pith. sign in

arxiv: 2305.09958 · v5 · submitted 2023-05-17 · 💻 cs.LG · cs.SI

SIGMA: An Efficient Heterophilous Graph Neural Network with Fast Global Aggregation

Haoyu Liu , Ningyi Liao , Siqiang Luo This is my paper

Pith reviewed 2026-05-24 08:42 UTC · model grok-4.3

classification 💻 cs.LG cs.SI
keywords graph neural networksheterophilySimRankglobal aggregationefficiencystructural similaritylarge graphs
0
0 comments X

The pith

SIGMA integrates SimRank into graph neural network aggregation to capture distant global similarities in heterophilous graphs with linear complexity.

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

The paper introduces SIGMA as a way to handle heterophily in graphs where connected nodes are often dissimilar. Conventional GNNs struggle because they aggregate locally and uniformly, so existing solutions add long-range aggregations but at high computational cost through iterations. SIGMA instead folds the SimRank similarity measure directly into the aggregation step. This lets it capture global structural similarities in a single efficient pass whose cost grows only linearly with the number of nodes. Experiments confirm it matches or exceeds prior methods while running substantially faster on large heterophilous graphs.

Core claim

SIGMA integrates the structural similarity measurement SimRank into GNN aggregation. Its theoretical analysis shows that this inherently captures distant global similarity even under heterophily, something conventional approaches achieve only after multiple iterative aggregations. The method requires only one-time computation with complexity linear in the node set size O(n).

What carries the argument

The SimRank integration into the aggregation step, which computes structural similarities across the graph once and uses them to distinguish nodes under heterophily.

If this is right

  • SIGMA achieves state-of-the-art performance on heterophilous graph datasets.
  • It delivers superior aggregation and overall efficiency compared to baselines.
  • It obtains 5 times acceleration on the large-scale heterophily dataset pokec with over 30 million edges.
  • The linear O(n) complexity enables application to graphs with millions of nodes without iterative overhead.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Replacing iterative global updates with a one-shot similarity computation could generalize to other graph tasks that need long-range information.
  • Future work might test whether different structural similarity measures yield similar efficiency gains under heterophily.
  • Large-scale graph learning pipelines could adopt this pattern to reduce memory and time costs when scaling beyond current limits.

Load-bearing premise

That adding SimRank to the aggregation step preserves the global similarity capture and linear complexity without introducing hidden iterative costs or reducing expressivity on heterophilous data.

What would settle it

A direct comparison on a heterophilous graph with tens of millions of edges showing whether SIGMA's accuracy after one pass equals that of an iterative baseline while using far less time and memory.

Figures

Figures reproduced from arXiv: 2305.09958 by Haoyu Liu, Ningyi Liao, Siqiang Luo.

Figure 1
Figure 1. Figure 1: All sub-figures are from Texas heterophily graph. (a) A toy example of global structural similarity. Two staffs inherit high similarity because they share similar neighbors intuitively. (b) Neighborhood-based local aggregation and (c) SIGMA aggregation. Node color represents aggregation score with respect to the center node (▲). Conventional aggregation focuses on neighboring nodes regardless of node label… view at source ↗
Figure 2
Figure 2. Figure 2: Patternes of SimRank scores over intra-class and [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Architecture of SIGMA. embeddings by incorporating the aggregation matrix such as [15, 16]. Instead, we only rely on a constant SimRank matrix S which can be efficiently calculated in precomputation. To generate node representations from graph topology and node attributes, we deploy a simple and effective heterophilous GNN architecture, derived from LINKX [17] that respectively embeds the adjacency matrix … view at source ↗
Figure 6
Figure 6. Figure 6: Effect of error parameter ϵ and top-k on graph pokec. against leading baselines, including MixHop, GCNII, LINKX, and GloGNN, in Fig.4. The results show that SIGMA achieves favorable convergence, attaining high accuracy in a short train￾ing time. Generally, SIGMA, MixHop, and LINKX converge quickly to their highest scores due to their simple designs, with SIGMA often achieving better ultimate accuracy and m… view at source ↗
Figure 5
Figure 5. Figure 5: Scalability Evaluation of SIGMA and GloGNN. X-axis 8 [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Trade-offs of runtime over the top-k scheme. The X￾axis represents the actual runtime, and the Y-axis represents accuracy. The value of k is indicated on each data point. determines the number of nodes selected for aggregation. In general, smaller values of ϵ and larger values of k increase computational complexity. On our empirical evaluation in Fig.6, we vary k ∈ {4, 8, . . . , 1024} and set ϵ ∈ {0.01, 0… view at source ↗
Figure 8
Figure 8. Figure 8: Homophily in node embeddings Z. X-axis corresponds to node index reordered by category labels, color along Y-axis represents values in the node embedding vector. TABLE XI: Exploration on iterative SIGMA. Model GENIUS ARXIV PENN94 TWITCH SNAP POKEC GCN-1 62.84 42.97 81.86 61.33 40.78 71.15 GCN-2 61.84 42.93 81.90 61.28 41.34 72.78 GCN-3 66.80 43.43 77.58 62.81 44.17 67.63 SIGMA-1 91.41 55.41 85.27 67.29 64.… view at source ↗
read the original abstract

Graph neural networks (GNNs) realize great success in graph learning but suffer from performance loss when meeting heterophily, i.e. neighboring nodes are dissimilar, due to their local and uniform aggregation. Existing attempts of heterophilous GNNs incorporate long-range or global aggregations to distinguish nodes in the graph. However, these aggregations usually require iteratively maintaining and updating full-graph information, which limits their efficiency when applying to large-scale graphs. In this paper, we propose SIGMA, an efficient global heterophilous GNN aggregation integrating the structural similarity measurement SimRank. Our theoretical analysis illustrates that SIGMA inherently captures distant global similarity even under heterophily, that conventional approaches can only achieve after iterative aggregations. Furthermore, it enjoys efficient one-time computation with a complexity only linear to the node set size $\mathcal{O}(n)$. Comprehensive evaluation demonstrates that SIGMA achieves state-of-the-art performance with superior aggregation and overall efficiency. Notably, it obtains $5\times$ acceleration on the large-scale heterophily dataset pokec with over 30 million edges compared to the best baseline aggregation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes SIGMA, a heterophilous GNN that integrates SimRank into the aggregation step to enable global structural similarity capture. It claims that this yields inherent distant-node similarity under heterophily (unlike local iterative baselines), with a one-time O(n) computation, SOTA accuracy, and 5× speedup on the 30M-edge pokec dataset.

Significance. If the O(n) reduction of SimRank is rigorously shown to preserve the similarity guarantee on general heterophilous graphs, the work would meaningfully advance scalable global aggregation for heterophily, directly addressing the efficiency barrier noted for prior long-range methods.

major comments (2)
  1. [theoretical analysis] The central O(n) claim for global aggregation (abstract and theoretical analysis) rests on an unshown reduction of SimRank; standard SimRank requires iterative fixed-point computation or quadratic matrix operations, so the manuscript must exhibit the exact closed-form or sampling equations that achieve strict linearity while retaining the distant-similarity guarantee under heterophily.
  2. [efficiency experiments] § on efficiency and experiments: the reported 5× acceleration versus the best baseline on pokec must be accompanied by explicit per-component timing (SimRank pre-computation vs. GNN forward pass) and confirmation that no hidden per-iteration or per-pair costs remain in the claimed one-time O(n) procedure.
minor comments (2)
  1. Notation for the SimRank-augmented aggregation operator should be defined explicitly (e.g., how the similarity matrix is folded into the message-passing rule) to allow reproduction.
  2. The abstract states “comprehensive evaluation” but the manuscript should add a table comparing wall-clock time and memory against all cited heterophilous baselines on the same hardware.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to improve clarity on the theoretical reduction and experimental details.

read point-by-point responses

  1. Referee: [theoretical analysis] The central O(n) claim for global aggregation (abstract and theoretical analysis) rests on an unshown reduction of SimRank; standard SimRank requires iterative fixed-point computation or quadratic matrix operations, so the manuscript must exhibit the exact closed-form or sampling equations that achieve strict linearity while retaining the distant-similarity guarantee under heterophily.

    Authors: Section 3.2 of the manuscript presents the SimRank integration via a one-pass structural similarity computation that avoids full iteration or quadratic operations by leveraging the graph's adjacency matrix in a linearized form. The distant-similarity guarantee under heterophily follows from the SimRank fixed-point properties applied globally in a single aggregation step. To address the request for explicitness, we will add the precise closed-form equations and the linearity derivation as a dedicated subsection in the revision. revision: yes

  2. Referee: [efficiency experiments] § on efficiency and experiments: the reported 5× acceleration versus the best baseline on pokec must be accompanied by explicit per-component timing (SimRank pre-computation vs. GNN forward pass) and confirmation that no hidden per-iteration or per-pair costs remain in the claimed one-time O(n) procedure.

    Authors: We will expand the efficiency experiments section with a breakdown table for the pokec dataset, explicitly reporting wall-clock times for the one-time SimRank pre-computation versus the subsequent GNN forward and backward passes. This will also include a statement confirming that all similarity values are materialized once with no per-iteration or per-pair recomputation during training or inference. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation integrates external SimRank measure

full rationale

The paper's central claims rest on integrating the established SimRank similarity measure into GNN aggregation, supported by a theoretical analysis of global similarity capture under heterophily and an asserted O(n) one-time computation. No self-definitional equations, fitted parameters renamed as predictions, or load-bearing self-citations that reduce the result to its own inputs appear in the abstract or described derivation. The approach treats SimRank as an independent input rather than deriving it from the model's own outputs or prior self-referential results, making the chain self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the central claim rests on the unelaborated integration of SimRank and the stated theoretical analysis.

pith-pipeline@v0.9.0 · 5727 in / 1094 out tokens · 18762 ms · 2026-05-24T08:42:25.577133+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages · 1 internal anchor

  1. [1]

    Semi-supervised classifica- tion with graph convolutional networks,

    M. Welling and T. N. Kipf, “Semi-supervised classifica- tion with graph convolutional networks,” in J. Interna- tional Conference on Learning Representations (ICLR 2017), 2016

    work page 2017

  2. [2]

    Inductive representation learning on large graphs,

    W. Hamilton, Z. Ying, and J. Leskovec, “Inductive representation learning on large graphs,” vol. 30, 2017

    work page 2017

  3. [3]

    Learning-based approaches for graph problems: a survey,

    K. S. Yow and S. Luo, “Learning-based approaches for graph problems: a survey,” arXiv preprint arXiv:2204.01057, 2022

    work page arXiv 2022

  4. [4]

    Spiking graph convolutional networks,

    Z. Zhu, J. Peng, J. Li, L. Chen, Q. Yu, and S. Luo, “Spiking graph convolutional networks,” arXiv preprint arXiv:2205.02767, 2022

    work page arXiv 2022

  5. [5]

    Fusing global domain information and local semantic information to classify financial documents,

    M. Fan, D. Cheng, F. Yang, S. Luo, Y . Luo, W. Qian, and A. Zhou, “Fusing global domain information and local semantic information to classify financial documents,” in Proceedings of the 29th ACM International Conference on Information & Knowledge Management , 2020, pp. 2413–2420

    work page 2020

  6. [6]

    Approximate graph propagation,

    H. Wang, M. He, Z. Wei, S. Wang, Y . Yuan, X. Du, and J.- R. Wen, “Approximate graph propagation,” inProceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining , 2021, pp. 1686–1696

    work page 2021

  7. [7]

    Scalable graph neural networks via bidirectional propagation,

    M. Chen, Z. Wei, B. Ding, Y . Li, Y . Yuan, X. Du, and J.-R. Wen, “Scalable graph neural networks via bidirectional propagation,” Advances in neural information processing systems, vol. 33, pp. 14 556–14 566, 2020

    work page 2020

  8. [8]

    Graph Attention Networks

    P. Veliˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Lio, and Y . Bengio, “Graph attention networks,” arXiv preprint arXiv:1710.10903, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  9. [9]

    Predict then propagate: Graph neural networks meet personal- ized pagerank,

    J. Klicpera, A. Bojchevski, and S. Günnemann, “Predict then propagate: Graph neural networks meet personal- ized pagerank,” International Conference on Learning Representations, 2018

    work page 2018

  10. [10]

    Beyond homophily in graph neural networks: Current limitations and effective designs,

    J. Zhu, Y . Yan, L. Zhao, M. Heimann, L. Akoglu, and D. Koutra, “Beyond homophily in graph neural networks: Current limitations and effective designs,” Advances in Neural Information Processing Systems, vol. 33, pp. 7793– 7804, 2020

    work page 2020

  11. [11]

    Mixhop: Higher-order graph convolutional architectures via sparsified neighborhood mixing,

    S. Abu-El-Haija, B. Perozzi, A. Kapoor, N. Alipour- fard, K. Lerman, H. Harutyunyan, G. Ver Steeg, and A. Galstyan, “Mixhop: Higher-order graph convolutional architectures via sparsified neighborhood mixing,” in international conference on machine learning . PMLR, 2019, pp. 21–29

    work page 2019

  12. [12]

    Geom-gcn: Geometric graph convolutional networks,

    H. Pei, B. Wei, K. C.-C. Chang, Y . Lei, and B. Yang, “Geom-gcn: Geometric graph convolutional networks,” International Conference on Learning Representations , 2020

    work page 2020

  13. [13]

    Graph pointer neural networks

    T. Yang, Y . Wang, Z. Yue, Y . Yang, Y . Tong, and J. Bai, “Graph pointer neural networks.” Proceedings of the ... AAAI Conference on Artificial Intelligence , 2022

    work page 2022

  14. [14]

    Non-local graph neural networks,

    M. Liu, Z. Wang, and S. Ji, “Non-local graph neural networks,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021

    work page 2021

  15. [15]

    Universal graph convolutional networks,

    D. Jin, Z. Yu, C. Huo, R. Wang, X. Wang, D. He, and J. Han, “Universal graph convolutional networks,” Advances in Neural Information Processing Systems , vol. 34, pp. 10 654–10 664, 2021

    work page 2021

  16. [16]

    Finding global homophily in graph neural networks when meeting heterophily,

    X. Li, R. Zhu, Y . Cheng, C. Shan, S. Luo, D. Li, and W. Qian, “Finding global homophily in graph neural networks when meeting heterophily,” International Con- ference on Machine Learning , 2022

    work page 2022

  17. [17]

    Large scale learning on non- homophilous graphs: New benchmarks and strong simple methods,

    D. Lim, F. Hohne, X. Li, S. L. Huang, V . Gupta, O. Bhalerao, and S. N. Lim, “Large scale learning on non- homophilous graphs: New benchmarks and strong simple methods,” Advances in Neural Information Processing Systems, vol. 34, pp. 20 887–20 902, 2021

    work page 2021

  18. [18]

    Breaking the limit of graph neural networks by improving the assortativity of graphs with local mixing patterns,

    S. Suresh, V . Budde, J. Neville, P. Li, and J. Ma, “Breaking the limit of graph neural networks by improving the assortativity of graphs with local mixing patterns,” Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining , 2021

    work page 2021

  19. [19]

    Simrank: a measure of structural- context similarity,

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

    work page 2002

  20. [20]

    Cast: a correlation-based adaptive spectral clustering algorithm on multi-scale data,

    X. Li, B. Kao, C. Shan, D. Yin, and M. Ester, “Cast: a correlation-based adaptive spectral clustering algorithm on multi-scale data,” in Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discov- ery & Data Mining , 2020, pp. 439–449

    work page 2020

  21. [21]

    Two sides of the same coin: Heterophily and oversmoothing in graph convolutional neural networks

    Y . Yan, M. Hashemi, K. Swersky, Y . Yang, and D. Koutra, “Two sides of the same coin: Heterophily and over- smoothing in graph convolutional neural networks,” arXiv preprint arXiv:2102.06462, 2021

    work page arXiv 2021

  22. [22]

    Graph Neural Networks Beyond Compromise Between Attribute and Topology,

    L. Yang, W. Zhou, W. Peng, B. Niu, J. Gu, C. Wang, X. Cao, and D. He, “Graph Neural Networks Beyond Compromise Between Attribute and Topology,” in Pro- ceedings of the ACM Web Conference 2022 . Virtual Event, Lyon France: ACM, Apr. 2022, pp. 1127–1135

    work page 2022

  23. [23]

    Exact single-source simrank computation on large graphs,

    H. Wang, Z. Wei, Y . Yuan, X. Du, and J.-R. Wen, “Exact single-source simrank computation on large graphs,” in Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data , 2020, pp. 653–663

    work page 2020

  24. [24]

    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,” in Proceedings of the 35th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research, J. Dy and A. Krause, Eds., vol. 80. PMLR, 10–15 Jul 2018, pp. 5453–5462

    work page 2018

  25. [25]

    Efficient top-k simrank-based simi- larity join,

    W. Tao and G. Li, “Efficient top-k simrank-based simi- larity join,” in Proceedings of the 2014 ACM SIGMOD International Conference on Management of Data , 2014, pp. 1603–1604

    work page 2014

  26. [26]

    Efficient simrank track- ing in dynamic graphs,

    Y . Wang, X. Lian, and L. Chen, “Efficient simrank track- ing in dynamic graphs,” in 2018 IEEE 34th international conference on data engineering (ICDE) . IEEE, 2018, pp. 545–556

    work page 2018

  27. [27]

    Gsim: A graph neural network based rele- vance measure for heterogeneous graphs,

    L. Luo, Y . Fang, M. Lu, X. Cao, X. Zhang, and 13 W. Zhang, “Gsim: A graph neural network based rele- vance measure for heterogeneous graphs,” arXiv preprint arXiv:2208.06144, 2022

    work page arXiv 2022

  28. [28]

    Massively parallel single- source simranks in o(logn) rounds,

    S. Luo and Z. Zhu, “Massively parallel single- source simranks in o(logn) rounds,” arXiv preprint arXiv:2304.04015, 2023

    work page arXiv 2023

  29. [29]

    Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks,

    W.-L. Chiang, X. Liu, S. Si, Y . Li, S. Bengio, and C.-J. Hsieh, “Cluster-gcn: An efficient algorithm for training deep and large graph convolutional networks,” in Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , 2019, pp. 257–266

    work page 2019

  30. [30]

    Adaptive universal generalized pagerank graph neural network,

    E. Chien, J. Peng, P. Li, and O. Milenkovic, “Adaptive universal generalized pagerank graph neural network,” in International Conference on Learning Representations ,

    work page

  31. [31]

    Available: https://openreview.net/forum? id=n6jl7fLxrP

    [Online]. Available: https://openreview.net/forum? id=n6jl7fLxrP

    work page

  32. [32]

    Dif- fusion Improves Graph Learning,

    J. Gasteiger, S. Weißenberger, and S. Günnemann, “Dif- fusion Improves Graph Learning,” in 32nd Advances in Neural Information Processing Systems , 2019

    work page 2019

  33. [33]

    Scaling graph neural networks with approximate pagerank,

    A. Bojchevski, J. Gasteiger, B. Perozzi, A. Kapoor, M. Blais, B. Rózemberczki, M. Lukasik, and S. Günne- mann, “Scaling graph neural networks with approximate pagerank,” in Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2020, pp. 2464–2473

    work page 2020

  34. [34]

    Do transformers really perform badly for graph representation?

    C. Ying, T. Cai, S. Luo, S. Zheng, G. Ke, D. He, Y . Shen, and T.-Y . Liu, “Do transformers really perform badly for graph representation?” Advances in neural information processing systems, vol. 34, pp. 28 877–28 888, 2021

    work page 2021

  35. [35]

    Simple and deep graph convolutional networks,

    M. Chen, Z. Wei, Z. Huang, B. Ding, and Y . Li, “Simple and deep graph convolutional networks,” in International Conference on Machine Learning . PMLR, 2020, pp. 1725–1735

    work page 2020

  36. [36]

    Simplifying graph convolutional networks,

    F. Wu, A. Souza, T. Zhang, C. Fifty, T. Yu, and K. Wein- berger, “Simplifying graph convolutional networks,” in Proceedings of the 36th International Conference on Machine Learning, K. Chaudhuri and R. Salakhutdinov, Eds., vol. 97, 6 2019, pp. 6861–6871

    work page 2019

  37. [37]

    Gbk-gnn: Gated bi-kernel graph neural networks for mod- eling both homophily and heterophily,

    L. Du, X. Shi, Q. Fu, X. Ma, H. Liu, S. Han, and D. Zhang, “Gbk-gnn: Gated bi-kernel graph neural networks for mod- eling both homophily and heterophily,” in Proceedings of the ACM Web Conference 2022 , 2022, pp. 1550–1558

    work page 2022

  38. [38]

    Powerful graph convolutional networks with adaptive propagation mechanism for homophily and heterophily,

    T. Wang, D. Jin, R. Wang, D. He, and Y . Huang, “Powerful graph convolutional networks with adaptive propagation mechanism for homophily and heterophily,” in Proceedings of the AAAI conference on artificial intelligence, vol. 36, no. 4, 2022, pp. 4210–4218

    work page 2022

  39. [39]

    Similarity-navigated graph neural networks for node classification,

    M. Zou, Z. Gan, R. Cao, C. Guan, and S. Leng, “Similarity-navigated graph neural networks for node classification,” Information Sciences , vol. 633, pp. 41– 69, 2023

    work page 2023

  40. [40]

    Predicting global label relationship matrix for graph neural networks under heterophily,

    L. Liang, X. Hu, Z. Xu, Z. Song, and I. King, “Predicting global label relationship matrix for graph neural networks under heterophily,” Advances in Neural Information Processing Systems, vol. 36, pp. 10 909–10 921, 2023

    work page 2023

  41. [41]

    Scara: Scalable graph neural networks with feature-oriented optimization,

    N. Liao, D. Mo, S. Luo, X. Li, and P. Yin, “Scara: Scalable graph neural networks with feature-oriented optimization,” Proceedings of the VLDB Endowment , vol. 15, no. 11, pp. 3240–3248, 2022

    work page 2022

  42. [42]

    LD2: Scalable heterophilous graph neural network with decoupled em- bedding,

    N. Liao, S. Luo, X. Li, and J. Shi, “LD2: Scalable heterophilous graph neural network with decoupled em- bedding,” in Advances in Neural Information Processing Systems, vol. 36, Dec 2023

    work page 2023

  43. [43]

    Scalable decoupling graph neural networks with feature-oriented optimization,

    N. Liao, D. Mo, S. Luo, X. Li, and P. Yin, “Scalable decoupling graph neural networks with feature-oriented optimization,” The VLDB Journal , vol. 33, no. 3, pp. 667–683, May 2024. [Online]. Available: https: //link.springer.com/article/10.1007/s00778-023-00829-6

    work page doi:10.1007/s00778-023-00829-6 2024

  44. [44]

    Bird: Efficient approximation of bidirectional hidden personalized pagerank,

    H. Liu and S. Luo, “Bird: Efficient approximation of bidirectional hidden personalized pagerank,” Proceedings of the VLDB Endowment , vol. 17, no. 9, pp. 2255–2268, 2024

    work page 2024

  45. [45]

    Agenda: Robust personalized pager- anks in evolving graphs,

    D. Mo and S. Luo, “Agenda: Robust personalized pager- anks in evolving graphs,” in Proceedings of the 30th ACM International Conference on Information & Knowledge Management, 2021, pp. 1315–1324

    work page 2021

  46. [46]

    Topology- monitorable contrastive learning on dynamic graphs,

    Z. Zhu, K. Wang, H. Liu, J. Li, and S. Luo, “Topology- monitorable contrastive learning on dynamic graphs,” in Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , 2024, pp. 4700– 4711

    work page 2024

  47. [47]

    Is heterophily a real nightmare for graph neural networks to do node classification?arXiv preprint arXiv:2109.05641,

    S. Luan, C. Hua, Q. Lu, J. Zhu, M. Zhao, S. Zhang, X.-W. Chang, and D. Precup, “Is heterophily a real nightmare for graph neural networks to do node classification?” arXiv preprint arXiv:2109.05641, 2021

    work page arXiv 2021

  48. [48]

    Multi- scale attributed node embedding,

    B. Rozemberczki, C. Allen, and R. Sarkar, “Multi- scale attributed node embedding,” Journal of Complex Networks, vol. 9, no. 2, p. cnab014, 2021

    work page 2021

  49. [49]

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

    A. Bojchevski and S. Günnemann, “Deep gaussian embedding of graphs: Unsupervised inductive learning via ranking,” in International Conference on Learning Representations, 2018. [Online]. Available: https:// openreview.net/forum?id=r1ZdKJ-0W

    work page 2018

  50. [50]

    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

    work page 2008

  51. [51]

    Query-driven active surveying for collective classification,

    G. Namata, B. London, L. Getoor, B. Huang, and U. Edu, “Query-driven active surveying for collective classification,” in 10th International Workshop on Mining and Learning with Graphs, vol. 8, 2012, p. 1

    work page 2012

  52. [52]

    On the use of arxiv as a dataset,

    C. B. Clement, M. Bierbaum, K. P. O’Keeffe, and A. A. Alemi, “On the use of arxiv as a dataset,” 2019

    work page 2019

  53. [53]

    Snap datasets: Stanford large network dataset collection,

    Jure Leskovec and Andrej Krevl, “Snap datasets: Stanford large network dataset collection,” http://snap.stanford.edu/ data, 2014. 14

    work page 2014