Pith. sign in

REVIEW 4 major objections 5 minor 50 references

TANGNN: a Concise, Scalable and Effective Graph Neural Networks with Top-m Attention Mechanism for Graph Representation Learning

T0 review · 4 major / 5 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 TANGNN adds a Top-m attention branch to neighborhood aggregation, giving each GNN layer both local and long-range information at near-linear cost.

desk verdict The top-m mechanism as implemented is not the per-node attention the paper claims, the experiments are too weak to rescue it, but the ArXivNet dataset and the anchor-vector efficiency trick are salvageable. read the letter →

arxiv 2411.15458 v1 pith:GK2GG6ND submitted 2024-11-23 cs.LG cs.AI

Jiawei E , Yinglong Zhang , Xuewen Xia , Xing Xu This is my paper · ORCID
classification cs.LGcs.AI
keywords graphneuralnetworkstop-mattentionneighborhoodaggregationrepresentationlearninglinkpredictioncitationsentimentregressionscalable
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

TANGNN is a graph neural network architecture designed to give each node a wide receptive field without the quadratic cost of graph transformers. It runs two aggregation branches in every layer: a standard sampled-neighborhood aggregation and a Top-m attention branch that selects the m most similar nodes and applies self-attention to them. The paper claims this two-branch design outperforms established GNNs and graph transformers on vertex classification, link prediction, citation-sentiment prediction, graph regression, and visualization, while running faster on large graphs. For the sentiment task it introduces ArXivNet, a citation network labeled positive, neutral, or negative, which the authors say is a first for GNN research. If the empirical claims hold, TANGNN offers a practical middle path between shallow local GNNs and expensive global attention models.

What carries the argument

The load-bearing object is the Top-m efficient algorithm: an auxiliary vector a, kept orthogonal to the mean node vector by the update $a = \mathrm{L2Norm}(a - (a^T \bar{g})\bar{g})$, scores every node by cosine similarity $s_n = a^T \hat{g}_n$, and sorting these scores produces one global list S whose top m entries form the attention set. This turns all-pairs similarity from $O(N^2)$ to $O(N)$. The other mechanism is the neighborhood aggregation branch, which samples a fixed number of neighbors and applies a mean aggregator in a GraphSAGE-style pattern. Each layer concatenates the two branch outputs and sends them through an MLP; TANGNN-LC additionally concatenates outputs across layers so the final representation keeps both shallow and deep information.

What would settle it

Run TANGNN on a graph where the global top-m set differs sharply from the per-node top-m set—for example, two dense clusters whose feature directions are orthogonal, so each cluster's most relevant nodes are inside the cluster but the global ranking is dominated by one cluster—and compare TANGNN's accuracy with a variant that computes the true per-node top-m. If the global-ranking model loses, the auxiliary-vector shortcut, not per-node attention, is what the implementation actually uses; inspecting the Top-m function in the released code to see whether it returns a per-node subset or the same sorted list S for all nodes would settle the question directly.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a GNN layer can combine local message passing with a cheap global-attention-style component by selecting the Top-m most similar nodes and applying scaled dot-product attention only within that small set. Similarity is not computed between all node pairs; instead each node's embedding is scored against one learnable auxiliary vector a using cosine similarity, the scores are sorted once, and the top m entries become the attention set. The two branches, neighborhood aggregation and Top-m attention, are concatenated and passed through an MLP at each layer, and the LC variant concatenates all layer outputs before a final MLP. The paper reports that the resulting model and its variants beat GCN, GraphSAGE, GAT, GIN, JK-Net, Graphormer, TransGNN, NAGphormer, SAT, DeepGraph, and SGFormer on the tasks it evaluates, and that the Top-m sampling keeps memory and runtime low enough for datasets where some graph transformers run out of memory.

Load-bearing premise

The load-bearing premise is that ranking every node by cosine similarity to one shared auxiliary vector a correctly identifies the nodes most relevant to each individual node, even though the ranking is the same for all nodes.

Editorial extensions

If this is right

  • Graph models can expand their receptive field to distant nodes without stacking many layers, which is the setting where oversmoothing typically degrades GNN accuracy.
  • Because similarity scoring is $O(N)$ and both branches sample a fixed number of nodes per layer, the model's memory and runtime scale better than graph transformers on large graphs; the paper shows Graphormer and SAT overflowing memory on Reddit while TANGNN converges.
  • The TANGNN-LC variant, which concatenates layer outputs, gives a direct way to keep early-layer local information in the final embedding, which the paper links to its best classification results.
  • Citation sentiment prediction on ArXivNet becomes a new benchmark task for GNNs, with edges labeled positive, neutral, or negative rather than only nodes or whole graphs.

Reading between the lines

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

  • The paper describes the Top-m set as node-specific, but its Section 4.3 implementation sorts all nodes by $a^T \hat{g}_n$ once and reuses the same sorted list for every node; the authors leave implicit that the attention branch is therefore aggregating a global set of prototype-similar nodes rather than per-node nearest neighbors.
  • A testable consequence is that TANGNN's gains could come from a shared global context rather than pairwise relevance, which would connect it to global-memory or graph-pooling designs; comparing it with a true per-node top-m variant would separate those explanations.
  • The auxiliary-vector transitivity assumption is the natural extension point: on graphs with strong cluster structure, a single direction a cannot represent all relevance relations, so a multi-vector or per-cluster extension is a straightforward experiment.
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 / 5 minor

Summary. The paper proposes TANGNN, a GNN layer that combines a sampled neighborhood aggregation component with a 'Top-m attention' component. The stated goal is to give each node a receptive field consisting of both its local neighbors and the m nodes most similar to it, without the quadratic cost of full graph transformers. An efficient variant replaces all-pairs similarity with a single auxiliary vector a and a sorting step, and a layer-concatenation variant TANGNN-LC is also presented. The authors introduce ArXivNet, a citation-sentiment dataset, and report experiments on node classification, link prediction, sentiment prediction, graph regression, and visualization against GCN, GraphSAGE, GAT, GIN, JK-Net, and several graph transformer baselines. The main claims are that TANGNN expands the receptive field while remaining scalable, and that it outperforms existing methods across multiple tasks.

Significance. If the mechanism worked as described, TANGNN would be an interesting intermediate point between local GNNs and global graph transformers, and ArXivNet would be a useful resource for citation-sentiment analysis. The paper also releases code and data, which is a concrete asset, and I see no circular reasoning: the choice of m is a hyperparameter selected by sensitivity analysis. However, the implementation in Section 4.3 does not realize the per-node similarity-based selection promised in the abstract and Section 4.1, and the reported experiments are too weak to establish the claimed superiority. The significance of the contribution therefore rests on correcting the mechanism and substantially strengthening the evaluation.

major comments (4)
  1. [§4.3, §4.2.2, Algorithm 3] The efficient Top-m algorithm is not the per-node attention mechanism described in Section 4.1. Eq. (10) computes a single score s_n = a^T ĝ_n for each node against one global auxiliary vector a, and Eq. (11) produces one sorted list S. If the first m entries of S are used, every node receives the same H_i_Top-m; if instead each node takes its rank-neighbors in S, the selected nodes are those with similar projections onto a, not the nodes most similar to v in feature space. In neither reading is the per-node Top-m selection of Section 4.1 and Figure 1 implemented. Moreover, Eqs. (1)-(4) derive query, key, and value matrices all from H_i_Top-m, so there is no query vector for the central node v; the attention weight A_uv in Eq. (4) is undefined as a node-specific weight unless v is itself in the selected set, which the algorithm does not guarantee. The 'transitive property of similarity' invoked to justify the anchor-vector proxy is an unvalidated assumption that can fail in high-dimensional spaces.
  2. [§5.1.2, Tables 3-8] The experimental evidence is reported as single runs without standard deviations, confidence intervals, or significance tests, and Section 5.1.2 fixes the learning rate, batch size, and depth uniformly for all baselines rather than tuning each method. The differences in the tables are often small, for example Table 3 at 10% training gives TANGNN-LC 0.9657 versus SGFormer 0.9570, which could easily be within run-to-run noise. As presented, the data do not support the abstract's claim that the method 'outperforms existing methods.'
  3. [§5.1.1, §5.2.1, Tables 3-6] The evaluation protocol is inconsistent across tasks and datasets: node classification uses AUROC on Cora and Citeseer under arbitrary 10% to 90% training splits (Tables 3 and 4), F1-micro on PubMed (Table 5), and accuracy on ArXivNet (Table 6), while graph regression uses MAE (Tables 7 and 8). For standard transductive benchmarks like Cora and Citeseer, AUROC with varied training fractions is not the usual protocol and makes comparison with prior work difficult. A consistent protocol with standard splits and the appropriate per-task metric is needed for the claimed cross-task superiority.
  4. [§4.2, Algorithm 4, Eq. (7)] The loss function in Eq. (7), L(p,q) = -sum(p log q + (1-p) log(1-q)), is inconsistent with Algorithm 4, where p ← softmax(g_v) and q ← one_hot(y). If q is a one-hot vector, log q is undefined for zero entries, and the text states that p is the desired output while q is the actual output, reversing the assignments in Algorithm 4. This makes the training objective ambiguous and prevents reproduction of the training procedure as specified.
minor comments (5)
  1. [§5.1.2] The baseline list mentions 'SAGEFormer,' but no such model appears in the results tables; this is likely a typo for 'SGFormer' and should be corrected.
  2. [§5.2.1, Figure 5] Figure 5 varies a parameter K without defining it, while Section 5.1 fixes the number of layers L to 2; the relationship between K and L should be clarified.
  3. [§5.2.1, Tables 3 and 5] The reported percentage improvements are not consistent with the tables: on Cora at 50% training, TANGNN-LC gives 0.9735 versus GraphSAGE 0.9056, an absolute gain of about 6.8 points rather than 12%, and on PubMed at 10% training the gain over TANGNN is about 0.4 points rather than 2%.
  4. [§4.3, Eq. (9)] There are typographical errors in the orthogonality proof, including '∥¯g2∥' instead of '∥¯g∥^2', and the statement that feature vectors with high similarity are mapped close to a after orthogonalizing to the mean is not generally true and should be justified or removed.
  5. [Tables 1-8] There are numerous formatting and spelling errors, including 'Dateset' in Table 1, 'ZIN C' in Section 5.1.1, 'GAT 07105' in Table 4, '01685' in Table 8, and 'Sigmod' instead of 'Sigmoid' in Section 4.2.1; a careful proofread is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: TANGNN's claims are empirical and benchmark-checked; the global-auxiliary Top-m selection is a heuristic approximation, not a circular derivation.

full rationale

TANGNN makes no first-principles claim that reduces to its inputs: the architecture is defined by Algorithms 1-3, and its performance claims are comparisons on external benchmarks (Cora, Citeseer, PubMed, Amazon, Reddit, ZINC, QM9, ArXivNet) with held-out training fractions. No fitted parameter is renamed as a prediction: the single hyperparameter m is selected by sensitivity analysis and then used for all experiments, which is standard tuning rather than a forced construction. The only derivation in Section 4.3 is the orthogonality verification for the auxiliary vector a, which is a direct algebraic consequence of the update rule; even if the unit-norm assumption on the average vector is questionable, that is a correctness defect, not circularity. The Top-m efficient algorithm's reliance on a single global auxiliary vector and the transitivity of similarity is an approximation of per-node Top-m selection; if the approximation fails, the implementation is unfaithful to the prose description, but the claim is not equivalent to its own input by construction. The self-citation to GraphSAGE++ appears only as related work and is not load-bearing.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The central claim rests on the anchor-vector transitivity assumption, the choice of m and sampling sizes, and the trustworthiness of the ArXivNet labels. The model itself is a standard trainable architecture with no hidden derivation.

free parameters (3)
  • m (top-m count) = 30
    Set to 30 after sensitivity analysis on Cora and PubMed (Section 5.2.7); controls the size of the attention receptive field and is central to the reported results.
  • neighbor sampling sizes = [20, 10] for two layers
    Chosen in Section 5.1 with no sensitivity analysis; determines the scope of local aggregation and affects the balance between local and global information.
  • number of layers L = 2
    Set to 2 due to computational burden (Section 5.1); depth directly influences the receptive field and the oversmoothing argument.
assumptions (4)
  • ad hoc to paper Cosine similarity to a shared auxiliary vector a is a valid proxy for pairwise node similarity (transitivity of similarity).
    Stated in Section 4.3: the higher the similarity between two nodes and a, the more similar these two nodes are. This is not proven and is generally false; it is the basis for replacing per-node top-m selection with a single global ranking.
  • domain assumption A fixed-size random sample of neighbors preserves the information needed for local aggregation.
    Adopted from GraphSAGE in Section 4.1 and Algorithm 2; standard but unverified for these datasets and downstream tasks.
  • domain assumption SPECTER embeddings and DictSentiBERT labels are reliable node features and ground-truth sentiment annotations for ArXivNet.
    Section 5.1.1 describes using SPECTER for features and DictSentiBERT to annotate sentiment; no human validation or agreement statistics are reported.
  • standard math Softmax attention, layer normalization, MLP, and Adam behave as standard background machinery.
    Used throughout Section 4.2; not in question for this review.
invented entities (1)
  • Auxiliary vector a
    purpose: Serves as an anchor whose cosine similarity to node embeddings ranks all nodes and defines the top-m receptive field for every node.
    Introduced in Section 4.3, Eqs. (8) to (10). No external falsifiable prediction is attached; its effectiveness rests on the unproven transitivity assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TANGNN: a Concise, Scalable and Effective Graph Neural Networks with Top-m Attention Mechanism for Graph Representation Learning." pith.science (2026). https://pith.science/paper/GK2GG6ND

@misc{pith2026241115458,
  author       = {Pith},
  title        = {Pith review of: TANGNN: a Concise, Scalable and Effective Graph Neural Networks with Top-m Attention Mechanism for Graph Representation Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GK2GG6ND}},
  note         = {Machine review of arXiv:2411.15458}
}
read the original abstract

In the field of deep learning, Graph Neural Networks (GNNs) and Graph Transformer models, with their outstanding performance and flexible architectural designs, have become leading technologies for processing structured data, especially graph data. Traditional GNNs often face challenges in capturing information from distant vertices effectively. In contrast, Graph Transformer models are particularly adept at managing long-distance node relationships. Despite these advantages, Graph Transformer models still encounter issues with computational and storage efficiency when scaled to large graph datasets. To address these challenges, we propose an innovative Graph Neural Network (GNN) architecture that integrates a Top-m attention mechanism aggregation component and a neighborhood aggregation component, effectively enhancing the model's ability to aggregate relevant information from both local and extended neighborhoods at each layer. This method not only improves computational efficiency but also enriches the node features, facilitating a deeper analysis of complex graph structures. Additionally, to assess the effectiveness of our proposed model, we have applied it to citation sentiment prediction, a novel task previously unexplored in the GNN field. Accordingly, we constructed a dedicated citation network, ArXivNet. In this dataset, we specifically annotated the sentiment polarity of the citations (positive, neutral, negative) to enable in-depth sentiment analysis. Our approach has shown superior performance across a variety of tasks including vertex classification, link prediction, sentiment prediction, graph regression, and visualization. It outperforms existing methods in terms of effectiveness, as demonstrated by experimental results on multiple datasets.

Figures

Figures reproduced from arXiv: 2411.15458 by the authors.

Figure 1
Figure 1. TANGNN Architecture common practice is to increase the number of layers in GNNs, allowing nodes to aggregate information from farther neighbors, thus expanding the model’s receptive field. However, this increase in layers leads to the problem of over￾smoothing. Transformer models have a global receptive field, but as the sequence length increases, the computational complexity and storage require￾ments grow quadratic… view at source ↗
Figure 2
Figure 2. Illustration of the Top-m Efficient Algorithm. An auxiliary vector a is introduced into the model to calculate the cosine similarity between node vector representations and the auxiliary vector, determining the similarity between nodes. This method utilizes the transfer property of similarity, reducing the necessary computational complexity. The value of m is set to 2, the two most similar nodes to each node are its… view at source ↗
Figure 3
Figure 3. TANGNN-LC architecture 16 [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: TANGNN-FLC architecture 18 [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: Accuracy values for vertex classification tasks on Cora and PubMed datasets vary with K values dataset is 10%). Overall, compared to other algorithms, TANGNN-LC and TANGNN both have achieved better performance in classification tasks [PITH_FULL_IMAGE:figures/full_fig_…
Figure 6
Figure 6. Figure 6: Accuracy values obtained by completing link prediction tasks on four datasets. a higher level compared to GraphSAGE in the initial stage and maintains a relatively stable state throughout the entire training process. 5.2.6. VISUALIZATION Firstly, we generate representa…
Figure 7
Figure 7. Figure 7: The time spent on link prediction tasks on the ArXivNet and Reddit datasets [PITH_FULL_IMAGE:figures/full_fig_p029_7.png]
Figure 8
Figure 8. Figure 8: The training process of TANGNN and GraphSAGE 29 [PITH_FULL_IMAGE:figures/full_fig_p029_8.png]
Figure 9
Figure 9. Figure 9: Visualization results on the PubMed dataset. 6. Conclusion In this study, we successfully developed the TANGNN framework, which significantly enhances the efficiency and accuracy of graph neural networks in processing large-scale graph data through an effective integra…
Figure 10
Figure 10. Figure 10: Selection of m for vertex classification tasks on the Cora and PubMed datasets data analysis, notably, TANGNN-LC outperformed existing methods on mul￾tiple evaluation metrics. Additionally, We developed the ArXivNet dataset by extracting papers from unarXive (Saier et…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 40 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  3. [3]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize "" * " " * ...

  4. [4]

    author Armah-Sekum, R. E. , author Szedmak, S. , & author Rousu, J. ( year 2024 ). title Protein function prediction through multi-view multi-label latent tensor reconstruction . journal BMC bioinformatics \/ , volume 25 \/ , pages 174

    work page 2024

  5. [5]

    , author Peters, M

    author Beltagy, I. , author Peters, M. E. , & author Cohan, A. ( year 2020 ). title Longformer: The long-document transformer . journal CoRR \/ , volume abs/2004.05150 \/

    arXiv 2020

  6. [6]

    , author Nasiri, E

    author Berahmand, K. , author Nasiri, E. , author Forouzandeh, S. , & author Li, Y. ( year 2022 ). title A preference random walk algorithm for link prediction through mutual influence nodes in complex networks . journal Journal of king saud university-computer and information sciences \/ , volume 34 \/ , pages 5375--5387

    work page 2022

  7. [7]

    , author Nasiri, E

    author Berahmand, K. , author Nasiri, E. , author Rostami, M. , & author Forouzandeh, S. ( year 2021 ). title A modified deepwalk method for link prediction in attributed social network . journal Computing \/ , volume 103 \/ , pages 2227--2249

    work page 2021

  8. [8]

    , author Alon, U

    author Brody, S. , author Alon, U. , & author Yahav, E. ( year 2022 ). title How attentive are graph attention networks? In booktitle International Conference on Learning Representations \/

    work page 2022

Show all 50 references
  1. [9]

    , & author Musco, C

    author Chanpuriya, S. , & author Musco, C. ( year 2020 ). title Infinitewalk: Deep network embeddings as laplacian embeddings with a nonlinearity . journal Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining \/ , (p. pages 1325–1333...

    2020

  2. [10]

    , author Lin, Y

    author Chen, D. , author Lin, Y. , author Li, W. , author Li, P. , author Zhou, J. , & author Sun, X. ( year 2020 ). title Measuring and relieving the over-smoothing problem for graph neural networks from the topological view . In booktitle Proceedings of the AAAI conference o...

    2020

  3. [11]

    , author O’Bray, L

    author Chen, D. , author O’Bray, L. , & author Borgwardt, K. ( year 2022 ). title Structure-aware transformer for graph representation learning . In booktitle Proceedings of the 39th International Conference on Machine Learning (ICML) \/ (pp. pages 3469--3489 ). organization PMLR

    2022

  4. [12]

    , author Gao, K

    author Chen, J. , author Gao, K. , author Li, G. , & author He, K. ( year 2023 ). title NAG phormer: A tokenized graph transformer for node classification in large graphs . In booktitle The Eleventh International Conference on Learning Representations \/

    2023

  5. [13]

    , author Feldman, S

    author Cohan, A. , author Feldman, S. , author Beltagy, I. , author Downey, D. , & author Weld, D. S. ( year 2020 ). title Specter: Document-level representation learning using citation-informed transformers . In booktitle ACL \/ (pp. pages 2270--2282 ). https://doi.org/10.186...

    2020 doi

  6. [14]

    author Dwivedi, V. P. , & author Bresson, X. ( year 2021 ). title A generalization of transformer networks to graphs . journal AAAI Workshop on Deep Learning on Graphs: Methods and Applications \/ ,

    2021

  7. [15]

    , author Denoyer, L

    author Gao, S. , author Denoyer, L. , & author Gallinari, P. ( year 2011 ). title Temporal link prediction by integrating content and structure information . In booktitle International Conference on Information and Knowledge Management \/

    2011

  8. [16]

    , & author Leskovec, J

    author Grover, A. , & author Leskovec, J. ( year 2016 ). title node2vec: Scalable feature learning for networks . journal Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining \/ , (p. pages 855–864 ). https://doi.org/10.1145/293967...

    2016

  9. [17]

    author Hamilton, W. L. , author Ying, Z. , & author Leskovec, J. ( year 2017 ). title Inductive representation learning on large graphs . In booktitle NIPS \/ (pp. pages 1025--1035 )

    2017

  10. [18]

    author Irwin, J. J. , & author Shoichet, B. K. ( year 2005 ). title Zinc- a free database of commercially available compounds for virtual screening . journal Journal of chemical information and modeling \/ , volume 45 \/ , pages 177--182

    2005

  11. [19]

    , author Zhang, Y

    author Jiawei, E. , author Zhang, Y. , author Yang, S. , author Wang, H. , author Xia, X. , & author Xu, X. ( year 2024 ). title Graphsage++: Weighted multi-scale gnn for graph representation learning . journal Neural Process. Lett. \/ , volume 56 \/ , pages 24

    2024

  12. [20]

    author Kipf, T. N. , & author Welling, M. ( year 2017 ). title Semi-supervised classification with graph convolutional networks . In booktitle International Conference on Learning Representations \/

    2017

  13. [21]

    , author Liu, Z

    author Lin, Y. , author Liu, Z. , author Sun, M. , author Liu, Y. , & author Zhu, X. ( year 2015 ). title Learning entity and relation embeddings for knowledge graph completion . In booktitle Proceedings of the AAAI conference on artificial intelligence \/ . volume volume 29

    2015

  14. [22]

    , & author Hinton, G

    author van der Maaten, L. , & author Hinton, G. E. ( year 2008 ). title Visualizing data using t-sne . journal Journal of Machine Learning Research \/ , volume 9 \/ , pages 2579--2605

    2008

  15. [23]

    , author Chen, R

    author Min, E. , author Chen, R. , author Bian, Y. , author Xu, T. , author Zhao, K. , author Huang, W. , author Zhao, P. , author Huang, J. , author Ananiadou, S. , & author Rong, Y. ( year 2022 ). title Transformer for graphs: An overview from architecture perspective . jour...

    2022 arXiv

  16. [24]

    , author Wenkel, F

    author Min, Y. , author Wenkel, F. , & author Wolf, G. ( year 2020 ). title Scattering gcn: Overcoming oversmoothness in graph convolutional networks . journal Advances in neural information processing systems \/ , volume 33 \/ , pages 14498--14508

    2020

  17. [25]

    , author Ragazzi, L

    author Moro, G. , author Ragazzi, L. , author Valgimigli, L. , author Frisoni, G. , author Sartori, C. , & author Marfia, G. ( year 2023 ). title Efficient memory-enhanced transformer for long-document summarization in low-resource regimes . journal Sensors \/ , volume 23 \/ ,...

    2023

  18. [26]

    , author Hieu, N

    author Nguyen, K. , author Hieu, N. M. , author Nguyen, V. D. , author Ho, N. , author Osher, S. , & author Nguyen, T. M. ( year 2023 ). title Revisiting over-smoothing and over-squashing using ollivier-ricci curvature . In booktitle International Conference on Machine Learnin...

    2023

  19. [27]

    , author Al-Rfou, R

    author Perozzi, B. , author Al-Rfou, R. , & author Skiena, S. ( year 2014 ). title Deepwalk: Online learning of social representations . In booktitle Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining \/ (pp. pages 701--710 )

    2014

  20. [28]

    , author Dral, P

    author Ramakrishnan, R. , author Dral, P. O. , author Rupp, M. , & author Von Lilienfeld, O. A. ( year 2014 ). title Quantum chemistry structures and properties of 134 kilo molecules . journal Scientific data \/ , volume 1 \/ , pages 1--7

    2014

  21. [29]

    , & author F \"a rber, M

    author Saier, T. , & author F \"a rber, M. ( year 2020 ). title unarxive: a large scholarly data set with publications’ full-text, annotated in-text citations, and links to metadata . journal Scientometrics \/ , volume 125 \/ , pages 3085 -- 3108

    2020

  22. [30]

    , author Krause, J

    author Saier, T. , author Krause, J. , & author F \"a rber, M. ( year 2023 ). title unarxive 2022: All arxiv publications pre-processed for nlp, including structured full-text and citation network . In booktitle 2023 ACM/IEEE Joint Conference on Digital Libraries (JCDL) \/ (pp...

    2023

  23. [31]

    , author Namata, G

    author Sen, P. , author Namata, G. , author Bilgic, M. , author Getoor, L. , author Galligher, B. , & author Eliassi-Rad, T. ( year 2008 ). title Collective classification in network data . journal AI magazine \/ , volume 29 \/ , pages 93--93

    2008

  24. [32]

    , author Aggarwal, C

    author Tang, J. , author Aggarwal, C. , & author Liu, H. ( year 2016 ). title Node classification in signed social networks . In booktitle Proceedings of the 2016 SIAM international conference on data mining \/ (pp. pages 54--62 ). organization SIAM

    2016

  25. [33]

    , author Qu, M

    author Tang, J. , author Qu, M. , & author Mei, Q. ( year 2015 ). title Pte: Predictive text embedding through large-scale heterogeneous text networks . In booktitle Proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining \/ (pp. page...

    2015

  26. [34]

    , author Shazeer, N

    author Vaswani, A. , author Shazeer, N. , author Parmar, N. , author Uszkoreit, J. , author Jones, L. , author Gomez, A. N. , author Kaiser, . , & author Polosukhin, I. ( year 2017 ). title Attention is all you need . journal Advances in neural information processing systems \...

    2017

  27. [35]

    , author Cucurull, G

    author Veli c kovi \' c , P. , author Cucurull, G. , author Casanova, A. , author Romero, A. , author Li \` o , P. , & author Bengio, Y. ( year 2018 ). title Graph Attention Networks . journal The Sixth International Conference on Learning Representations \/ , . note Accepted ...

    2018

  28. [36]

    , author Zhang, P

    author Wang, H. , author Zhang, P. , & author Xing, E. P. ( year 2020 ). title Word shape matters: Robust machine translation with visual embedding . journal CoRR \/ , volume abs/2010.09997 \/

    2020 arXiv

  29. [37]

    , author Souza, A

    author Wu, F. , author Souza, A. , author Zhang, T. , author Fifty, C. , author Yu, T. , & author Weinberger, K. ( year 2019 ). title Simplifying graph convolutional networks . In booktitle Proceedings of the 36th International Conference on Machine Learning \/ (pp. pages 6861...

    2019

  30. [38]

    , author Zhao, W

    author Wu, Q. , author Zhao, W. , author Li, Z. , author Wipf, D. P. , & author Yan, J. ( year 2022 ). title Nodeformer: A scalable graph structure learning transformer for node classification . journal Advances in Neural Information Processing Systems \/ , volume 35 \/ , page...

    2022

  31. [39]

    , author Zhao, W

    author Wu, Q. , author Zhao, W. , author Yang, C. , author Zhang, H. , author Nie, F. , author Jiang, H. , author Bian, Y. , & author Yan, J. ( year 2024 ). title Simplifying and empowering transformers for large-graph representations . journal Advances in Neural Information P...

    2024

  32. [40]

    , author Hu, W

    author Xu, K. , author Hu, W. , author Leskovec, J. , & author Jegelka, S. ( year 2019 ). title How powerful are graph neural networks? In booktitle International Conference on Learning Representations \/

    2019

  33. [41]

    , author Li, C

    author Xu, K. , author Li, C. , author Tian, Y. , author Sonobe, T. , author Kawarabayashi, K.-i. , & author Jegelka, S. ( year 2018 ). title Representation learning on graphs with jumping knowledge networks . In booktitle International conference on machine learning \/ (pp. p...

    2018

  34. [42]

    , author Dai, W

    author Yang, R. , author Dai, W. , author Li, C. , author Zou, J. , & author Xiong, H. ( year 2023 ). title Tackling over-smoothing in graph convolutional networks with em-based joint topology optimization and node classification . journal IEEE Transactions on Signal and Infor...

    2023

  35. [43]

    , author Cai, T

    author Ying, C. , author Cai, T. , author Luo, S. , author Zheng, S. , author Ke, G. , author He, D. , author Shen, Y. , & author Liu, T.-Y. ( year 2021 ). title Do transformers really perform badly for graph representation? journal Advances in neural information processing sy...

    2021

  36. [44]

    , author Dai, W

    author You, C. , author Dai, W. , author Min, Y. , author Liu, F. , author Zhang, X. , author Feng, C. , author Clifton, D. A. , author Zhou, S. K. , author Staib, L. H. , & author Duncan, J. S. ( year 2023 a ). title Rethinking semi-supervised medical image segmentation: A va...

    2023

  37. [45]

    , author Dai, W

    author You, C. , author Dai, W. , author Min, Y. , author Staib, L. H. , & author Duncan, J. S. ( year 2023 b ). title Implicit anatomical rendering for medical image segmentation with stochastic experts . journal Medical image computing and computer-assisted intervention : MI...

    2023

  38. [46]

    , & author Hua, B

    author Yu, D. , & author Hua, B. ( year 2023 ). title Sentiment classification of scientific citation based on modified bert attention by sentiment dictionary . In booktitle EEKE/AII@ JCDL \/ (pp. pages 59--64 )

    2023

  39. [47]

    , author Ren, X

    author Yu, X. , author Ren, X. , author Sun, Y. , author Gu, Q. , author Sturt, B. , author Khandelwal, U. , author Norick, B. , & author Han, J. ( year 2014 ). title Personalized entity recommendation: A heterogeneous information network approach . In booktitle Proceedings of...

    2014

  40. [48]

    , author Yan, Y

    author Zhang, P. , author Yan, Y. , author Li, C. , author Wang, S. , author Xie, X. , & author Kim, S. ( year 2023 ). title Can transformer and gnn help each other? journal CoRR \/ , volume abs/2308.14355 \/ . https://doi.org/10.48550/arXiv.2308.14355

    2023 doi arXiv

  41. [49]

    , author Ma, S

    author Zhao, H. , author Ma, S. , author Zhang, D. , author Deng, Z.-H. , & author Wei, F. ( year 2023 ). title Are more layers beneficial to graph transformers? In booktitle International Conference on Learning Representations \/

    2023

  42. [50]

    , author Yu, C

    author Zhuo, W. , author Yu, C. , & author Tan, G. ( year 2021 ). title Graph neural networks with feature and structure aware random walk . journal ArXiv \/ , volume abs/2111.10102 \/

    2021 arXiv

Pith tools

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