Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

One Model for One Graph: A New Perspective for Pretraining with Cross-domain Graphs

T0 review · 4 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Pretraining one model per graph and fusing the top few experts at inference transfers across graph domains better than training a single shared model.

desk verdict The per-graph expert bank idea is genuinely new and the transfer tables are consistently favorable, but Appendix D's theoretical guarantees are invalid, so the paper is an empirical claim in need of careful verification, not a proven method. read the letter →

arxiv 2412.00315 v2 pith:SPRDKVOA submitted 2024-11-30 cs.LG cs.AIstat.ML

Jingzhe Liu , Haitao Mao , Zhikai Chen , Bingheng Li , Wenqi Fan , Mingxuan Ju , Tong Zhao , Neil Shah
show 1 more author
Jiliang Tang
This is my paper · ORCID
classification cs.LGcs.AIstat.ML
keywords GraphNeuralNetworksCross-datasetpretrainingModelbankmergingNegativetransferZero-shotnodeclassificationLinkpredictionText-attributedgraphs
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 pretraining a single GNN on a bundle of different graphs is the wrong default: graphs from different domains conflict, so the shared model loses on each one. Instead, the authors propose training one source model per pretraining graph, storing them in a bank, and at inference time letting per-graph 'scoring' modules pick the few most relevant experts and merge their parameters. They call this pipeline 'one model for one graph' and implement it as OMOG. Tested across ten text-attributed graphs in zero-shot and few-shot settings, the fused model ranks first on almost every node-classification and link-prediction benchmark relative to shared-model baselines. If this holds up, a growing collection of pretrained graph experts can be reused for new domains by selecting and merging rather than retraining.

What carries the argument

The load-bearing pieces are the per-graph source model plus its paired scoring module, and the top-K weighted parameter fusion. Each source model is a small transformer trained with a graph contrastive objective on multi-hop SGC embeddings; each scoring module is an MLP trained so that, after filtering, features of its own domain land near the domain centroid while the mask itself does not. The relevance score is the cosine similarity of the encoded test embedding to a source graph's centroid, which the authors show approximates an inverse KL divergence under Gaussian assumptions, making the softmax-weighted fusion an approximation to Bayesian model averaging.

What would settle it

Run OMOG's leave-one-out protocol and, for each test graph, compare the fused top-K model's accuracy against the best single source model in the bank. Because Theorem 2 promises the fused model matches or exceeds that best expert, any test graph where the fused model is significantly worse would falsify the guarantee. A more direct test of the scoring assumption: take a test graph, permute its node labels randomly, and check whether the scoring module's top-K selection still picks the most structurally similar source graphs.

Watch

Extended reading notes

Core claim

The central claim is that cross-domain graph transfer succeeds not by distilling all graphs into one model but by keeping one model per graph and fusing the most relevant ones for each test graph. OMOG encodes node text with a language model to get a unified feature space, aggregates local structure with simplified graph convolutions (SGC), pretrains a contrastive transformer per graph to get a source model, and trains an associated MLP scoring module that filters which features look native to that graph. At inference, every scoring module scores the test graph, the top-K source models are selected and their parameters averaged with softmax weights. The paper proves this weighted merging is equivalent to Bayesian model averaging and that, under its assumptions, the fused model matches or beats the best individual expert in expectation. In leave-one-out experiments on ten text-attributed graphs, the fused model ranks first on 8 of 9 zero-shot node-classification datasets, all 9 link-prediction datasets, and 8 of 10 few-shot classification datasets.

Load-bearing premise

The entire theoretical guarantee rests on the relevance score being an approximate inverse KL divergence between the test graph's and a source graph's output distributions, and specifically on those distributions being Gaussian with aligned means; empirically it also assumes the shared text-embedding space is aligned well enough that per-graph experts are comparable.

Editorial extensions

If this is right

  • New pretraining graphs can be added to the bank by training one extra source and scoring model, without touching the existing experts.
  • Low-relevance source models are suppressed at inference, so adding more diverse pretraining graphs should not drag down performance the way joint training does.
  • Because fusion is a weighted average of parameters, the same bank can be reused for future test graphs without any fine-tuning.
  • The approach is computationally cheaper than large-LLM backbones, since each expert is a small transformer and experts can be trained in parallel.

Reading between the lines

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

  • The Gaussian equal-mean assumption in the appendix is strong; if a test graph's predictive distribution is far from Gaussian or its mean is misaligned, the BMA equivalence may not hold, and the scoring scores could mis-rank experts.
  • The bank inherits the quality of the text-embedding encoder; using a stronger aligned encoder should improve all experts, giving a scaling path that the paper does not explore.
  • Parameter averaging implicitly assumes the experts occupy a compatible linear region; if future experts are more heterogeneous, permutation-aligned merging (model soups) may be needed.
  • A genuinely novel domain with low relevance to every source model would reduce top-K selection to near-random choice; the paper does not test this out-of-bank-distribution case.
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

4 major / 4 minor

Summary. This paper proposes OMOG, a cross-domain graph pretraining framework built on a 'one model for one graph' pipeline. In the pretraining stage, OMOG trains one source model and one associated scoring model on each pretraining graph, using SentenceBERT features and SGC aggregation. At inference, the scoring models assign relevance scores to a test graph, the top-K source models are selected, softmax weights are computed, and the corresponding model parameters are merged by weighted averaging. The paper reports zero-shot node classification and link prediction results (Table 1), few-shot node classification results (Table 2), ablations of the scoring design (Figures 4-6), and computational cost comparisons (Table 3). The manuscript states two theorems claiming that the merging strategy is equivalent to Bayesian Model Averaging and that top-K selection and merging guarantee that the fused model matches or outperforms the best individual source model in expectation, with proofs in Appendix D.

Significance. If the empirical and theoretical claims were correct, OMOG would be a simple and potentially useful alternative to 'one model for all graphs' pretraining: per-graph experts avoid interference during pretraining, the model bank is incrementally extensible, and the reported training cost is low (Table 3). The top-K versus random-K versus least-K ablation in Figure 5 is genuine evidence that the learned relevance scores carry information, and the scaling behavior in Figure 6 is consistent with the intended negative-transfer avoidance. However, the formal guarantee is load-bearing for the paper's explanation of why negative transfer is mitigated, and that guarantee is not established: the proofs in Appendix D contain algebraic errors, and the key relation between the relevance score and inverse predictive variance is assumed rather than derived. The experimental comparison also has protocol asymmetries and no variance reporting. For these reasons, the contribution as currently stated cannot be accepted.

major comments (4)
  1. [Appendix D, Lemma D.5] The KL-divergence approximation is algebraically wrong. For equal-mean Gaussians with r = sigma_i^2 / sigma_test^2, the KL divergence is 0.5(r - 1 - log r). Since log r ≈ r - 1 to first order, the first-order terms cancel, and the leading-order approximation is quadratic in (r - 1), not 0.5(sigma_i^2 - sigma_test^2). Consequently, the claimed relation v_i ≈ -1/(2 sigma_i^2) does not follow even under the paper's own Assumption D.4.
  2. [Appendix D, Definition D.2 and Proposition D.6] The BMA equivalence is both assumed and based on a false approximation. Definition D.2 simply assumes that the relevance score v_i is approximately the negative KL divergence between output distributions; this is the very property that would need to be derived from the scoring model. Proposition D.6 then uses exp(-x) ≈ 1/x for small x, but the correct first-order approximation is exp(-x) ≈ 1 - x. Even substituting v_i = -1/(2 sigma_i^2) gives softmax weights proportional to exp(-1/(2 sigma_i^2)), which are not proportional to 1/sigma_i^2. Therefore Theorem 1 is unsupported.
  3. [Appendix D, Theorem D.7] The proof of Theorem 2 asserts Var(f_fused) <= min_i Var(f_i) and that the ensemble bias does not increase, but neither inequality is valid in general. For a weighted average with nonnegative weights summing to one, the variance can be as large as the maximum variance of the individual models (and larger under positive correlation), not the minimum, and the bias is a weighted average of the individual biases, which can exceed the minimum bias. The step E[f_fused] ≈ E[f_best] is simply assumed. Top-K selection does not repair these inequalities. Hence the advertised guarantee that the fused model outperforms or matches the best individual source model in expectation is not proved.
  4. [Section 4.2, Tables 1 and 2] The empirical comparison is not sufficiently controlled. LLaGA is pretrained on only Arxiv and Products rather than on the nine-graph leave-one-out pool used to train OMOG, while Prodigy uses its original MAG240M pretraining corpus; these are different pretraining conditions. In addition, no standard deviations, number of seeds, or significance tests are reported for any table. As a result, the reported margins cannot be attributed unambiguously to the OMOG design rather than to pretraining-data choices or run-to-run variation.
minor comments (4)
  1. [Title page] The name 'Mingxuan Ju' appears twice in the author list and should be deduplicated.
  2. [Throughout] There are numerous typos and misspellings, including 'vallina', 'texture-attributed', 'repectively', 'empolys', and 'pertaining'; the manuscript needs a careful proofreading pass.
  3. [Equation (4)] The denominator of the contrastive loss appears malformed as written; please check the summation indices and the factor of 2 so that the expression matches the intended InfoNCE form.
  4. [Figures 4-6] The ablation figures would be much more informative with error bars and a statement of the number of random seeds, especially because the 'No Source', 'No Score', and 'No SGC' variants are not described in enough detail to know what replacement strategy was used.

Circularity Check

3 steps flagged · score 8.0 of 10

Theorem 1 and Theorem 2 are forced by Appendix D assumptions: the inverse-KL relevance score is assumed, the softmax-to-BMA step uses a false approximation, and the fusion guarantee is assumed as 'E[f_fused]≈E[f_best]'; the empirical benchmark results remain independent and non-circular.

  1. self definitional [Appendix D, Definition D.2 and Lemma D.5]
    "We assume that v_i serves as an approximate inverse measure of the KL divergence between the output distributions: v_i≈−D_KL(P(y|x,G_test)||P(y|x,G_i)). ... Since σ_test^2 is constant across models, the OMOG similarity score v_i can be approximated as: v_i≈−1/(2σ_i^2). Thus, the similarity score v_i serves as an inverse proxy for the predictive variance σ_i^2, which aligns with Bayesian Model Averaging (BMA) theory."

    The claimed derivation of the inverse-variance proxy starts from an assumption that v_i already equals the negative KL divergence, i.e., from the Bayesian-model-comparison quantity the theorem is supposed to justify. Moreover, under the paper's own equal-mean Gaussian formula, the KL is approximately 0.5(σ_i^2−σ_test^2), not 0.5/σ_i^2; the σ_test^2 term is discarded and the reciprocal form is inserted without justification. The 'alignment with BMA' is therefore loaded into the definition rather than derived from the scoring model.

  2. self definitional [Appendix D, Proposition D.6]
    "Substituting v_i≈−1/(2σ_i^2) into the softmax function: w_i=exp(−1/(2σ_i^2))/Σ_j exp(−1/(2σ_j^2)). Using the approximation exp(−x)≈1/x for small x: w_i≈(1/σ_i^2)/Σ_j(1/σ_j^2). This is exactly the Bayesian Model Averaging weight, proving that OMOG’s merging strategy is theoretically equivalent to BMA."

    The proof selects the approximation exp(−x)≈1/x, which is mathematically false (the small-x expansion is exp(−x)≈1−x), precisely to force the softmax weights into the inverse-variance BMA form. Even if v_i were exactly −1/(2σ_i^2), softmax would give exp(−0.5/σ_i^2)-shaped weights, not 1/σ_i^2 weights. Thus the claimed equivalence to BMA is not a consequence of the scoring model; it is the desired conclusion manufactured by an unjustified approximation.

1 more flagged steps
  1. other [Appendix D, Theorem D.7]
    "Since the bias of a weighted ensemble is at most the minimum bias among the base models, we assume: E[f_fused(x)]≈E[f_best(x)]. ... Since Var(f_fused(x)) is minimized and Bias^2 does not increase significantly due to Top-K selection, we conclude: E[(f_fused(x)−f*(x))^2]≤min_i E[(f_i(x)−f*(x))^2]."

    The theorem's guarantee—that the fused model matches or beats the best individual model in expectation—is explicitly assumed in the proof via 'we assume E[f_fused(x)]≈E[f_best(x)]' and the variance inequality Var(f_fused)≤min_i Var(f_i) is asserted without proof. For arbitrary softmax weights this variance inequality is not generally true. Consequently, Theorem 2's conclusion is a restatement of the proof's assumptions rather than a demonstrated result, making the advertised 'guarantee' circular.

full rationale

The circularity is concentrated in the formal theory of Appendix D, which is presented in Section 3.2 as the mechanism by which OMOG avoids negative transfer. Definition D.2 assumes the learned relevance score is an approximate negative KL divergence, and the subsequent 'proofs' of BMA equivalence and the Top-K fusion guarantee do not derive this property from the scoring objective; they either insert it by assumption or force it with an invalid approximation (exp(−x)≈1/x) and an unproved bias/variance claim. In this sense, Theorems 1 and 2 are not first-principles results but re-statements of the paper's assumptions. The self-citations to the authors' prior benchmark and structural-disparity work, while present, are not the load-bearing circular element here; the same negative-transfer phenomenon is also attributed to independent work, so I do not base the score on self-citation. The extensive zero-shot and few-shot comparisons against OneForAll, LLaGA, AnyGraph, ZeroG, Prodigy, and other baselines, together with the ablations in Figures 4-6, are self-contained empirical evidence and are not circular. The score reflects the partial circularity of the central theoretical guarantee, not the empirical evaluation, which remains independent.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The method rests on an aligned-language embedding space, a fixed SGC aggregation depth, and several scoring-model assumptions. The theoretical appendix adds Gaussian and mean-alignment assumptions and an unproved variance bound; these are the main ad hoc load-bearing premises. The empirical protocol also assumes leave-one-out pretraining is a fair proxy for real transfer.

free parameters (5)
  • SGC hop count alpha = 4
    Set to 4 in Appendix C; controls how much structural neighborhood information is concatenated into node embeddings and is chosen by hand rather than derived.
  • text embedding dimension = 384
    Set to 384 in Appendix C to balance efficiency and performance.
  • number of fused experts K = 2
    Appendix C chooses top-2 source models during fusion; Figure 6 explores K but the default is hand-selected.
  • learning rate = 0.0001
    Adam initial learning rate, standard but hand-chosen.
  • feature mask ratio = 0.5
    Randomly masks half of node features for contrastive views, following Zhu et al. [40].
assumptions (6)
  • domain assumption Unified text feature space from SentenceBERT aligns node semantics across domains.
    Section 3.1 states a unified feature space is the requisite and adopts SentenceBERT; if text embeddings do not align domains, the whole pretraining bank is built on unaligned features.
  • domain assumption SGC with alpha=4 preserves enough structural information for every source and target graph.
    Equation (2) concatenates H^(0) through H^(4); no analysis shows four hops is sufficient across citation, e-commerce, and knowledge graphs.
  • ad hoc to paper Relevance score v_i approximates inverse KL between test and source output distributions.
    Appendix D Definition D.2 asserts this directly; it is not derived from the cosine-similarity definition.
  • ad hoc to paper Predictive distributions are Gaussian with equal means across graphs.
    Appendix D Assumption D.4 and the proof assume mu_test = mu_i, which essentially assumes the models are already aligned.
  • ad hoc to paper Fused model bias and variance satisfy Var(fused) <= min_i Var(f_i) and bias no larger than minimum bias.
    Theorem D.7 proof states these without proof; they are generally false for arbitrary weighted averages, so the theorem is not established.
  • domain assumption Negative transfer always hurts joint training (Assumption D.3).
    Assumption D.3 postulates joint model loss exceeds every relevant single model; this motivates the bank design but is not empirically established in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of One Model for One Graph: A New Perspective for Pretraining with Cross-domain Graphs." pith.science (2026). https://pith.science/paper/SPRDKVOA

@misc{pith2026241200315,
  author       = {Pith},
  title        = {Pith review of: One Model for One Graph: A New Perspective for Pretraining with Cross-domain Graphs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SPRDKVOA}},
  note         = {Machine review of arXiv:2412.00315}
}
read the original abstract

Graph Neural Networks (GNNs) have emerged as a powerful tool to capture intricate network patterns, achieving success across different domains. However, existing GNNs require careful domain-specific architecture designs and training from scratch on each dataset, leading to an expertise-intensive process with difficulty in generalizing across graphs from different domains. Therefore, it can be hard for practitioners to infer which GNN model can generalize well to graphs from their domains. To address this challenge, we propose a novel cross-domain pretraining framework, "one model for one graph," which overcomes the limitations of previous approaches that failed to use a single GNN to capture diverse graph patterns across domains with significant gaps. Specifically, we pretrain a bank of expert models, with each one corresponding to a specific dataset. When inferring to a new graph, gating functions choose a subset of experts to effectively integrate prior model knowledge while avoiding negative transfer. Extensive experiments consistently demonstrate the superiority of our proposed method on both link prediction and node classification tasks.

Figures

Figures reproduced from arXiv: 2412.00315 by the authors.

Figure 1
Figure 1. Existing “one model for all graphs" pipeline vs. the proposed “one model for one graph" pipeline . [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. An illustration of the pretraining stage. The first step encodes the node attributes with language models and then [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. An illustration of the inference stage. We input the test graph features into each scoring model to calculate the [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The impact for key components on OMOG Top KNo weightsRandom K Least K 39 40 41 42 Accuracy Node classification Top KNo weightsRandom K Least K 40 42 44 46 Hits@100 Link prediction [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: The performance of different scoring module de [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: The effect of the number of source models. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Beyond Feature and Structure Alignment: Learning Transferable Propagation Knowledge for Graph Foundation Models

    cs.LG 2026-07 conditional novelty 6.0 of 10

    ProGFM transfers graph knowledge across domains by learning a prototype bank of per-edge, per-dimension propagation strengths and using them to modulate message passing on unseen graphs.

  2. H$^2$GFM: Towards unifying Homogeneity and Heterogeneity on Text-Attributed Graphs

    cs.LG 2025-06 conditional novelty 6.0 of 10

    A graph foundation model with text-encoded meta-relations and a mixture of context-adaptive transformers improves accuracy across homogeneous and heterogeneous text-attributed graphs.

Reference graph

Works this paper leans on

42 extracted references · 9 canonical work pages · cited by 2 Pith papers

  1. [1]

    Fedor Borisyuk, Shihai He, Yunbo Ouyang, Morteza Ramezani, Peng Du, Xiaochen Hou, Chengming Jiang, Nitin Pasumarthy, Priya Bannur, Birjodh Tiwana, et al

  2. [2]

    Runjin Chen, Tong Zhao, Ajay Jaiswal, Neil Shah, and Zhangyang Wang. 2024. LLaGA: Large Language and Graph Assistant. arXiv preprint arXiv:2402.08170 (2024)

    arXiv 2024

  3. [3]

    Zhikai Chen, Haitao Mao, Hang Li, Wei Jin, Haifang Wen, Xiaochi Wei, Shuaiqiang Wang, Dawei Yin, Wenqi Fan, Hui Liu, and Jiliang Tang. 2023. Exploring the Potential of Large Language Models (LLMs) in Learning on Graphs. ArXiv abs/2307.03393 (2023)

    arXiv 2023

  4. [4]

    Zhikai Chen, Haitao Mao, Jingzhe Liu, Yu Song, Bingheng Li, Wei Jin, Bahare Fatemi, Anton Tsitsulin, Bryan Perozzi, Hui Liu, et al. 2024. Text-space Graph Foundation Models: Comprehensive Benchmarks and New Insights. arXiv preprint arXiv:2406.10727 (2024)

    arXiv 2024

  5. [5]

    Wenqi Fan, Yao Ma, Qing Li, Yuan He, Eric Zhao, Jiliang Tang, and Dawei Yin

  6. [6]

    Yuxin Guo, Cheng Yang, Yuluo Chen, Jixi Liu, Chuan Shi, and Junping Du. 2023. A Data-centric Framework to Endow Graph Neural Networks with Out-Of- Distribution Detection Ability. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 638–648

    work page 2023

  7. [7]

    Yufei He and Bryan Hooi. 2024. UniGraph: Learning a Cross-Domain Graph Foundation Model From Natural Language. arXiv preprint arXiv:2402.13630 (2024)

    arXiv 2024

  8. [8]

    Zhenyu Hou, Haozhan Li, Yukuo Cen, Jie Tang, and Yuxiao Dong. 2024. GraphAlign: Pretraining One Graph Neural Network on Multiple Graphs via Feature Alignment. arXiv preprint arXiv:2406.02953 (2024)

    arXiv 2024

Show all 42 references
  1. [9]

    Zhenyu Hou, Xiao Liu, Yukuo Cen, Yuxiao Dong, Hongxia Yang, Chunjie Wang, and Jie Tang. 2022. Graphmae: Self-supervised masked graph autoencoders. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 594–604

    2022

  2. [10]

    Qian Huang, Hongyu Ren, Peng Chen, Gregor Kržmanc, Daniel Zeng, Percy Liang, and Jure Leskovec. 2023. PRODIGY: Enabling In-context Learning Over Graphs. arXiv preprint arXiv:2305.12600 (2023)

    2023 arXiv

  3. [11]

    Wei Jin, Xiaorui Liu, Xiangyu Zhao, Yao Ma, Neil Shah, and Jiliang Tang. 2021. Automated self-supervised learning for graphs. arXiv preprint arXiv:2106.05470 (2021)

    2021 arXiv

  4. [12]

    Mingxuan Ju, Tong Zhao, Qianlong Wen, Wenhao Yu, Neil Shah, Yanfang Ye, and Chuxu Zhang. 2023. Multi-task Self-supervised Graph Neural Networks Enable Stronger Task Generalization. (2023)

    2023

  5. [13]

    Diederik P Kingma. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

    2014 arXiv

  6. [14]

    Mingxin Li et al. 2024. Tackling Task Conflict and Data Imbalance through Model Merging. arXiv preprint arXiv:2410.15035 (2024). https://arxiv.org/abs/2410.15035

    2024 arXiv

  7. [15]

    Yuhan Li, Peisong Wang, Zhixun Li, Jeffrey Xu Yu, and Jia Li. 2024. Zerog: Investigating cross-dataset zero-shot transferability in graphs. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 1725–1735

    2024

  8. [16]

    Hao Liu, Jiarui Feng, Lecheng Kong, Ningyue Liang, Dacheng Tao, Yixin Chen, and Muhan Zhang. 2023. One for All: Towards Training One Graph Model for All Classification Tasks. arXiv preprint arXiv:2310.00149 (2023)

    2023 arXiv

  9. [17]

    Zemin Liu, Xingtong Yu, Yuan Fang, and Xinming Zhang. 2023. Graphprompt: Unifying pre-training and downstream tasks for graph neural networks. In Pro- ceedings of the ACM Web Conference 2023 . 417–428

    2023

  10. [18]

    Yao Ma, Xiaorui Liu, Neil Shah, and Jiliang Tang. 2021. Is homophily a necessity for graph neural networks? arXiv preprint arXiv:2106.06134 (2021)

    2021 arXiv

  11. [19]

    Yao Ma and Jiliang Tang. 2021. Deep learning on graphs . Cambridge University Press

    2021

  12. [20]

    Haitao Mao, Zhikai Chen, Wei Jin, Haoyu Han, Yao Ma, Tong Zhao, Neil Shah, and Jiliang Tang. 2023. Demystifying Structural Disparity in Graph Neural Networks: Can One Size Fit All? arXiv preprint arXiv:2306.01323 (2023)

    2023 arXiv

  13. [21]

    Haitao Mao, Zhikai Chen, Wenzhuo Tang, Jianan Zhao, Yao Ma, Tong Zhao, Neil Shah, Michael Galkin, and Jiliang Tang. 2024. Graph foundation models. arXiv preprint arXiv:2402.02216 (2024)

    2024 arXiv

  14. [22]

    Michael S Matena and Colin A Raffel. 2022. Merging models with fisher-weighted averaging. Advances in Neural Information Processing Systems 35 (2022), 17703– 17716

    2022

  15. [23]

    Jiezhong Qiu, Qibin Chen, Yuxiao Dong, Jing Zhang, Hongxia Yang, Ming Ding, Kuansan Wang, and Jie Tang. 2020. Gcc: Graph contrastive coding for graph neural network pre-training. InProceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining...

    2020

  16. [24]

    Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084 (2019)

    2019 arXiv

  17. [25]

    Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. 2017. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. arXiv preprint arXiv:1701.06538 (2017)

    2017 arXiv

  18. [26]

    Mingchen Sun, Kaixiong Zhou, Xin He, Ying Wang, and Xin Wang. 2022. GPPT: Graph Pre-training and Prompt Tuning to Generalize Graph Neural Networks. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (Washington DC, USA) (KDD ’22). Associati...

    2022

  19. [27]

    Xiangguo Sun, Hong Cheng, Jia Li, Bo Liu, and Jihong Guan. 2023. All in One: Multi-Task Prompting for Graph Neural Networks. (2023)

    2023

  20. [28]

    Jiabin Tang, Yuhao Yang, Wei Wei, Lei Shi, Lixin Su, Suqi Cheng, Dawei Yin, and Chao Huang. 2023. Graphgpt: Graph instruction tuning for large language models. arXiv preprint arXiv:2310.13023 (2023)

    2023 arXiv

  21. [29]

    Xianfeng Tang, Yozen Liu, Neil Shah, Xiaolin Shi, Prasenjit Mitra, and Suhang Wang. 2020. Knowing your fate: Friendship, action and temporal explanations for user engagement prediction on social apps. In Proceedings of the 26th ACM SIGKDD international conference on knowledge ...

    2020

  22. [30]

    A Vaswani. 2017. Attention is all you need. Advances in Neural Information Processing Systems (2017)

    2017

  23. [31]

    Felix Wu, Amauri Souza, Tianyi Zhang, Christopher Fifty, Tao Yu, and Kilian Weinberger. 2019. Simplifying graph convolutional networks. In International conference on machine learning . PMLR, 6861–6871

    2019

  24. [32]

    Lianghao Xia and Chao Huang. 2024. AnyGraph: Graph Foundation Model in the Wild. arXiv preprint arXiv:2408.10700 (2024)

    2024 arXiv

  25. [33]

    Jiarong Xu, Renhong Huang, Xin Jiang, Yuxuan Cao, Carl Yang, Chunping Wang, and Yang Yang. 2023. Better with less: A data-active perspective on pre-training graph neural networks. Advances in Neural Information Processing Systems 36 (2023), 56946–56978

    2023

  26. [34]

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

    2016

  27. [35]

    Chengxuan Ying, Tianle Cai, Shengjie Luo, Shuxin Zheng, Guolin Ke, Di He, Yanming Shen, and Tie-Yan Liu. 2021. Do Transformers Really Perform Badly for Graph Representation?. In Thirty-Fifth Conference on Neural Information Processing Systems. https://openreview.net/forum?id=O...

    2021

  28. [36]

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

    2018

  29. [38]

    Haihong Zhao, Aochuan Chen, Xiangguo Sun, Hong Cheng, and Jia Li. 2024. All in one and one for all: A simple yet effective method towards cross-domain graph pretraining. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 4443–4454

    2024

  30. [39]

    Jianan Zhao, Hesham Mostafa, Michael Galkin, Michael Bronstein, Zhaocheng Zhu, and Jian Tang. 2024. GraphAny: A Foundation Model for Node Classification on Any Graph. arXiv preprint arXiv:2405.20445 (2024)

    2024 arXiv

  31. [40]

    Yanqiao Zhu, Yichen Xu, Qiang Liu, and Shu Wu. 2021. An Empirical Study of Graph Contrastive Learning. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2)

    2021

  32. [41]

    Yanqiao Zhu, Xu Yichen, Yu Feng, Liu Qiang, Wu Shu, and Wang Liang. 2020. Deep graph contrastive representation learning. arXiv preprint arXiv:2006.04131 (2020)

    2020 arXiv

  33. [2019]

    In The world wide web conference

    Graph neural networks for social recommendation. In The world wide web conference. 417–426

  34. [2024]

    LiGNN: Graph Neural Networks at LinkedIn.arXiv preprint arXiv:2402.11139 (2024)

    2024 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.