Pith. sign in

REVIEW 3 major objections 7 minor 44 references

Density-aware Walks for Coordinated Campaign Detection

T0 review · 3 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Density-biased random walks, not raw node features, let graph networks recognize coordinated Twitter campaigns, reaching 85.2% binary accuracy.

desk verdict A genuine but narrowly tested contribution: density-weighted walks help on LEN, but the paper never shows density weighting is what causes the gain. read the letter →

arxiv 2506.13912 v1 pith:LEG3DQBC submitted 2025-06-16 cs.SI cs.LG

Atul Anand Gopalakrishnan , Jakir Hossain , Tuğrulcan Elmas , Ahmet Erdem Sar{i}yüce This is my paper
classification cs.SIcs.LG
keywords coordinatedcampaigndetectionrandomweightedwalksgraphclassificationdensity-awareembeddingsmessage-passingneuralnetworksLENdatasetephemeralastroturfingsocialnetworkanalysis
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

The paper argues that coordinated inauthentic campaigns on Twitter leave a detectable footprint: their engagement graphs are denser than organic ones, and that density can be turned into a learning signal. It introduces DECODE, which runs random walks biased by local density measures (degree, core number, truss number), encodes the walked sequences with Skip-gram to get node embeddings, and feeds those embeddings to message-passing neural networks for graph classification. On the LEN dataset of 314 engagement networks around the 2023 Turkish elections, the best configuration reaches 0.852 accuracy and 0.877 F1 for campaign-vs-noncampaign classification, beating the strongest node-feature baseline by about 12 percentage points. The paper's core claim is that density-aware walks are the source of the gain, not merely the random-walk embedding pipeline or the richer node features.

What carries the argument

The central mechanism is the density-biased random weighted walk (RWW). At each step, the next node is sampled from the current node's neighbors with probabilities proportional either to their density (if the current node's density exceeds a threshold $\tau$) or to the inverse (if it falls below $\tau$), so a walk tends to stay within regions of similar local density. Density is measured by one of three metrics, node degree, k-core number, or (for edges, averaged to nodes) k-truss number, and the threshold is set to 0.5, the median density, or the midpoint of the density range. The resulting walks are encoded with Skip-gram into 128-dimensional embeddings, which are concatenated with the dataset's raw node features and fed into an MPNN (GCN, GAT, GIN, or GraphSAGE) followed by mean pooling to produce a graph-level embedding for classification.

What would settle it

Re-run the same MPNN experiments on LEN with embeddings from an unweighted random walk (same walk length, window, and dimension) in place of DECODE's RWW; if the accuracy gap over node features disappears or shrinks to noise, the density bias is not the operative mechanism.

Watch

Extended reading notes

Core claim

The central claim is that local density, expressed through degree, core number, or truss number, is a discriminative signal for coordinated campaign networks, and that injecting it into node embeddings via density-biased random weighted walks lets standard message-passing networks classify campaign graphs far more accurately than when they use raw node features alone. Under the best setup, GraphSAGE on degree-based walks achieves 0.852 ± 0.010 accuracy and 0.877 ± 0.010 F1 on binary campaign detection, surpassing the best previous baseline on LEN by 0.117 and 0.112 respectively. For classifying campaign subtypes, GIN on degree-based walks reaches 0.679 ± 0.001 accuracy, 0.045 above the node-feature baseline, though macro-F1 stays slightly below the baseline because rare subtypes remain hard to distinguish. The authors also find that the choice of threshold separating high- and low-density nodes is model-dependent, and that combining density-aware embeddings with raw node features helps in most multiclass settings.

Load-bearing premise

The load-bearing premise is that the density bias in the random walks, not the random-walk embedding pipeline in general, is what improves classification, and the paper never tests standard unweighted random-walk embeddings under the same setup.

Editorial extensions

If this is right

  • If density-aware embeddings are the operative signal, the method transfers to other settings where coordinated manipulation makes engagement networks denser than organic ones.
  • GraphSAGE with degree-based RWW is the best binary detector on LEN, implying that simple local density suffices without expensive truss computation.
  • Combining the density-aware embeddings with raw node features is usually better than either alone for multiclass campaign-type classification.
  • The model-dependent optimal threshold suggests that a fixed global threshold may be suboptimal, pointing to adaptive or per-graph threshold selection as a direct next step.

Reading between the lines

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

  • The paper does not compare against standard unweighted random-walk embeddings (e.g., DeepWalk or node2vec) under the same MPNN protocol; if those match DECODE's accuracy, the density bias would not be the cause of the gains.
  • The density signal could be tested outside Twitter: any graph-classification task where manipulated networks are denser than organic ones (e.g., coordinated link farms, fraudulent financial networks) is a natural extension.
  • A stronger test of the mechanism would run the same RWW pipeline with a density metric that is not monotonically related to degree, or on datasets where campaign graphs are sparser; failure to hold would mean the method relies on the density contrast in LEN specifically.
  • Since the threshold choice is model-dependent, an automated threshold selection or a learned transition weighting could make the method more robust across models.
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

3 major / 7 minor

Summary. DECODE is a graph classification method that augments MPNNs with density-aware node embeddings generated by biased random walks. The walks bias transitions by degree, core number, or truss number, using a threshold τ that switches between preferring high- or low-density neighbors. Skip-gram embeddings of these walks are concatenated with raw LEN node features and passed to GCN, GAT, GIN, or GraphSAGE. On the LEN dataset, the authors report binary classification accuracy 0.852±0.010 and F1 0.877±0.010 for GraphSAGE with degree-based RWW, and multiclass accuracy 0.679±0.001 for GIN with degree-based RWW, exceeding raw-node-feature baselines from [16].

Significance. The paper addresses a real and timely problem, and the proposed encoding of local density into random-walk features is simple and potentially useful. The code is publicly available, and the density statistics in Table 2 provide sensible motivation for the approach. If the reported gains are truly attributable to density-aware transition probabilities, this would be a practical contribution to coordinated campaign detection. However, the missing unweighted-random-walk baseline and the opaque τ selection protocol currently prevent the reader from attributing the gains to the proposed mechanism. The contribution is therefore promising but not yet established.

major comments (3)
  1. [§5.1, Tables 3–6] The experimental design does not include an unweighted random-walk baseline. The DECODE pipeline introduces two ingredients at once: (i) random-walk/Skip-gram structural features and (ii) density-biased transition probabilities. The paper's stated contribution is specifically ingredient (ii) (Sections 1 and 4), but all reported comparisons are against MPNNs trained on raw LEN node features. A control using DeepWalk or node2vec embeddings with the same walk length (100), window (4), embedding dimension (128), MPNN architectures, and hyperparameter tuning is needed to isolate the density-weighting mechanism. Without it, the 0.852/0.877 binary result is equally consistent with any random-walk embedding being beneficial on LEN.
  2. [§5.1, Tables 3–6] The threshold-selection protocol for τ is not described as validation-based. The text lists three candidate values (0.5, median, midpoint) but each table cell reports a single best τ per model/density pair, and no statement explains whether this selection used a held-out validation split or the test set. If the latter, the reported accuracies and F1 scores are optimistically biased by selection on the test set. Please specify the split protocol, report performance for all three τ values (or a validation-selected τ) with standard deviations, and state how many seeds/folds were used.
  3. [§5.1 and §5.2] It is unclear whether the 'NF' baseline numbers in Tables 3–6 were re-trained in this paper or copied from [16]. The text says 'we compare our method against four baselines' and also says 'surpassing the best baseline in [16]'. If the NF rows are taken from [16], differences in implementation, seeds, or train/validation splits could confound the comparison. If they were re-trained, please say so explicitly and describe the protocol (same folds, same hardware, same tuning). This is needed to validate the headline improvement of 0.117 in accuracy and 0.112 in F1.
minor comments (7)
  1. [Algorithm 1] 'Top(w)' is not defined; it presumably refers to the last node in the walk. Please define it.
  2. [Algorithm 1] The loop 'for t = 1 to L' plus the initial node produces a walk of length L+1; state whether L is the number of transitions or the total walk length.
  3. [Tables 5 and 6] Table 5 (GIN Truss row) and Table 6 (GAT NF+RWW rows) list 'mean' as a threshold, but §5.1 defines only 0.5, median, and midpoint; this appears to be a typo.
  4. [Section 1 and Section 4] The text contains duplicated words: 'using the the density-aware embeddings' in the contributions and 'embeddings embeddings' in Section 4. Please fix these typos.
  5. [Section 3.1] The notation 'where V is the number of vertices' should read 'where V is the set of vertices'.
  6. [Section 5.1] The paper does not state the evaluation protocol (number of folds/seeds, split proportions). Please provide these details or refer explicitly to the protocol in [16].
  7. [Conclusion] The conclusion mentions 'betweeness centrality'; this should be 'betweenness centrality'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: DECODE's density-weighted walk embeddings are evaluated empirically against independent baselines; missing DeepWalk/node2vec controls are experimental gaps, not definitional reductions.

full rationale

The paper's claimed derivation chain is empirical rather than formal: it constructs RWW embeddings from degree/core/truss densities (Algorithm 1), encodes them with Skip-gram, and trains MPNNs on LEN. The density-aware embeddings are not defined in terms of the classification labels, and the RWW transition weights depend only on the density function and threshold, not on y. Thus the headline result (GraphSAGE + degree RWW, accuracy 0.852, F1 0.877) is an empirical comparison, not an identity with the input statistics. The self-citations to [16] provide the LEN dataset and baseline numbers, and [10] provides the ephemeral astroturfing concept; these are external artifacts, not assumed theorems that force the conclusion. The main evaluation gap—no DeepWalk/node2vec control under the same MPNN protocol—means the specific contribution of density-bias over generic random-walk embeddings is not isolated, and the best-of-three threshold reporting without a described validation split risks selection bias. These are experimental-completeness and statistical concerns, not circular reductions by construction. No load-bearing step reduces to its own input by definition or by self-citation, so no circularity is found.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The method rests on the domain belief that campaign networks are denser, on the transfer of Skip-gram random-walk embeddings to graphs, and on the reliability of the LEN ground truth. The main hand-chosen parameter is the density threshold, with several MPNN hyperparameters tuned empirically. No new entities or forces are introduced.

free parameters (6)
  • Density threshold tau = Selected per model from {0.5, median, midpoint}
    Controls whether the random walk prefers higher- or lower-density neighbors. The best value varies by model and density metric, and it is chosen based on empirical performance.
  • Walk length L = 100
    Fixed walking length used to generate random weighted walks before Skip-gram encoding.
  • Skip-gram embedding size = 128
    Dimensionality of the node embeddings produced by Skip-gram.
  • Skip-gram window = 4
    Number of neighboring nodes in the random walk used to encode each node.
  • MPNN hidden size = Best of {128, 256, 512, 1024}
    Tuned as in the prior LEN paper [16], with no explicit validation procedure described.
  • MPNN learning rate = Best of {0.001, 0.0001, 0.00001}
    Tuned as in the prior LEN paper [16], with no explicit validation procedure described.
assumptions (4)
  • domain assumption Campaign graphs in LEN are denser than non-campaign graphs.
    This is the main motivation for density-aware walks. Table 2 shows higher mean degree, core number, and truss number for campaign graphs overall, but this may be a dataset-specific property.
  • domain assumption Skip-gram on random walks produces structurally meaningful node embeddings.
    Borrowed directly from DeepWalk and node2vec; the paper does not re-derive or test this assumption.
  • domain assumption The LEN ground-truth labels are correct.
    The ground truth relies on ephemeral astroturfing detection from the authors' prior work [16], and is not independently verified in this paper.
  • domain assumption The directed engagement graph can be treated as undirected for density computation.
    The paper states it ignores edge directions for k-core and k-truss definitions, which discards direction information that may be relevant to engagement semantics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Density-aware Walks for Coordinated Campaign Detection." pith.science (2026). https://pith.science/paper/LEG3DQBC

@misc{pith2026250613912,
  author       = {Pith},
  title        = {Pith review of: Density-aware Walks for Coordinated Campaign Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LEG3DQBC}},
  note         = {Machine review of arXiv:2506.13912}
}
read the original abstract

Coordinated campaigns frequently exploit social media platforms by artificially amplifying topics, making inauthentic trends appear organic, and misleading users into engagement. Distinguishing these coordinated efforts from genuine public discourse remains a significant challenge due to the sophisticated nature of such attacks. Our work focuses on detecting coordinated campaigns by modeling the problem as a graph classification task. We leverage the recently introduced Large Engagement Networks (LEN) dataset, which contains over 300 networks capturing engagement patterns from both fake and authentic trends on Twitter prior to the 2023 Turkish elections. The graphs in LEN were constructed by collecting interactions related to campaigns that stemmed from ephemeral astroturfing. Established graph neural networks (GNNs) struggle to accurately classify campaign graphs, highlighting the challenges posed by LEN due to the large size of its networks. To address this, we introduce a new graph classification method that leverages the density of local network structures. We propose a random weighted walk (RWW) approach in which node transitions are biased by local density measures such as degree, core number, or truss number. These RWWs are encoded using the Skip-gram model, producing density-aware structural embeddings for the nodes. Training message-passing neural networks (MPNNs) on these density-aware embeddings yields superior results compared to the simpler node features available in the dataset, with nearly a 12\% and 5\% improvement in accuracy for binary and multiclass classification, respectively. Our findings demonstrate that incorporating density-aware structural encoding with MPNNs provides a robust framework for identifying coordinated inauthentic behavior on social media networks such as Twitter.

Figures

Figures reproduced from arXiv: 2506.13912 by the authors.

Figure 1
Figure 1. An overview of DECODE. Density-based random weighted walk captures [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Receiver Operating Characteristic (ROC) curves for campaign vs. non [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Confusion matrices for multi-class classification. The best performing con [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 35 canonical work pages

  1. [16]

    Gopalakrishnan, A.A., Hossain, J., Elmas, T., Sariyuce, A.E.: Large engagement networks for classifying coordinated campaigns and organic twitter trends (2025), https://api.semanticscholar.org/CorpusID:276742534

    work page 2025

  2. [1]

    handle.net/10477/79221

    Center for computational research, university at buffalo (2025), http://hdl. handle.net/10477/79221. Accessed: 2025-04-14

    work page 2025

  3. [2]

    arXiv preprint cs/0310049 (2003)

    Batagelj, V., Zaversnik, M.: An o (m) algorithm for cores decomposition of net- works. arXiv preprint cs/0310049 (2003)

    arXiv 2003

  4. [3]

    IEEE Transactions on Pattern Analysis and Machine Intelligence45(1), 657–668 (2022)

    Bouritsas, G., Frasca, F., Zafeiriou, S., Bronstein, M.M.: Improving graph neural network expressivity via subgraph isomorphism counting. IEEE Transactions on Pattern Analysis and Machine Intelligence45(1), 657–668 (2022)

    work page 2022

  5. [4]

    In: International conference on machine learning

    Chen, D., O’Bray, L., Borgwardt, K.: Structure-aware transformer for graph rep- resentation learning. In: International conference on machine learning. pp. 3469–

  6. [5]

    National secu- rity agency technical report16(3.1), 1–29 (2008)

    Cohen, J.: Trusses: Cohesive subgraphs for social network analysis. National secu- rity agency technical report16(3.1), 1–29 (2008)

    work page 2008

  7. [6]

    arXiv preprint arXiv:2012.09699 (2020)

    Dwivedi, V.P., Bresson, X.: A generalization of transformer networks to graphs. arXiv preprint arXiv:2012.09699 (2020)

    arXiv 2020

  8. [7]

    Advances in Neural Information Processing Sys- tems 35, 22326–22340 (2022)

    Dwivedi, V.P., Rampášek, L., Galkin, M., Parviz, A., Wolf, G., Luu, A.T., Beaini, D.: Long range graph benchmark. Advances in Neural Information Processing Sys- tems 35, 22326–22340 (2022)

    work page 2022

Show all 44 references
  1. [8]

    In: Companion Proceedings of the ACM Web Conference 2023

    Elmas, T.: Analyzing activity and suspension patterns of twitter bots attacking turkish twitter trends by a longitudinal dataset. In: Companion Proceedings of the ACM Web Conference 2023. pp. 1404–1412 (2023)

    2023

  2. [9]

    In: Proceedings of the International AAAI Conference on Web and Social Media

    Elmas, T., Overdorf, R., Aberer, K.: Characterizing retweet bots: The case of black market accounts. In: Proceedings of the International AAAI Conference on Web and Social Media. vol. 16, pp. 171–182 (2022)

    2022

  3. [10]

    In: 2021 IEEE European symposium on security and privacy (EuroS&P)

    Elmas,T.,Overdorf,R.,Özkalay,A.F.,Aberer,K.:Ephemeralastroturfingattacks: The case of fake twitter trends. In: 2021 IEEE European symposium on security and privacy (EuroS&P). pp. 403–422. IEEE (2021)

    2021

  4. [11]

    In: Proceedings of the International AAAI Confer- ence on Web and Social Media

    Elmas, T., Randl, M., Attia, Y.: # teamfollowback: Detection & analysis of follow back accounts on social media. In: Proceedings of the International AAAI Confer- ence on Web and Social Media. vol. 18, pp. 381–393 (2024)

    2024

  5. [12]

    Computational and Mathematical Organization Theory30(2), 161–172 (2024) Density-aware Walks for Coordinated Campaign Detection 15

    Erhardt, K., Pentland, A.: Hidden messages: mapping nations’ media campaigns. Computational and Mathematical Organization Theory30(2), 161–172 (2024) Density-aware Walks for Coordinated Campaign Detection 15

    2024

  6. [13]

    Trans- actions on Machine Learning Research

    Feldman, O., Boyarski, A., Feldman, S., Kogan, D., Mendelson, A., Baskin, C.: Weisfeiler and leman go infinite: Spectral and combinatorial pre-colorings. Trans- actions on Machine Learning Research

  7. [14]

    Social Net- works 1(3), 215–239 (1978)

    Freeman, L.C.: Centrality in social networks conceptual clarification. Social Net- works 1(3), 215–239 (1978)

    1978

  8. [15]

    Scientific Reports13(1), 22571 (2023)

    Gabriel, N.A., Broniatowski, D.A., Johnson, N.F.: Inductive detection of influence operations via graph learning. Scientific Reports13(1), 22571 (2023)

    2023

  9. [17]

    In: Pro- ceedings of the 22nd ACM SIGKDD international conference on Knowledge dis- covery and data mining

    Grover, A., Leskovec, J.: node2vec: Scalable feature learning for networks. In: Pro- ceedings of the 22nd ACM SIGKDD international conference on Knowledge dis- covery and data mining. pp. 855–864 (2016)

    2016

  10. [18]

    Advances in neural information processing systems30 (2017)

    Hamilton, W., Ying, Z., Leskovec, J.: Inductive representation learning on large graphs. Advances in neural information processing systems30 (2017)

    2017

  11. [19]

    Computational and Mathematical Organization Theory30(2), 127–147 (2024)

    Jacobs, C.S., Carley, K.M.: # whatisdemocracy: finding key actors in a chinese influence campaign. Computational and Mathematical Organization Theory30(2), 127–147 (2024)

    2024

  12. [20]

    Proceedings of the ACM on Human-computer Interaction5(CSCW2), 1–19 (2021)

    Jakesch, M., Garimella, K., Eckles, D., Naaman, M.: Trend alert: A cross-platform organization manipulated twitter trends in the indian general election. Proceedings of the ACM on Human-computer Interaction5(CSCW2), 1–19 (2021)

    2021

  13. [21]

    arXiv preprint arXiv:2109.14872 (2021)

    Kausar, S., Tahir, B., Mehmood, M.A.: Towards understanding trends manipula- tion in pakistan twitter. arXiv preprint arXiv:2109.14872 (2021)

    2021 arXiv

  14. [22]

    In: International Conference on Learning Representations (2017)

    Kipf, T.N., Welling, M.: Semi-supervised classification with graph convolutional networks. In: International Conference on Learning Representations (2017)

    2017

  15. [23]

    Advances in Neural Information Pro- cessing Systems 34, 21618–21629 (2021)

    Kreuzer, D., Beaini, D., Hamilton, W., Létourneau, V., Tossou, P.: Rethinking graph transformers with spectral attention. Advances in Neural Information Pro- cessing Systems 34, 21618–21629 (2021)

    2021

  16. [24]

    Advances in Neural Information Processing Systems33, 4465–4478 (2020)

    Li, P., Wang, Y., Wang, H., Leskovec, J.: Distance encoding: Design provably more powerful neural networks for graph representation learning. Advances in Neural Information Processing Systems33, 4465–4478 (2020)

    2020

  17. [25]

    In: Proceedings of the ACM Web Conference 2024

    Luceri, L., Pantè, V., Burghardt, K., Ferrara, E.: Unmasking the web of deceit: Uncovering coordinated activity to expose information operations on twitter. In: Proceedings of the ACM Web Conference 2024. pp. 2530–2541 (2024)

    2024

  18. [26]

    arXiv preprint arXiv:1301.3781 (2013)

    Mikolov, T., Chen, K., Corrado, G., Dean, J.: Efficient estimation of word repre- sentations in vector space. arXiv preprint arXiv:1301.3781 (2013)

    2013 arXiv

  19. [27]

    Ong, J.C., Cabañes, J.V.: Architects of networked disinformation: Behind the scenes of troll accounts and fake news production in the philippines (2018)

    2018

  20. [28]

    In: Companion proceedings of the web conference

    Pacheco, D., Flammini, A., Menczer, F.: Unveiling coordinated groups behind white helmets disinformation. In: Companion proceedings of the web conference

  21. [29]

    NATO Strategic Communication Centre of Excellence (2022)

    Pamment, J., Smith, V.: Attributing information influence operations: Identifying those responsible for malicious behaviour online. NATO Strategic Communication Centre of Excellence (2022)

    2022

  22. [30]

    In: Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining

    Perozzi, B., Al-Rfou, R., Skiena, S.: Deepwalk: Online learning of social represen- tations. In: Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining. pp. 701–710 (2014)

    2014

  23. [31]

    In: Proceedings of the In- ternational AAAI Conference on Web and Social Media

    Pote, M., Elmas, T., Flammini, A., Menczer, F.: Coordinated reply attacks in influence operations: Characterization and detection. In: Proceedings of the In- ternational AAAI Conference on Web and Social Media. vol. 19, pp. 1586–1598 (2025) 16 A.A. Gopalakrishnan et al

    2025

  24. [32]

    Rampášek, L., Galkin, M., Dwivedi, V.P., Luu, A.T., Wolf, G., Beaini, D.: Recipe forageneral,powerful,scalablegraphtransformer.AdvancesinNeuralInformation Processing Systems 35, 14501–14515 (2022)

    2022

  25. [33]

    In: Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining

    Ribeiro, L.F., Saverese, P.H., Figueiredo, D.R.: struc2vec: Learning node repre- sentations from structural identity. In: Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining. pp. 385–394 (2017)

    2017

  26. [34]

    ACM Transactions on the Web (TWEB) 11(3), 1–27 (2017)

    Sariyüce, A.E., Seshadhri, C., Pinar, A., Çatalyürek, Ü.V.: Nucleus decomposi- tions for identifying hierarchy of dense subgraphs. ACM Transactions on the Web (TWEB) 11(3), 1–27 (2017)

    2017

  27. [35]

    Social networks5(3), 269– 287 (1983)

    Seidman, S.B.: Network structure and minimum degree. Social networks5(3), 269– 287 (1983)

    1983

  28. [36]

    In: International Conference on Learning Representations (2018)

    Veličković, P., Cucurull, G., Casanova, A., Romero, A., Liò, P., Bengio, Y.: Graph attention networks. In: International Conference on Learning Representations (2018)

    2018

  29. [37]

    In: Proceedings of the international AAAI conference on web and social media

    Vishnuprasad, P.S., Nogara, G., Cardoso, F., Cresci, S., Giordano, S., Luceri, L.: Tracking fringe and coordinated activity on twitter leading up to the us capitol attack. In: Proceedings of the international AAAI conference on web and social media. vol. 18, pp. 1557–1570 (2024)

    2024

  30. [38]

    In: Proceedings of the AAAI conference on artificial intelligence

    Wang, X., Cui, P., Wang, J., Pei, J., Zhu, W., Yang, S.: Community preserving net- work embedding. In: Proceedings of the AAAI conference on artificial intelligence. vol. 31 (2017)

    2017

  31. [39]

    Social Network Analysis and Mining11(1), 111 (2021)

    Weber, D., Neumann, F.: Amplifying influence through coordinated behaviour in social networks. Social Network Analysis and Mining11(1), 111 (2021)

    2021

  32. [40]

    Xu, K., Hu, W., Leskovec, J., Jegelka, S.: How powerful are graph neural networks? In: International Conference on Learning Representations

  33. [41]

    Ying, C., Cai, T., Luo, S., Zheng, S., Ke, G., He, D., Shen, Y., Liu, T.Y.: Do transformers really perform badly for graph representation? Advances in neural information processing systems34, 28877–28888 (2021)

    2021

  34. [42]

    In: Companion proceedings of the 2019 world wide web conference

    Zannettou,S.,Caulfield,T.,DeCristofaro,E.,Sirivianos,M.,Stringhini,G.,Black- burn, J.: Disinformation warfare: Understanding state-sponsored trolls on twitter and their influence on the web. In: Companion proceedings of the 2019 world wide web conference. pp. 218–226 (2019)

    2019

  35. [43]

    In: ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

    Zeng, D., Chen, W., Liu, W., Zhou, L., Qu, H.: Rethinking random walk in graph representation learning. In: ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). pp. 1–5. IEEE (2023)

    2023

  36. [44]

    Advances in Neural Information Processing Systems 36, 16172–16206 (2023)

    Zhou, C., Wang, X., Zhang, M.: Facilitating graph neural networks with random walk on simplicial complexes. Advances in Neural Information Processing Systems 36, 16172–16206 (2023)

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.