pith. sign in

arxiv: 2202.07082 · v4 · submitted 2022-02-14 · 💻 cs.LG

Graph Neural Networks for Graphs with Heterophily: A Survey

Xin Zheng , Yi Wang , Yixin Liu , Ming Li , Miao Zhang , Di Jin , Philip S. Yu , Shirui Pan This is my paper

Pith reviewed 2026-05-24 12:14 UTC · model grok-4.3

classification 💻 cs.LG
keywords graph neural networksheterophilysurveytaxonomygraph learningnode classificationheterophilic graphs
0
0 comments X

The pith

A systematic taxonomy organizes existing GNN models built for graphs where connected nodes have different labels.

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

The paper reviews graph neural networks developed for heterophily, the case where nodes with different labels tend to connect. It groups current models under one taxonomy, supplies summaries and analyses for each group, and traces how heterophily relates to other graph research areas. The goal is to make it easier to design stronger GNNs for applications that encounter this property. A reader would care because many real graphs in social, biological, and citation networks violate the homophily assumption that standard GNNs rely on.

Core claim

The paper claims that heterophilic GNNs can be governed by a systematic taxonomy, supplies general summaries and detailed analyses of the models under that taxonomy, maps heterophily onto other graph domains, and identifies open directions for future work.

What carries the argument

A systematic taxonomy of heterophilic GNN models that classifies them according to how they address label dissimilarity between linked nodes.

If this is right

  • Models can be compared and extended more systematically once placed inside the taxonomy categories.
  • Design choices in each category can be reused to address heterophily in node classification and related tasks.
  • Links between heterophily and other graph problems can guide transfer of techniques across domains.
  • Future models can be developed by filling gaps the taxonomy makes visible.

Where Pith is reading between the lines

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

  • The taxonomy may reveal which architectural families remain under-explored for heterophily.
  • Insights from the survey could be tested on dynamic or temporal graphs that also exhibit changing label patterns.
  • The same organizing lens might apply to other non-homophilic settings such as signed graphs or graphs with noisy labels.

Load-bearing premise

That a taxonomy built from today's published models will help researchers create more effective GNNs for practical heterophilic graphs.

What would settle it

A new heterophilic GNN that cannot be assigned to any category in the taxonomy, or a controlled study in which following the taxonomy produces no measurable gain in model performance on heterophilic benchmarks.

Figures

Figures reproduced from arXiv: 2202.07082 by Di Jin, Miao Zhang, Ming Li, Philip S. Yu, Shirui Pan, Xin Zheng, Yi Wang, Yixin Liu.

Figure 1
Figure 1. Figure 1: Examples of homophilic and heterophilic graphs (Left: (a) [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The algorithm taxonomy of GNNs with heterophily. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
read the original abstract

Recent years have witnessed fast developments of graph neural networks (GNNs) that have benefited myriad graph analytic tasks and applications. Most GNNs rely on the homophily assumption that nodes belonging to the same class are more likely to be connected. However, as a ubiquitous graph property in numerous real-world scenarios, heterophily, i.e., nodes with different labels tend to be linked, significantly limits the performance of tailor-made homophilic GNNs. Hence, GNNs for heterophilic graphs are gaining increasing research attention to enhance graph learning with heterophily. In this paper, we provide a comprehensive review of GNNs for heterophilic graphs. Specifically, we propose a systematic taxonomy that governs existing heterophilic GNN models, along with general summaries and detailed analyses. Furthermore, we discuss the relationship between heterophily and various graph research domains, aiming to facilitate the development of more effective GNNs across a spectrum of practical applications and learning tasks in the graph research community. In the end, we point out potential directions to advance and inspire future research and applications on heterophilic graph learning with GNNs.

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

0 major / 2 minor

Summary. The manuscript is a survey on graph neural networks (GNNs) for heterophilic graphs, where connected nodes tend to have dissimilar labels. It proposes a systematic taxonomy to organize existing heterophilic GNN models, provides summaries and detailed analyses of these models, discusses relationships between heterophily and other graph research domains (such as various learning tasks and applications), and identifies potential future research directions to advance heterophilic graph learning.

Significance. If the taxonomy is comprehensive, internally consistent, and accurately derived from the cited literature, the survey would provide a useful organizing framework for a growing subfield of GNN research. This could help researchers identify gaps and connections across domains, supporting the development of more effective models for real-world graphs that violate the homophily assumption. The explicit discussion of cross-domain relationships is a constructive element.

minor comments (2)
  1. [Abstract and taxonomy section] The abstract states the taxonomy 'governs existing heterophilic GNN models' but the manuscript should add an explicit statement of inclusion criteria and coverage (e.g., number of models reviewed per category) early in the taxonomy section to allow readers to assess completeness.
  2. [Analyses subsection] Some model descriptions in the analyses could benefit from a standardized comparison table (e.g., key mechanisms, complexity, and reported performance on heterophily benchmarks) to make the detailed analyses easier to navigate.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the constructive summary and positive evaluation of our survey. The recommendation for minor revision is noted. No specific major comments were provided in the report, so we have no points to address point-by-point at this stage. We will make any minor adjustments as needed during revision.

Circularity Check

0 steps flagged

No significant circularity in survey taxonomy

full rationale

This is a literature survey paper with no derivations, equations, fitted parameters, or predictive claims. The central contribution is a proposed taxonomy of existing heterophilic GNN models drawn from cited works, plus discussion of relationships to other domains. No load-bearing step reduces to self-definition, fitted input, or self-citation chain; the taxonomy is presented as an organizational summary of prior literature rather than a deductive result. The work is self-contained as a review and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The survey rests on the domain assumption that heterophily is a common and performance-limiting property in real graphs; no free parameters or invented entities are introduced.

axioms (1)
  • domain assumption Heterophily, i.e., nodes with different labels tend to be linked, is a ubiquitous graph property in numerous real-world scenarios.
    Invoked in the abstract as the core motivation limiting standard GNN performance.

pith-pipeline@v0.9.0 · 5745 in / 1066 out tokens · 21295 ms · 2026-05-24T12:14:45.859664+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.

Forward citations

Cited by 8 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Differentially Private Motif-Preserving Multi-modal Hashing

    cs.IR 2026-05 unverdicted novelty 7.0

    DMP-MH clips degrees to control triangle sensitivity, synthesizes an edge-DP graph with Noisy Mirror Descent, and distills it into dual-stream hash networks, beating private baselines by up to 11.4 mAP on MIRFlickr-25...

  2. Neighbourhood Transformer: Switchable Attention for Monophily-Aware Graph Learning

    cs.LG 2026-04 unverdicted novelty 7.0

    Neighbourhood Transformers apply local self-attention for monophily-aware graph learning, guarantee expressiveness at least as strong as message-passing GNNs, and outperform prior methods on node classification across...

  3. Modeling Heterophily in Multiplex Graphs: An Adaptive Approach for Node Classification

    cs.LG 2026-05 unverdicted novelty 6.0

    A new method for node classification in multiplex graphs adapts to mixed homophily and heterophily per dimension via compatibility matrices and Chebyshev-approximated product filters optimized with proximal gradients.

  4. Debiased Multimodal Personality Understanding through Dual Causal Intervention

    cs.AI 2026-05 unverdicted novelty 6.0

    DCAN uses dual causal adjustments with prototype and mediator dictionaries to debias multimodal personality understanding, reporting higher accuracy and fairness on CFI-V2 and a new DMSP dataset.

  5. HeterSEED: Semantics-Structure Decoupling for Heterogeneous Graph Learning under Heterophily

    cs.LG 2026-05 unverdicted novelty 6.0

    HeterSEED decouples semantics from structure in heterogeneous graphs under heterophily using separate channels and adaptive fusion, proving higher expressiveness and lower bias than standard HGNNs while outperforming ...

  6. Attention-based graph neural networks: a survey

    cs.SI 2026-05 unverdicted novelty 5.0

    The survey groups attention-based GNNs into three stages—graph recurrent attention networks, graph attention networks, and graph transformers—while reviewing architectures and future directions.

  7. Graph Rewiring in GNNs to Mitigate Over-Squashing and Over-Smoothing: A Survey

    cs.LG 2026-05 unverdicted novelty 2.0

    A survey reviewing graph rewiring methods that modify topology to mitigate over-squashing and over-smoothing in GNNs.

  8. Graph Rewiring in GNNs to Mitigate Over-Squashing and Over-Smoothing: A Survey

    cs.LG 2024-11 unverdicted novelty 2.0

    A survey compiling graph rewiring techniques for mitigating over-squashing and over-smoothing in GNNs.

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages · cited by 8 Pith papers

  1. [1]

    Mixhop: Higher-order graph convolutional archi- tectures via sparsified neighborhood mixing

    [Abu-El-Haija et al., 2019] Sami Abu-El-Haija, Bryan Per- ozzi, Amol Kapoor, Nazanin Alipourfard, Kristina Ler- man, Hrayr Harutyunyan, Greg Ver Steeg, and Aram Gal- styan. Mixhop: Higher-order graph convolutional archi- tectures via sparsified neighborhood mixing. In Inter- national Conference on Machine Learning , pages 21–29. PMLR,

    work page 2019

  2. [2]

    Beyond low-frequency information in graph convolutional networks

    [Bo et al., 2021] Deyu Bo, Xiao Wang, Chuan Shi, and Huawei Shen. Beyond low-frequency information in graph convolutional networks. In Proceedings of the AAAI Con- ference on Artificial Intelligence, pages 3950–3957,

    work page 2021

  3. [3]

    Simple and deep graph convolutional networks

    [Chen et al., 2020] Ming Chen, Zhewei Wei, Zengfeng Huang, Bolin Ding, and Yaliang Li. Simple and deep graph convolutional networks. In International Confer- ence on Machine Learning , pages 1725–1735. PMLR,

    work page 2020

  4. [4]

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

    [Chiang et al., 2019] Wei-Lin Chiang, Xuanqing Liu, Si Si, Yang Li, Samy Bengio, and Cho-Jui 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 Discov- ery & Data Mining, pages 257–266,

    work page 2019

  5. [5]

    Adaptive universal generalized pager- ank graph neural network

    [Chien et al., 2020] Eli Chien, Jianhao Peng, Pan Li, and Ol- gica Milenkovic. Adaptive universal generalized pager- ank graph neural network. In International Conference on Learning Representations,

    work page 2020

  6. [6]

    Ho- mophily and missing links in citation networks

    [Ciotti et al., 2016] Valerio Ciotti, Moreno Bonaventura, Vincenzo Nicosia, Pietro Panzarasa, and Vito Latora. Ho- mophily and missing links in citation networks. EPJ Data Science, 5:1–14,

    work page 2016

  7. [7]

    Using fuzzy logic to leverage html markup for web page representation

    [Garc´ıa-Plaza et al., 2016] Alberto P Garc ´ıa-Plaza, V ´ıctor Fresno, Raquel Mart ´ınez Unanue, and Arkaitz Zubiaga. Using fuzzy logic to leverage html markup for web page representation. IEEE Transactions on Fuzzy Systems , 25(4):919–933,

    work page 2016

  8. [8]

    Inductive representation learning on large graphs

    [Hamilton et al., 2017] William L Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large graphs. In Advances in Neural Information Processing Systems, pages 1025–1035,

    work page 2017

  9. [9]

    Block modeling-guided graph convolutional neural net- works

    [He et al., 2022] Dongxiao He, Chundong Liang, Huixin Liu, Mingxiang Wen, Pengfei Jiao, and Zhiyong Feng. Block modeling-guided graph convolutional neural net- works. In Proceedings of the AAAI Conference on Arti- ficial Intelligence,

    work page 2022

  10. [10]

    Semi-supervised classification with graph convolutional networks

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

    work page 2017

  11. [11]

    Deeper insights into graph convolutional networks for semi-supervised learning

    [Li et al., 2018] Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for semi-supervised learning. In Proceedings of the AAAI Conference on Artificial Intelligence,

    work page 2018

  12. [12]

    [Liu et al., 2021] M. Liu, Z. Wang, and S. Ji. Non-local graph neural networks. IEEE Transactions on Pattern Analysis & Machine Intelligence,

    work page 2021

  13. [13]

    Towards un- supervised deep graph structure learning

    [Liu et al., 2022] Yixin Liu, Yu Zheng, Daokun Zhang, Hongxu Chen, Hao Peng, and Shirui Pan. Towards un- supervised deep graph structure learning. In Proceedings of the 31st international conference on World Wide Web ,

    work page 2022

  14. [14]

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

    [Luan et al., 2021] Sitao Luan, Chenqing Hua, Qincheng Lu, Jiaqi Zhu, Mingde Zhao, Shuyuan Zhang, Xiao-Wen Chang, and Doina Precup. Is heterophily a real nightmare for graph neural networks to do node classification?arXiv preprint arXiv:2109.05641,

    work page arXiv 2021

  15. [15]

    Parameterized explainer for graph neural network

    [Luo et al., 2020] Dongsheng Luo, Wei Cheng, Dongkuan Xu, Wenchao Yu, Bo Zong, Haifeng Chen, and Xiang Zhang. Parameterized explainer for graph neural network. Advances in Neural Information Processing Systems , 33,

    work page 2020

  16. [16]

    Learning disentangled representations for recommendation

    [Ma et al., 2019] Jianxin Ma, Chang Zhou, Peng Cui, Hongxia Yang, and Wenwu Zhu. Learning disentangled representations for recommendation. In Advances in Neu- ral Information Processing Systems , pages 5711–5722,

    work page 2019

  17. [17]

    Is homophily a necessity for graph neural net- works? In International Conference on Learning Repre- sentations,

    [Ma et al., 2022] Yao Ma, Xiaorui Liu, Neil Shah, and Jil- iang Tang. Is homophily a necessity for graph neural net- works? In International Conference on Learning Repre- sentations,

    work page 2022

  18. [18]

    Netprobe: a fast and scalable system for fraud detection in online auction networks

    [Pandit et al., 2007] Shashank Pandit, Duen Horng Chau, Samuel Wang, and Christos Faloutsos. Netprobe: a fast and scalable system for fraud detection in online auction networks. In Proceedings of the 16th international confer- ence on World Wide Web, pages 201–210,

    work page 2007

  19. [19]

    Geom-gcn: Geomet- ric graph convolutional networks

    [Pei et al., 2020] Hongbin Pei, Bingzhe Wei, Kevin Chen- Chuan Chang, Yu Lei, and Bo Yang. Geom-gcn: Geomet- ric graph convolutional networks. InInternational Confer- ence on Learning Representations,

    work page 2020

  20. [20]

    Reinforced, incre- mental and cross-lingual event detection from social mes- sages

    [Peng et al., 2022] Hao Peng, Ruitong Zhang, Shaoning Li, Yuwei Cao, Shirui Pan, and Philip Yu. Reinforced, incre- mental and cross-lingual event detection from social mes- sages. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence,

    work page 2022

  21. [21]

    Multi-scale attributed node embed- ding

    [Rozemberczki et al., 2021] Benedek Rozemberczki, Carl Allen, and Rik Sarkar. Multi-scale attributed node embed- ding. Journal of Complex Networks, 9(2):cnab014,

    work page 2021

  22. [22]

    Adversarial attack and defense on graph data: A survey.arXiv preprint arXiv:1812.10528,

    [Sun et al., 2018] Lichao Sun, Yingtong Dou, Carl Yang, Ji Wang, Philip S Yu, Lifang He, and Bo Li. Adversarial attack and defense on graph data: A survey.arXiv preprint arXiv:1812.10528,

    work page arXiv 2018

  23. [23]

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

    [Suresh et al., 2021] Susheel Suresh, Vinith Budde, Jennifer Neville, Pan Li, and Jianzhu Ma. Breaking the limit of graph neural networks by improving the assortativity of graphs with local mixing patterns. In Proceedings of the ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pages 1541–1551,

    work page 2021

  24. [24]

    Social influence analysis in large-scale networks

    [Tang et al., 2009] Jie Tang, Jimeng Sun, Chi Wang, and Zi Yang. Social influence analysis in large-scale networks. In Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery & data mining, pages 807–816,

    work page 2009

  25. [25]

    Composition-based multi-relational graph convolutional networks

    [Vashishth et al., 2019] Shikhar Vashishth, Soumya Sanyal, Vikram Nitin, and Partha Talukdar. Composition-based multi-relational graph convolutional networks. In Interna- tional Conference on Learning Representations,

    work page 2019

  26. [26]

    Graph attention networks

    [Veliˇckovi´c et al., 2018] Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li `o, and Yoshua Bengio. Graph attention networks. In Interna- tional Conference on Learning Representations,

    work page 2018

  27. [27]

    Tree de- composed graph neural network

    [Wang and Derr, 2021] Yu Wang and Tyler Derr. Tree de- composed graph neural network. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, pages 2040–2049,

    work page 2021

  28. [28]

    Powerful graph convolutioal networks with adaptive propagation mechanism for ho- mophily and heterophily

    [Wang et al., 2022] Tao Wang, Rui Wang, Di Jin, Dongx- iao He, and Yuxiao Huang. Powerful graph convolutioal networks with adaptive propagation mechanism for ho- mophily and heterophily. InProceedings of the AAAI Con- ference on Artificial Intelligence,

    work page 2022

  29. [29]

    Simpli- fying graph convolutional networks

    [Wu et al., 2019] Felix Wu, Amauri Souza, Tianyi Zhang, Christopher Fifty, Tao Yu, and Kilian Weinberger. Simpli- fying graph convolutional networks. InInternational Con- ference on Machine Learning , pages 6861–6871. PMLR,

    work page 2019

  30. [30]

    Representation learning on graphs with jumping knowledge networks

    [Xu et al., 2018] Keyulu Xu, Chengtao Li, Yonglong Tian, Tomohiro Sonobe, Ken-ichi Kawarabayashi, and Stefanie Jegelka. Representation learning on graphs with jumping knowledge networks. In International Conference on Ma- chine Learning, pages 5453–5462. PMLR,

    work page 2018

  31. [31]

    How powerful are graph neural net- works? In International Conference on Learning Rep- resentations,

    [Xu et al., 2019] Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural net- works? In International Conference on Learning Rep- resentations,

    work page 2019

  32. [32]

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

    [Yan et al., 2021] Yujun Yan, Milad Hashemi, Kevin Swer- sky, Yaoqing Yang, and Danai Koutra. Two sides of the same coin: Heterophily and oversmoothing in graph convolutional neural networks. arXiv preprint arXiv:2102.06462,

    work page arXiv 2021

  33. [33]

    Di- verse message passing for attribute with heterophily

    [Yang et al., 2021] Liang Yang, Mengzhe Li, Liyang Liu, Chuan Wang, Xiaochun Cao, Yuanfang Guo, et al. Di- verse message passing for attribute with heterophily. In Advances in Neural Information Processing Systems, vol- ume 34,

    work page 2021

  34. [34]

    Graph pointer neural networks

    [Yang et al., 2022] Tianmeng Yang, Yujing Wang, Zhihan Yue, Yaming Yang, Yunhai Tong, and Jing Bai. Graph pointer neural networks. In Proceedings of the AAAI Con- ference on Artificial Intelligence,

    work page 2022

  35. [35]

    Graph convolutional neural networks for web- scale recommender systems

    [Ying et al., 2018] Rex Ying, Ruining He, Kaifeng Chen, Pong Eksombatchai, William L Hamilton, and Jure Leskovec. Graph convolutional neural networks for web- scale recommender systems. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, pages 974–983,

    work page 2018

  36. [36]

    Gnnexplainer: Gener- ating explanations for graph neural networks

    [Ying et al., 2019] Rex Ying, Dylan Bourgeois, Jiaxuan You, Marinka Zitnik, and Jure Leskovec. Gnnexplainer: Gener- ating explanations for graph neural networks. Advances in Neural Information Processing Systems, 32:9240,

    work page 2019

  37. [37]

    Graph- saint: Graph sampling based inductive learning method

    [Zeng et al., 2019] Hanqing Zeng, Hongkuan Zhou, Ajitesh Srivastava, Rajgopal Kannan, and Viktor Prasanna. Graph- saint: Graph sampling based inductive learning method. In International Conference on Learning Representations,

    work page 2019

  38. [38]

    Graph neural networks with heterophily

    [Zhu et al., 2021] Jiong Zhu, Ryan A Rossi, Anup B Rao, Tung Mai, Nedim Lipka, Nesreen K Ahmed, and Danai Koutra. Graph neural networks with heterophily. In Pro- ceedings of the AAAI Conference on Artificial Intelligence, 2021

    work page 2021