)-(14) of Proof 1. The paper provides no justification; in general the error events overlap.
  • ad hoc to paper The fusion weights are normalized, W_c + W_c = 1.
    Eq. (16) requires 1 - W_c in place of W_c; the normalization is never stated. The conclusion fails without it.
  • domain assumption The directed line graph transformation preserves all information needed for link prediction and adds useful link-to-link context for GCNs.
    Section 3.3.2 and contribution 2. No proof is given; the transformed graph is d_hat times larger, and the paper does not demonstrate that this improves aggregation over standard GCNs on the original graph.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Directed Link Prediction using GNN with Local and Global Feature Fusion." pith.science (2026). https://pith.science/paper/WHJP3LNV

@misc{pith2026250620235,
  author       = {Pith},
  title        = {Pith review of: Directed Link Prediction using GNN with Local and Global Feature Fusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WHJP3LNV}},
  note         = {Machine review of arXiv:2506.20235}
}
read the original abstract

Link prediction is a classical problem in graph analysis with many practical applications. For directed graphs, recently developed deep learning approaches typically analyze node similarities through contrastive learning and aggregate neighborhood information through graph convolutions. In this work, we propose a novel graph neural network (GNN) framework to fuse feature embedding with community information. We theoretically demonstrate that such hybrid features can improve the performance of directed link prediction. To utilize such features efficiently, we also propose an approach to transform input graphs into directed line graphs so that nodes in the transformed graph can aggregate more information during graph convolutions. Experiments on benchmark datasets show that our approach outperforms the state-of-the-art in most cases when 30%, 40%, 50%, and 60% of the connected links are used as training data, respectively.

Figures

Figures reproduced from arXiv: 2506.20235 by the authors.

Figure 1
Figure 1. An illustration of the proposed FFD. The target link is marked as a directed red dash line, and the corresponding target node in the directed [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The feature of each node, for example, (vi , vj ) in the line graph, is obtained by concatenating corresponding node features, [OH[i], OH[j]], in the input graph. If a subgraph has nˆ nodes with average degree ˆd, the corresponding number of nodes in the transformed directed line graph is nˆ ˆd. Thus, for each graph convolution operation, the transformed input graph is ˆd times larger than the original input graph. … view at source ↗
Figure 2
Figure 2. Illustration of a directed graph transformed into a directed line [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figures from the paper (2 more)
Figure 3
Figure 3. Figure 3: AUC and AP results compared to baseline methods (50Tr). [PITH_FULL_IMAGE:figures/full_fig_p010_3.png]
Figure 4
Figure 4. Figure 4: The values of g(p, q, K) − 1 corresponding to different values of p and q when K is fixed. neural networks usually learn the neighbors around target links or learn the global features in a directed graph. In this work, we use a feature fusion method for directed link p…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

107 extracted references · 67 canonical work pages

  1. [1]

    A pattern based supervised link predic- tion in directed complex networks,

    E. B ¨ut ¨un and M. Kaya, “A pattern based supervised link predic- tion in directed complex networks,”Physica A: Statistical Mechan- ics and its Applications, vol. 525, pp. 1136–1145, 2019

    2019

  2. [2]

    Predictive network representation learning for link prediction,

    Z. Wang, C. Chen, and W. Li, “Predictive network representation learning for link prediction,” inProceedings of the 40th international ACM SIGIR conference on research and development in information retrieval, 2017, pp. 969–972

    2017

  3. [3]

    Level-2 node clustering coefficient-based link prediction,

    A. Kumar, S. S. Singh, K. Singh, and B. Biswas, “Level-2 node clustering coefficient-based link prediction,”Applied Intelligence, vol. 49, no. 7, pp. 2762–2779, 2019

    2019

  4. [4]

    Collaborative filtering approach to link prediction,

    Y.-L. Lee and T. Zhou, “Collaborative filtering approach to link prediction,”Physica A: Statistical Mechanics and its Applications, vol. 578, p. 126107, 2021

    2021

  5. [5]

    Applying asp for knowledge-based link prediction with explanation generation in feature-rich networks,

    C ¸ . G ¨uven, D. Seipel, and M. Atzmueller, “Applying asp for knowledge-based link prediction with explanation generation in feature-rich networks,”IEEE Transactions on Network Science and Engineering, vol. 8, no. 2, pp. 1305–1315, 2020

    2020

  6. [6]

    Correcting exposure bias for link recommendation,

    S. Gupta, H. Wang, Z. Lipton, and Y. Wang, “Correcting exposure bias for link recommendation,” inInternational Conference on Machine Learning. PMLR, 2021, pp. 3953–3963

    2021

  7. [7]

    Extension of neighbor-based link prediction methods for directed, weighted and temporal social networks,

    E. B ¨ut ¨un, M. Kaya, and R. Alhajj, “Extension of neighbor-based link prediction methods for directed, weighted and temporal social networks,”Information Sciences, vol. 463, pp. 152–165, 2018

    2018

  8. [8]

    Link prediction in signed social networks: From status theory to motif families,

    S.-Y. Liu, J. Xiao, and X.-K. Xu, “Link prediction in signed social networks: From status theory to motif families,”IEEE Transactions on Network Science and Engineering, vol. 7, no. 3, pp. 1724–1735, 2019

    2019

Show all 107 references
  1. [9]

    Domain knowledge-based link prediction in customer-product bipartite graph for product recommendation,

    L. Zhang, J. Li, Q. Zhang, F. Meng, and W. Teng, “Domain knowledge-based link prediction in customer-product bipartite graph for product recommendation,”International Journal of Infor- mation Technology & Decision Making, vol. 18, no. 01, pp. 311–338, 2019

    2019

  2. [10]

    Se-gru: Structure embedded gated recurrent unit neural networks for temporal link prediction,

    Y. Yin, Y. Wu, X. Yang, W. Zhang, and X. Yuan, “Se-gru: Structure embedded gated recurrent unit neural networks for temporal link prediction,”IEEE Transactions on Network Science and Engineering, vol. 9, no. 4, pp. 2495–2509, 2022

    2022

  3. [11]

    Nhp: Neural hypergraph link prediction,

    N. Yadati, V . Nitin, M. Nimishakavi, P . Yadav, A. Louis, and P . Talukdar, “Nhp: Neural hypergraph link prediction,” inPro- ceedings of the 29th ACM International Conference on Information & Knowledge Management, 2020, pp. 1705–1714

    2020

  4. [12]

    Modeling relational data with graph convolutional networks,

    M. Schlichtkrull, T. N. Kipf, P . Bloem, R. v. d. Berg, I. Titov, and M. Welling, “Modeling relational data with graph convolutional networks,” inEuropean semantic web conference. Springer, 2018, pp. 593–607

    2018

  5. [13]

    Pme: projected metric embedding on heterogeneous networks for link prediction,

    H. Chen, H. Yin, W. Wang, H. Wang, Q. V . H. Nguyen, and X. Li, JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 12 “Pme: projected metric embedding on heterogeneous networks for link prediction,” inProceedings of the 24th ACM SIGKDD international conference on knowle...

    2015

  6. [14]

    Boostgap- fill: improving the fidelity of metabolic network reconstructions through integrated constraint and pattern-based methods,

    T. Oyetunde, M. Zhang, Y. Chen, Y. Tang, and C. Lo, “Boostgap- fill: improving the fidelity of metabolic network reconstructions through integrated constraint and pattern-based methods,”Bioin- formatics, vol. 33, no. 4, pp. 608–611, 2017

    2017

  7. [15]

    Link prediction based on graph neu- ral networks,

    M. Zhang and Y. Chen, “Link prediction based on graph neu- ral networks,”Advances in neural information processing systems, vol. 31, 2018

    2018

  8. [16]

    Link prediction using node information on local paths,

    F. Aziz, H. Gul, I. Muhammad, and I. Uddin, “Link prediction using node information on local paths,”Physica A: Statistical Mechanics and its Applications, vol. 557, p. 124980, 2020

    2020

  9. [17]

    Link prediction based on local information considering preferential attachment,

    S. Zeng, “Link prediction based on local information considering preferential attachment,”Physica A: Statistical Mechanics and its Applications, vol. 443, pp. 537–542, 2016

    2016

  10. [18]

    A link prediction similarity index based on enhanced local path method,

    W. Chen and Y. Zhou, “A link prediction similarity index based on enhanced local path method,” in2021 40th Chinese Control Conference (CCC). IEEE, 2021, pp. 753–757

    2021

  11. [19]

    Efficient simrank-based similarity join over large graphs,

    W. Zheng, L. Zou, Y. Feng, L. Chen, and D. Zhao, “Efficient simrank-based similarity join over large graphs,”Proceedings of the VLDB Endowment, vol. 6, no. 7, pp. 493–504, 2013

    2013

  12. [20]

    A link pre- diction method based on topological nearest-neighbors similarity in directed networks,

    F. Guo, W. Zhou, Z. Wang, C. Ju, S. Ji, and Q. Lu, “A link pre- diction method based on topological nearest-neighbors similarity in directed networks,”Journal of Computational Science, vol. 69, p. 102002, 2023

    2023

  13. [21]

    A hybrid method of link prediction in directed graphs,

    H. Ghorbanzadeh, A. Sheikhahmadi, M. Jalili, and S. Sulaimany, “A hybrid method of link prediction in directed graphs,”Expert Systems with Applications, vol. 165, p. 113896, 2021

    2021

  14. [22]

    Financial contagion in banking net- works with community structure,

    G. Torri and R. Giacometti, “Financial contagion in banking net- works with community structure,”Communications in Nonlinear Science and Numerical Simulation, vol. 117, p. 106924, 2023

    2023

  15. [23]

    Graph regularized nonnegative matrix factorization for community detection in attributed networks,

    K. Berahmand, M. Mohammadi, F. Saberi-Movahed, Y. Li, and Y. Xu, “Graph regularized nonnegative matrix factorization for community detection in attributed networks,”IEEE Transactions on Network Science and Engineering, vol. 10, no. 1, pp. 372–385, 2022

    2022

  16. [24]

    A novel metric for community detection,

    K.-k. Shang, M. Small, Y. Wang, D. Yin, and S. Li, “A novel metric for community detection,”Europhysics Letters, vol. 129, no. 6, p. 68002, 2020

    2020

  17. [25]

    Stacking models for nearly optimal link prediction in complex networks,

    A. Ghasemian, H. Hosseinmardi, A. Galstyan, E. M. Airoldi, and A. Clauset, “Stacking models for nearly optimal link prediction in complex networks,”Proceedings of the National Academy of Sciences, vol. 117, no. 38, pp. 23 393–23 400, 2020

    2020

  18. [26]

    Network structure from rich but noisy data,

    M. Newman, “Network structure from rich but noisy data,” Nature Physics, vol. 14, pp. 542–545, 2018

    2018

  19. [27]

    Finding missing edges in networks based on their community structure,

    B. Yan and S. Gregory, “Finding missing edges in networks based on their community structure,”Phys. Rev. E, vol. 85, p. 056112, May 2012

    2012

  20. [28]

    Link prediction for long-circle-like networks,

    K.-k. Shang and M. Small, “Link prediction for long-circle-like networks,”Physical review E, vol. 105, no. 2, p. 024311, 2022

    2022

  21. [29]

    Disentangling homophily, community structure, and triadic closure in networks,

    T. P . Peixoto, “Disentangling homophily, community structure, and triadic closure in networks,”Physical Review X, vol. 12, no. 1, p. 011004, 2022

    2022

  22. [30]

    Directed graph contrastive learning,

    Z. Tong, Y. Liang, H. Ding, Y. Dai, X. Li, and C. Wang, “Directed graph contrastive learning,”Advances in Neural Information Pro- cessing Systems, vol. 34, 2021

    2021

  23. [31]

    A directed link prediction method using graph convolutional net- work based on social ranking theory,

    Z. Wu, H. Chen, J. Zhang, S. Liu, R. Huang, and Y. Pei, “A directed link prediction method using graph convolutional net- work based on social ranking theory,”Intelligent Data Analysis, vol. 25, no. 3, pp. 739–757, 2021

    2021

  24. [32]

    A weighted symmetric graph embedding approach for link pre- diction in undirected graphs,

    Z. Wang, Y. Chai, C. Sun, X. Rui, H. Mi, X. Zhang, and S. Y. Philip, “A weighted symmetric graph embedding approach for link pre- diction in undirected graphs,”IEEE Transactions on Cybernetics, 2022

    2022

  25. [33]

    Accurate link prediction method based on path length between a pair of unlinked nodes and their degree,

    J. Ayoub, D. Lotfi, M. El Marraki, and A. Hammouch, “Accurate link prediction method based on path length between a pair of unlinked nodes and their degree,”Social Network Analysis and Mining, vol. 10, pp. 1–13, 2020

    2020

  26. [34]

    Line graph neural networks for link prediction,

    L. Cai, J. Li, J. Wang, and S. Ji, “Line graph neural networks for link prediction,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021

    2021

  27. [35]

    Atomistic line graph neural network for improved materials property predictions,

    K. Choudhary and B. DeCost, “Atomistic line graph neural network for improved materials property predictions,”npj Com- putational Materials, vol. 7, no. 1, p. 185, 2021

    2021

  28. [36]

    Relation structure- aware heterogeneous graph neural network,

    S. Zhu, C. Zhou, S. Pan, X. Zhu, and B. Wang, “Relation structure- aware heterogeneous graph neural network,” in2019 IEEE inter- national conference on data mining (ICDM). IEEE, 2019, pp. 1534– 1539

    2019

  29. [37]

    Supervised community de- tection with line graph neural networks,

    Z. Chen, X. Li, and J. Bruna, “Supervised community de- tection with line graph neural networks,”arXiv preprint arXiv:1705.08415, 2017

    2017 arXiv

  30. [38]

    Variational graph auto-encoders,

    T. N. Kipf and M. Welling, “Variational graph auto-encoders,” arXiv preprint arXiv:1611.07308, 2016

    2016 arXiv

  31. [39]

    Graph contrastive learning with generative adversarial network,

    C. Wu, C. Wang, J. Xu, Z. Liu, K. Zheng, X. Wang, Y. Song, and K. Gai, “Graph contrastive learning with generative adversarial network,” inProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023, pp. 2721–2730

    2023

  32. [40]

    Link prediction based on higher-order structure extraction and autoencoder learning in directed networks,

    T. Yi, S. Zhang, Z. Bu, J. Du, and C. Fang, “Link prediction based on higher-order structure extraction and autoencoder learning in directed networks,”Knowledge-Based Systems, vol. 241, p. 108241, 2022

    2022

  33. [41]

    Toward link predictability of complex networks,

    L. L ¨u, L. Pan, T. Zhou, Y.-C. Zhang, and H. E. Stanley, “Toward link predictability of complex networks,”Proceedings of the Na- tional Academy of Sciences, vol. 112, no. 8, pp. 2325–2330, 2015

    2015

  34. [42]

    An evidential link prediction method and link predictability based on shannon entropy,

    L. Yin, H. Zheng, T. Bian, and Y. Deng, “An evidential link prediction method and link predictability based on shannon entropy,”Physica A: Statistical Mechanics and its Applications, vol. 482, pp. 699–712, 2017

    2017

  35. [43]

    A new status index derived from sociometric analysis,

    L. Katz, “A new status index derived from sociometric analysis,” Psychometrika, vol. 18, no. 1, pp. 39–43, 1953

    1953

  36. [44]

    Reprint of: The anatomy of a large-scale hypertextual web search engine,

    S. Brin and L. Page, “Reprint of: The anatomy of a large-scale hypertextual web search engine,”Computer networks, vol. 56, no. 18, pp. 3825–3833, 2012

    2012

  37. [45]

    Simrank: a measure of structural-context similarity,

    G. Jeh and J. Widom, “Simrank: a measure of structural-context similarity,” inProceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining, 2002, pp. 538– 543

    2002

  38. [46]

    Adaptive similarity function with structural features of network embedding for missing link prediction,

    C. Zhang, K.-K. Shang, and J. Qiao, “Adaptive similarity function with structural features of network embedding for missing link prediction,”Complexity, vol. 2021, no. 1, p. 1277579, 2021

    2021

  39. [47]

    Deepwalk: Online learning of social representations,

    B. Perozzi, R. Al-Rfou, and S. Skiena, “Deepwalk: Online learning of social representations,” inProceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2014, p. 701¨C710

    2014

  40. [48]

    Line: Large-scale information network embedding,

    J. Tang, M. Qu, M. Wang, M. Zhang, J. Yan, and Q. Mei, “Line: Large-scale information network embedding,” inProceedings of the 24th international conference on world wide web, 2015, pp. 1067– 1077

    2015

  41. [49]

    Network embedding as matrix factorization: Unifying deepwalk, line, pte, and node2vec,

    J. Qiu, Y. Dong, H. Ma, J. Li, K. Wang, and J. Tang, “Network embedding as matrix factorization: Unifying deepwalk, line, pte, and node2vec,” inProceedings of the eleventh ACM international conference on web search and data mining, 2018, pp. 459–467

    2018

  42. [50]

    Adversarial directed graph embedding,

    S. Zhu, J. Li, H. Peng, S. Wang, and L. He, “Adversarial directed graph embedding,” inProceedings of the AAAI conference on artifi- cial intelligence, vol. 35, no. 5, 2021, pp. 4741–4748

    2021

  43. [51]

    Directed graph auto-encoders,

    G. Kollias, V . Kalantzis, T. Id ´e, A. Lozano, and N. Abe, “Directed graph auto-encoders,” inProceedings of the AAAI conference on artificial intelligence, vol. 36, no. 7, 2022, pp. 7211–7219

    2022

  44. [52]

    Gravity-inspired graph autoencoders for directed link prediction,

    G. Salha, S. Limnios, R. Hennequin, V .-A. Tran, and M. Vazir- giannis, “Gravity-inspired graph autoencoders for directed link prediction,” inProceedings of the 28th ACM international conference on information and knowledge management, 2019, pp. 589–598

    2019

  45. [53]

    Structpool: Structured graph pooling via conditional random fields,

    H. Yuan and S. Ji, “Structpool: Structured graph pooling via conditional random fields,” inProceedings of the 8th International Conference on Learning Representations, 2020

    2020

  46. [54]

    Hierarchical graph representation learning with differentiable pooling,

    Z. Ying, J. You, C. Morris, X. Ren, W. Hamilton, and J. Leskovec, “Hierarchical graph representation learning with differentiable pooling,”Advances in neural information processing systems, vol. 31, 2018

    2018

  47. [55]

    Semi-supervised classification with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,”arXiv preprint arXiv:1609.02907, 2016

    2016 arXiv

  48. [56]

    Guest editorial: Deep neural networks for graphs: Theory, models, algorithms, and applications,

    M. Li, A. Micheli, Y. G. Wang, S. Pan, P . Li ´o, G. S. Gnecco, and M. Sanguineti, “Guest editorial: Deep neural networks for graphs: Theory, models, algorithms, and applications,”IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 4, pp. 4367–4372, 2024

    2024

  49. [57]

    Haqjsk: Hierarchical-aligned quantum jensen-shannon kernels for graph classification,

    L. Bai, L. Cui, Y. Wang, M. Li, J. Li, S. Y. Philip, and E. R. Hancock, “Haqjsk: Hierarchical-aligned quantum jensen-shannon kernels for graph classification,”IEEE Transactions on Knowledge and Data Engineering, 2024. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 13

    2024

  50. [58]

    Educross: Dual adversarial bipartite hypergraph learning for cross-modal retrieval in multimodal educational slides,

    M. Li, S. Zhou, Y. Chen, C. Huang, and Y. Jiang, “Educross: Dual adversarial bipartite hypergraph learning for cross-modal retrieval in multimodal educational slides,”Information Fusion, vol. 109, p. 102428, 2024

    2024

  51. [59]

    Multimodal graph learn- ing based on 3d haar semi-tight framelet for student engagement prediction,

    M. Li, X. Zhuang, L. Bai, and W. Ding, “Multimodal graph learn- ing based on 3d haar semi-tight framelet for student engagement prediction,”Information Fusion, vol. 105, p. 102224, 2024

    2024

  52. [60]

    Convolutional neural networks on graphs with fast localized spectral filtering,

    M. Defferrard, X. Bresson, and P . Vandergheynst, “Convolutional neural networks on graphs with fast localized spectral filtering,” Advances in neural information processing systems, vol. 29, 2016

    2016

  53. [61]

    Inductive representation learning on large graphs,

    W. Hamilton, Z. Ying, and J. Leskovec, “Inductive representation learning on large graphs,”Advances in neural information process- ing systems, vol. 30, 2017

    2017

  54. [62]

    Geometric deep learning on graphs and manifolds using mixture model cnns,

    F. Monti, D. Boscaini, J. Masci, E. Rodola, J. Svoboda, and M. M. Bronstein, “Geometric deep learning on graphs and manifolds using mixture model cnns,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 5115–5124

    2017

  55. [63]

    Neural message passing for quantum chemistry,

    J. Gilmer, S. S. Schoenholz, P . F. Riley, O. Vinyals, and G. E. Dahl, “Neural message passing for quantum chemistry,” inInterna- tional conference on machine learning. PMLR, 2017, pp. 1263–1272

    2017

  56. [64]

    Graph contrastive learning with augmentations,

    Y. You, T. Chen, Y. Sui, T. Chen, Z. Wang, and Y. Shen, “Graph contrastive learning with augmentations,”Advances in Neural Information Processing Systems, vol. 33, pp. 5812–5823, 2020

    2020

  57. [65]

    Graph contrastive learn- ing automated,

    Y. You, T. Chen, Y. Shen, and Z. Wang, “Graph contrastive learn- ing automated,” inInternational Conference on Machine Learning. PMLR, 2021, pp. 12 121–12 132

    2021

  58. [66]

    Graph contrastive learning with adaptive augmentation,

    Y. Zhu, Y. Xu, F. Yu, Q. Liu, S. Wu, and L. Wang, “Graph contrastive learning with adaptive augmentation,” inProceedings of the Web Conference 2021, 2021, pp. 2069–2080

    2021

  59. [67]

    What makes for good views for contrastive learning?

    Y. Tian, C. Sun, B. Poole, D. Krishnan, C. Schmid, and P . Isola, “What makes for good views for contrastive learning?”Advances in Neural Information Processing Systems, vol. 33, pp. 6827–6839, 2020

    2020

  60. [68]

    What should not be contrastive in contrastive learning,

    T. Xiao, X. Wang, A. A. Efros, and T. Darrell, “What should not be contrastive in contrastive learning,”arXiv preprint arXiv:2008.05659, 2020

    2008 arXiv

  61. [69]

    Curriculum learning of multiple tasks,

    A. Pentina, V . Sharmanska, and C. H. Lampert, “Curriculum learning of multiple tasks,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015, pp. 5492–5500

    2015

  62. [70]

    Curriculum learning of visual attribute clusters for multi-task classification,

    N. Sarafianos, T. Giannakopoulos, C. Nikou, and I. A. Kakadiaris, “Curriculum learning of visual attribute clusters for multi-task classification,”Pattern Recognition, vol. 80, pp. 94–108, 2018

    2018

  63. [71]

    Finding community structure in very large networks,

    A. Clauset, M. E. J. Newman, and C. Moore, “Finding community structure in very large networks,”Phys. Rev. E, vol. 70, p. 066111, Dec 2004. [Online]. Available: https: //link.aps.org/doi/10.1103/PhysRevE.70.066111

    2004 doi

  64. [72]

    A community detection algo- rithm based on graph compression for large-scale social net- works,

    X. Zhao, J. Liang, and J. Wang, “A community detection algo- rithm based on graph compression for large-scale social net- works,”Information Sciences, vol. 551, pp. 358–372, 2021

    2021

  65. [73]

    Costa: covariance-preserving feature augmentation for graph con- trastive learning,

    Y. Zhang, H. Zhu, Z. Song, P . Koniusz, and I. King, “Costa: covariance-preserving feature augmentation for graph con- trastive learning,” inProceedings of the 28th ACM SIGKDD Confer- ence on Knowledge Discovery and Data Mining, 2022, pp. 2524–2534

    2022

  66. [74]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems, vol. 30, 2017

    2017

  67. [75]

    Graph attention networks,

    P . Velickovic, G. Cucurull, A. Casanova, A. Romero, P . Lio, Y. Bengioet al., “Graph attention networks,”stat, vol. 1050, no. 20, pp. 10–48 550, 2017

    2017

  68. [76]

    Representation learning on graphs with jumping knowledge networks,

    K. Xu, C. Li, Y. Tian, T. Sonobe, K.-i. Kawarabayashi, and S. Jegelka, “Representation learning on graphs with jumping knowledge networks,” inInternational conference on machine learn- ing. PMLR, 2018, pp. 5453–5462

    2018

  69. [77]

    Asymptotically efficient estimators for stochastic blockmodels: The naive mle, the rank- constrained mle, and the spectral estimator,

    M. Tang, J. Cape, and C. E. Priebe, “Asymptotically efficient estimators for stochastic blockmodels: The naive mle, the rank- constrained mle, and the spectral estimator,”Bernoulli, vol. 28, no. 2, pp. 1049–1073, 2022

    2022

  70. [78]

    An iterative clustering algorithm for the contextual stochastic block model with optimal- ity guarantees,

    G. Braun, H. Tyagi, and C. Biernacki, “An iterative clustering algorithm for the contextual stochastic block model with optimal- ity guarantees,” inInternational Conference on Machine Learning. PMLR, 2022, pp. 2257–2291

    2022

  71. [79]

    Stochastic blockmodels and community structure in networks,

    B. Karrer and M. E. Newman, “Stochastic blockmodels and community structure in networks,”Physical review E, vol. 83, no. 1, p. 016107, 2011

    2011

  72. [80]

    Flp-id: Fuzzy-based link prediction in multiplex social networks using information diffusion perspective,

    S. S. Singh, D. Srivastva, A. Kumar, and V . Srivastava, “Flp-id: Fuzzy-based link prediction in multiplex social networks using information diffusion perspective,”Knowledge-Based Systems, vol. 248, p. 108821, 2022

    2022

  73. [81]

    Collaborative representation learning for nodes and relations via heterogeneous graph neural network,

    W. Li, L. Ni, J. Wang, and C. Wang, “Collaborative representation learning for nodes and relations via heterogeneous graph neural network,”Knowledge-Based Systems, vol. 255, p. 109673, 2022

    2022

  74. [82]

    Link prediction in complex networks based on significance of higher- order path index (shopi),

    A. Kumar, S. Mishra, S. S. Singh, K. Singh, and B. Biswas, “Link prediction in complex networks based on significance of higher- order path index (shopi),”Physica A: Statistical Mechanics and its Applications, vol. 545, p. 123790, 2020

    2020

  75. [83]

    Deep gaussian embedding of attributed graphs: Unsupervised inductive learning via ranking,

    A. Bojchevski and S. G ¨unnemann, “Deep gaussian embedding of attributed graphs: Unsupervised inductive learning via ranking,” arXiv preprint arXiv:1707.03815, 2017

    2017 arXiv

  76. [84]

    Pitfalls of graph neural network evaluation,

    O. Shchur, M. Mumme, A. Bojchevski, and S. G ¨unnemann, “Pitfalls of graph neural network evaluation,”arXiv preprint arXiv:1811.05868, 2018

    2018 arXiv

  77. [85]

    Collective classification in network data,

    P . Sen, G. Namata, M. Bilgic, L. Getoor, B. Galligher, and T. Eliassi- Rad, “Collective classification in network data,”AI magazine, vol. 29, no. 3, pp. 93–93, 2008

    2008

  78. [86]

    Inductive link prediction with interactive structure learning on attributed graph,

    S. Yang, B. Hu, Z. Zhang, W. Sun, Y. Wang, J. Zhou, H. Shan, Y. Cao, B. Ye, Y. Fanget al., “Inductive link prediction with interactive structure learning on attributed graph,” inMachine Learning and Knowledge Discovery in Databases. Research Track: European Conference, ECML PK...

    2021

  79. [87]

    A breadth-first search algorithm accelerator based on csci graph data format,

    H. Ren, J. Deng, B. Zhang, Z. Ye, and X. Fu, “A breadth-first search algorithm accelerator based on csci graph data format,” in2022 4th International Conference on Natural Language Processing (ICNLP). IEEE, 2022, pp. 636–640

    2022

  80. [88]

    Clustering and preferential attachment in grow- ing networks,

    M. E. Newman, “Clustering and preferential attachment in grow- ing networks,”Physical review E, vol. 64, no. 2, p. 025102, 2001

    2001

  81. [89]

    G. G. Chowdhury,Introduction to modern information retrieval. Facet publishing, 2010

    2010

  82. [90]

    Evaluation of features for social contact prediction,

    A. Samad, M. A. Islam, M. A. Iqbal, M. Aleem, and J. U. Arshed, “Evaluation of features for social contact prediction,” in2017 13th International Conference on Emerging Technologies (ICET). IEEE, 2017, pp. 1–6

    2017

  83. [91]

    A method of establishing groups of equal ampli- tude in plant sociology based on similarity of species content and its application to analyses of the vegetation on danish commons,

    T. Sorensen, “A method of establishing groups of equal ampli- tude in plant sociology based on similarity of species content and its application to analyses of the vegetation on danish commons,” Biologiske skrifter, vol. 5, pp. 1–34, 1948

    1948

  84. [92]

    Hierarchical organization of modularity in metabolic networks,

    E. Ravasz, A. L. Somera, D. A. Mongru, Z. N. Oltvai, and A.-L. Barab´asi, “Hierarchical organization of modularity in metabolic networks,”science, vol. 297, no. 5586, pp. 1551–1555, 2002

    2002

  85. [93]

    Predicting missing links via local information,

    T. Zhou, L. L ¨u, and Y.-C. Zhang, “Predicting missing links via local information,”The European Physical Journal B, vol. 71, pp. 623–630, 2009

    2009

  86. [94]

    Vertex similarity in networks,

    E. A. Leicht, P . Holme, and M. E. Newman, “Vertex similarity in networks,”Physical Review E, vol. 73, no. 2, p. 026120, 2006

    2006

  87. [95]

    Emergence of scaling in random networks,

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

    1999

  88. [96]

    Friends and neighbors on the web,

    L. A. Adamic and E. Adar, “Friends and neighbors on the web,” Social networks, vol. 25, no. 3, pp. 211–230, 2003

    2003

  89. [97]

    Resistance distance,

    D. J. Klein and M. Randic, “Resistance distance,”Journal of Mathematical Chemistry, vol. 12, pp. 81–95, 1993. [Online]. Available: https://api.semanticscholar.org/CorpusID:16382100

    1993

  90. [98]

    Random- walk computation of similarities between nodes of a graph with application to collaborative recommendation,

    F. Fouss, A. Pirotte, J.-M. Renders, and M. Saerens, “Random- walk computation of similarities between nodes of a graph with application to collaborative recommendation,”IEEE Transactions on knowledge and data engineering, vol. 19, no. 3, pp. 355–369, 2007

    2007

  91. [99]

    Clusterlp: A novel cluster-aware link prediction model in undirected and directed graphs,

    S. Zhang, W. Zhang, Z. Bu, and X. Zhang, “Clusterlp: A novel cluster-aware link prediction model in undirected and directed graphs,”International Journal of Approximate Reasoning, p. 109216, 2024

    2024

  92. [100]

    node2vec: Scalable feature learning for networks,

    A. Grover and J. Leskovec, “node2vec: Scalable feature learning for networks,” inProceedings of the 22nd ACM SIGKDD interna- tional conference on Knowledge discovery and data mining, 2016, pp. 855–864

    2016

  93. [101]

    Adversari- ally regularized graph autoencoder for graph embedding,

    S. Pan, R. Hu, G. Long, J. Jiang, L. Yao, and C. Zhang, “Adversari- ally regularized graph autoencoder for graph embedding,”arXiv preprint arXiv:1802.04407, 2018

    2018 arXiv

  94. [102]

    Graph infoclust: Maximizing coarse-grain mutual information in graphs,

    C. Mavromatis and G. Karypis, “Graph infoclust: Maximizing coarse-grain mutual information in graphs,” inPacific-Asia Con- ference on Knowledge Discovery and Data Mining. Springer, 2021, pp. 541–553

    2021

  95. [103]

    Modularity-aware graph autoencoders for joint JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 14 community detection and link prediction,

    G. Salha-Galvan, J. F. Lutzeyer, G. Dasoulas, R. Hennequin, and M. Vazirgiannis, “Modularity-aware graph autoencoders for joint JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 14 community detection and link prediction,”Neural Networks, vol. 153, pp. 474–495, 2022

    2015

  96. [104]

    What’s behind the mask: Understanding masked graph modeling for graph autoencoders,

    J. Li, R. Wu, W. Sun, L. Chen, S. Tian, L. Zhu, C. Meng, Z. Zheng, and W. Wang, “What’s behind the mask: Understanding masked graph modeling for graph autoencoders,” inProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2023, pp. 1268–1279

    2023

  97. [105]

    S2gae: Self-supervised graph autoencoders are generalizable learners with graph masking,

    Q. Tan, N. Liu, X. Huang, S.-H. Choi, L. Li, R. Chen, and X. Hu, “S2gae: Self-supervised graph autoencoders are generalizable learners with graph masking,” inProceedings of the sixteenth ACM international conference on web search and data mining, 2023, pp. 787–795

    2023

  98. [106]

    Core– periphery detection based on masked bayesian nonnegative matrix factorization,

    Z. Wang, R. Yuan, J. Fu, K.-C. Wong, and C. Peng, “Core– periphery detection based on masked bayesian nonnegative matrix factorization,”IEEE Transactions on Computational Social Systems, 2024

    2024

  99. [107]

    Finding core–periphery structures in large networks,

    X. Shen, Y. Han, W. Li, K.-C. Wong, and C. Peng, “Finding core–periphery structures in large networks,”Physica A: Statistical Mechanics and its Applications, vol. 581, p. 126224, 2021. Y uyang ZhangYuyang Zhang received the B.Eng. degree from Taizhou University, Taizhou, China...

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.