Pith. sign in

REVIEW 4 major objections 5 minor 51 references

Towards Propagation-aware Representation Learning for Supervised Social Media Graph Analytics

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a single representation-learning architecture can serve rumor detection, social-bot detection, and information-diffusion prediction at state-of-the-art level, provided its encoder is made propagation-aware through a Ma

desk verdict Useful unified backbone with strong experiments, but the central kinetic loss has a load-bearing indexing error that makes its physical claim unverified as written. read the letter →

arxiv 2509.01124 v1 pith:Y3MQDVNK submitted 2025-09-01 cs.SI

Wei Jiang , Tong Chen , Wei Yuan , Xiangyu Zhao , Quoc Viet Hung Nguyen , Hongzhi Yin This is my paper · ORCID
classification cs.SI
keywords informationpropagationgraphkinetic-guidedlossMarkovchaintransmissionmodelsocialmediaanalyticsrumordetectionbotdiffusionpredictionzero-shottransfer
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

RPRL tries to show that one representation-learning architecture can serve rumor detection, social-bot detection, and information diffusion prediction, provided it is guided by a principled model of how information spreads. The paper abstracts all three tasks into an information propagation graph where nodes move from an unknown state to positive or negative states, then derives a propagation-aware encoder and a kinetic-guided loss from a Markov chain transmission model. On six datasets, the same backbone achieves state-of-the-art accuracy, balanced accuracy, Hit@K, and MAP@K across the three tasks, and outperforms baselines under zero-shot and one/five-shot transfer. The broader point is that a physics-style inductive bias—information only flows along edges from informed to uninformed nodes—can act as a shared, noise-robust supervisory signal for social graph learning.

What carries the argument

The load-bearing object is the information propagation graph G_IP = (V,E,S), S={U,I1,I2}, plus the Markov chain kinetic model adapted from [32]. The propagation-aware encoder applies a hop mask m_j^{(t)} = 1{d_j <= t} that zeros out embeddings beyond hop t, then passes masked embeddings through shared transformer blocks to output state probabilities per time step. The kinetic-guided loss differentiates those probabilities with unit time per hop and minimizes the residual against the ODE dU/dt = -U Σ_k β_k Σ_j a_vj I_k,v(t), dI_k,v/dt = U β_k Σ_j a_vj I_k,v(t). This injects the prior that information travels only along edges, from informed to uninformed nodes, as label-free physical supervisi

What would settle it

Train RPRL on a social graph with ground-truth adoption timestamps and compare the model's predicted hop-indexed state probabilities against the empirical fractions of nodes in U, I1, I2 at each hop. If the kinetic loss is truly propagation-aware, the two should track each other; large systematic deviations with high task accuracy would show the loss acts only as a regularizer. A second test: randomly permute edge directions while preserving hop distances from the ego node—if performance barely changes, the 'propagation' prior is not reading edge direction.

Watch

Extended reading notes

Core claim

RPRL is a single backbone for rumor detection, social-bot detection, and diffusion prediction. The paper's claim: all three tasks instantiate one object—the information propagation graph, where nodes start in an unknown state U and flip to one of two informative states I1/I2 as information spreads from an ego node. Two data-driven encoders (structure-agnostic context, graph) run in parallel with a propagation-aware encoder that outputs hop-indexed state probabilities. A kinetic-guided loss from a Markov chain transmission model forces these probabilities to track an infection-style ODE: unknown nodes become informed only through informed neighbors. No labeled states are needed; this physics-

Load-bearing premise

That the model's predicted softmax probabilities can be read as the physical state variables U_v(t) and I_k,v(t) in a Markov chain ODE, and that one hop of graph distance equals one unit of propagation time; if that identification is wrong, the kinetic loss degrades into a generic regularizer.

Editorial extensions

If this is right

  • One architecture with task-specific heads and a shared kinetic loss replaces three separate model families; the paper reports state-of-the-art accuracy, balanced accuracy, F1, Hit@K, and MAP@K on all six datasets.
  • Cross-dataset pre-training works: pre-training on one dataset and fine-tuning on another improves performance, while applying the same schedule to baselines often hurts them, suggesting the propagation-aware representation transfers better.
  • The kinetic loss is the source of the gain, not the architecture alone: removing the propagation-aware encoder and kinetic loss (RPRL w/o pe) consistently drops performance, and replacing the Markov chain model with a regular kinetic model (RPRL-rg) underperforms.
  • Zero-shot transfer is strong in rumor and bot detection, where RPRL without any target labels beats or approaches baselines given one or five shots; in diffusion prediction, one shot is enough to overtake all baselines.

Reading between the lines

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

  • If the softmax-as-state-probability identification is valid, the kinetic loss could be used as a self-supervised pretext task on unlabeled propagation graphs, making the framework a candidate pretraining objective for social graph foundation models.
  • The hop-distance-as-time assumption ties the method to BFS-like propagation; on graphs with long-range edges or non-tree cascades, the mask m_j^{(t)} may desynchronize from true adoption times, so a time-stamped variant of the mask would be a natural stress test.
  • The paper's ODE, as printed, sums over neighbors j while indexing the node's own state I_k,v; replacing I_k,v with I_k,j yields the standard neighbor-driven susceptible-infectious model, and the corrected form is what the physical interpretation requires. This distinction is worth checking in the released code.
  • Because the kinetic model generalizes to more than two informative states or continuous-valued states, the same loss could extend to multi-class or intensity-based propagation, though the paper only evaluates binary states.
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 RPRL, a unified architecture for supervised social media graph analytics covering graph classification, node classification, and link prediction. The framework combines a data-driven dual encoder (structure-agnostic context encoder plus graph encoder) with a propagation-aware encoder that applies hop-distance-based masking, and a kinetic-guided loss Lp supposed to be derived from a Markov chain-based transmission model. The final objective is L = Ls + λLp with task-specific heads. Experiments on six datasets compare RPRL with task-specific baselines, including cross-dataset pre-training and zero-/few-shot transfer, and report state-of-the-art results across all three tasks.

Significance. If the kinetic derivation were correct, the paper would offer a generally useful inductive bias and a practical unified architecture for social media graph analytics. The released code, the breadth of tasks, and the empirical gains—especially in zero-/few-shot transfer—are commendable. However, the central physical-grounding claim is not currently supported: the printed ODE and loss do not encode the neighbor-driven transmission mechanism attributed to the cited kinetic models, and the identification of classifier probabilities with physical state variables is asserted rather than derived. The noise-robustness claim is also not tested directly. The empirical core may still be salvageable, but the paper's main novelty needs substantial repair or reframing.

major comments (4)
  1. [Section IV-B, Eqs. (8) and (10)] The printed kinetic model sums over neighbors j but evaluates the infectious state at the node v itself: the RHS of Eq. (8) contains Σ_j a_{v,j} I_{k,v}(t), which equals deg(v)·I_{k,v}(t). This is a degree-scaled self-transition, not the neighbor-driven transmission term used in the cited contact-process models, which require Σ_j a_{v,j} I_{k,j}(t). The same indexing error appears in the kinetic-guided loss Eq. (10), where the residual uses Î_{k,i}(t) inside the adjacency sum. As written, Lp does not encode the physical mechanism claimed. If the implementation corrects the index to I_{k,j}(t), an erratum is needed; if it uses the printed form, the ablation RPRL vs. RPRL-rg does not test the claimed Markov-chain guidance.
  2. [Section IV-B, Eqs. (8)–(11)] The derivation assumes without justification that the softmax probabilities emitted by the propagation-aware encoder can be treated as the physical state variables U_v(t) and I_{k,v}(t), and that hop distance equals continuous propagation time. Since β_k is also learned, the kinetic-guided loss is, as it stands, a self-consistency regularizer: the model's own probabilities appear on both sides of the residual, with a learned coefficient. This may still be a useful inductive bias, but the 'first-principle' framing and the physical interpretation of Lp are not established. The authors should either provide a derivation connecting the encoder's output distribution to the state variables of the kinetic model, or explicitly reframe Lp as a temporal-consistency regularizer.
  3. [Abstract and Section V] The paper repeatedly claims robustness to noisy social media data, but no experiment injects noise into node features, graph edges, or labels. The cross-dataset pre-training results show transferability, not noise robustness. To support the abstract's claim and the introduction's assertion that the kinetic model 'addresses' vulnerability to data noise, the authors should add noise-injection experiments (e.g., random edge dropout, feature perturbation, or label noise) or temper the claim.
  4. [Section V-D, Table IV] The ablation narrative states that removing the propagation-aware embedding module leads to a 'consistent and notable performance drop,' but Table IV shows RPRL w/o pe has slightly higher Accuracy than RPRL on DRWeibo (89.58 vs 89.53), with overlapping standard deviations. Moreover, RPRL w/o pe removes both the propagation-aware encoder and the kinetic-guided loss, so it cannot isolate the contribution of Lp. The RPRL-rg variant is also never defined (what is the 'regular kinetic model'?). The ablation needs to specify these variants and ideally include a version that keeps the propagation-aware encoder but removes only Lp, or vice versa.
minor comments (5)
  1. [Section IV-B, Eq. (7)] Notation is inconsistent: the state set is written S = {U, l1, l2} but elsewhere the informative states are I1 and I2. Also, the argument of softmax uses h_out(t)_i without defining h_out; presumably it is z(t)_i or a projection of it.
  2. [Section V-C, Figure 3] Zero-shot and few-shot results are presented only in figures without numerical tables. Given that strong transferability is a headline claim, numerical values with standard deviations should be provided.
  3. [Section V-E, Figure 4] Hyperparameter sensitivity curves are shown without error bars or repeated-run variance; it is hard to judge whether the observed trends are significant.
  4. [General] The code is released, which is appreciated, but the paper should state the exact random seeds, number of runs, and hardware compute time for reproducibility. Currently only 'NVIDIA RTX 4090' is mentioned.
  5. [General] Minor typos: 'Represenation' in the Index Terms; 'Pr ob' in Figure 2; 'the objective of this task is to use G and the labels' in Section III-A.

Circularity Check

2 steps flagged · score 6.0 of 10

The kinetic-guided loss is a self-consistency regularizer: Eq. (8) indexes the transmission term by the node's own state in a neighbor sum, so the 'Markov-chain physics' reduces to a degree-scaled self-transition; Eq. (10) then fits the model's own softmax outputs to that self-transition with a learned beta.

  1. other [Section IV-B, Eq. (8) (Markov Chain-based Kinetic Model)]
    "dIk,v(t)/dt = Uv(t)βk Σ_{j∈N} a_{v,j} Ik,v(t)"

    As printed, the infectious state in the transmission term is I_{k,v}(t), which does not depend on the summation index j. Therefore Σ_{j∈N} a_{v,j} I_{k,v}(t) = deg(v)·I_{k,v}(t): a degree-scaled self-transition, not the neighbor-driven transmission Σ_j a_{v,j} I_{k,j}(t) of the contact-process models cited in [31]–[33]. The propagation 'physics' is thus defined in terms of the node's own state probability, so the model cannot be said to derive from the cited Markov-chain transmission mechanism.

  2. fitted input called prediction [Section IV-B, Eq. (10)-(11) (Kinetic-guided Loss Function)]
    "Given these predictions of probabilities, we then compute the kinetic-guided loss by minimizing the residuals defined by the kinetic model: ... where β̂_k can be learned by a linear layer with propagation-aware embedding zout(t)_i."

    The loss is the squared residual between the finite-difference of the softmax outputs and a function of those same softmax outputs at time t, with β̂_k learned rather than fixed by the physical model. Since no observed states or external constants enter the residual, optimizing Lp can drive it near zero by adapting β̂_k to the model's own predictions. The 'agreement with the Markov-chain ODE' is therefore enforced by construction, and the claimed physical supervision is a self-consistency regularizer rather than an independent first-principles prediction.

full rationale

The empirical core of the paper is not circular: all three tasks are evaluated against external baselines, the graph encoder follows [39] and the kinetic model is adapted from the external source [32], and no load-bearing self-citation or uniqueness theorem is used. The circularity concern is internal to the derivation of the central novelty. Eq. (8) as printed has Σ_{j∈N} a_{v,j} I_{k,v}(t), which equals deg(v)·I_{k,v}(t); the cited neighbor-driven kinetic models require I_{k,j}(t). Eq. (10) then uses the same softmax probabilities on both sides of the residual with a learned β̂, so the kinetic-guided loss reduces to a self-consistency constraint. This undermines the claim that the objective is 'derived from a Markov chain-based transmission model' as written. A corrected implementation (neighbor index I_{k,j} and independently fixed β) could repair the derivation; as printed, the physical grounding is not established. Because the SOTA results are externally benchmarked, the score is 6 rather than 8-10.

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

The kinetic model, the hop-time equivalence, and the identification of predicted probabilities with physical states are all assumed without independent support. beta, gamma, and lambda are fitted or tuned, and the invented state variables have no external validation.

free parameters (3)
  • beta_k (transition coefficients) = learned via a linear layer from propagation-aware embeddings (Section IV-B)
    The kinetic loss's infection rates are not specified by the physical model; they are learned from data, so the 'physics' is partially fitted.
  • gamma (fusion weight) = tuned per task, e.g., 0.5 on DRWeibo, 0.7 on MGTAB (Section V-E)
    Controls graph versus context encoder balance; chosen by validation performance, not derived.
  • lambda (kinetic loss weight) = tuned per task, peaks vary, e.g., 0.5 on DRWeibo and MGTAB, higher on Christianity (Section V-E)
    Controls how much physics constraint is applied; selected by validation performance.
assumptions (4)
  • domain assumption The Markov chain kinetic model in Eq (8) describes real information propagation in social media graphs.
    Section IV-B adopts the model from [32] and applies it to rumor, bot, and cascade data; no validation that these tasks obey the ODE.
  • domain assumption Hop distance from the ego node equals propagation time: nodes at hop i activate at time step t_i.
    Defined in Section III-B and Algorithm 1; ties discrete time to graph distance without evidence.
  • ad hoc to paper Softmax probabilities from the propagation-aware encoder can be treated as the state variables U and I_k in the kinetic ODE.
    Equations (9) and (10) substitute predicted probabilities into the ODE; this identification is assumed, with no labeled states or calibration to justify it.
  • standard math The forward difference quotient with Delta t = 1 approximates the ODE time derivative.
    Equation (11); standard numerical approximation, but coarse for integer hop steps.
invented entities (1)
  • Information propagation graph states U, I1, I2
    purpose: Unify the three tasks under a common state transition story and provide the state probabilities used in the kinetic loss.
    U, I1, and I2 are a modeling abstraction; the paper gives no outside falsifiable handle such as observable state annotations for them.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Propagation-aware Representation Learning for Supervised Social Media Graph Analytics." pith.science (2026). https://pith.science/paper/Y3MQDVNK

@misc{pith2026250901124,
  author       = {Pith},
  title        = {Pith review of: Towards Propagation-aware Representation Learning for Supervised Social Media Graph Analytics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y3MQDVNK}},
  note         = {Machine review of arXiv:2509.01124}
}
read the original abstract

Social media platforms generate vast, complex graph-structured data, facilitating diverse tasks such as rumor detection, bot identification, and influence modeling. Real-world applications like public opinion monitoring and stock trading -- which have a strong attachment to social media -- demand models that are performant across diverse tasks and datasets. However, most existing solutions are purely data-driven, exhibiting vulnerability to the inherent noise within social media data. Moreover, the reliance on task-specific model design challenges efficient reuse of the same model architecture on different tasks, incurring repetitive engineering efforts. To address these challenges in social media graph analytics, we propose a general representation learning framework that integrates a dual-encoder structure with a kinetic-guided propagation module. In addition to jointly modeling structural and contextual information with two encoders, our framework innovatively captures the information propagation dynamics within social media graphs by integrating principled kinetic knowledge. By deriving a propagation-aware encoder and corresponding optimization objective from a Markov chain-based transmission model, the representation learning pipeline receives a boost in its robustness to noisy data and versatility in diverse tasks. Extensive experiments verify that our approach achieves state-of-the-art performance with a unified architecture on a variety of social media graph mining tasks spanning graph classification, node classification, and link prediction. Besides, our solution exhibits strong zero-shot and few-shot transferability across datasets, demonstrating practicality when handling data-scarce tasks.

Figures

Figures reproduced from arXiv: 2509.01124 by the authors.

Figure 1
Figure 1. Information Propagation Graph. This graph is an [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Overview of Robust Propagation-aware Representation Learning Framework [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Performance of RPRL and baselines across three tasks under different zero- and few-shot settings. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Hyperparameter sensitivity analysis: The influences of [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 48 canonical work pages

  1. [1]

    What is twitter, a social network or a news media?

    H. Kwak, C. Lee, H. Park, and S. Moon, “What is twitter, a social network or a news media?” in Proceedings of the 19th international conference on World wide web , 2010, pp. 591–600

    work page 2010

  2. [2]

    Comparision of microblogging service between sina weibo and twitter,

    S. Chen, H. Zhang, M. Lin, and S. Lv, “Comparision of microblogging service between sina weibo and twitter,” in Proceedings of 2011 In- ternational Conference on Computer Science and Network Technology , vol. 4. IEEE, 2011, pp. 2259–2263

    work page 2011

  3. [3]

    Challenging low homophily in social recommendation,

    W. Jiang, X. Gao, G. Xu, T. Chen, and H. Yin, “Challenging low homophily in social recommendation,” in Proceedings of the ACM Web Conference 2024, 2024, pp. 3476–3484

    work page 2024

  4. [4]

    Social influence-based group representation learning for group recommenda- tion,

    H. Yin, Q. Wang, K. Zheng, Z. Li, J. Yang, and X. Zhou, “Social influence-based group representation learning for group recommenda- tion,” in 2019 IEEE 35th International Conference on Data Engineering (ICDE). IEEE, 2019, pp. 566–577

    work page 2019

  5. [5]

    Computing crowd consensus with partial agreement,

    N. Q. V . Hung, H. H. Viet, N. T. Tam, M. Weidlich, H. Yin, and X. Zhou, “Computing crowd consensus with partial agreement,” IEEE Transactions on Knowledge and Data Engineering , vol. 30, no. 1, pp. 1–14, 2017

    work page 2017

  6. [6]

    Self- supervised hypergraph representation learning for sociological analysis,

    X. Sun, H. Cheng, B. Liu, J. Li, H. Chen, G. Xu, and H. Yin, “Self- supervised hypergraph representation learning for sociological analysis,” IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 11, pp. 11 860–11 871, 2023

    work page 2023

  7. [7]

    Hate speech detection with generalizable target-aware fairness,

    T. Chen, D. Wang, X. Liang, M. Risius, G. Demartini, and H. Yin, “Hate speech detection with generalizable target-aware fairness,” in Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , 2024, pp. 365–375

    work page 2024

  8. [8]

    Bourne: Boot- strapped self-supervised learning framework for unified graph anomaly detection,

    J. Liu, M. He, X. Shang, J. Shi, B. Cui, and H. Yin, “Bourne: Boot- strapped self-supervised learning framework for unified graph anomaly detection,” in 2024 IEEE 40th International Conference on Data Engi- neering (ICDE). IEEE, 2024, pp. 2820–2833

    work page 2024

Show all 51 references
  1. [9]

    Accelerating scalable graph neural network inference with node-adaptive propagation,

    X. Gao, W. Zhang, J. Yu, Y . Shao, Q. V . H. Nguyen, B. Cui, and H. Yin, “Accelerating scalable graph neural network inference with node-adaptive propagation,” in2024 IEEE 40th International Conference on Data Engineering (ICDE) . IEEE, 2024, pp. 3042–3055

    2024

  2. [10]

    A survey on fake news and rumour detection techniques,

    A. Bondielli and F. Marcelloni, “A survey on fake news and rumour detection techniques,” Information sciences, vol. 497, pp. 38–55, 2019

    2019

  3. [11]

    A decade of social bot detection,

    S. Cresci, “A decade of social bot detection,” Communications of the ACM, vol. 63, no. 10, pp. 72–83, 2020

    2020

  4. [12]

    A survey on information diffusion in online social networks: Models and methods,

    M. Li, X. Wang, K. Gao, and S. Zhang, “A survey on information diffusion in online social networks: Models and methods,” Information, vol. 8, no. 4, p. 118, 2017

    2017

  5. [13]

    Propagation tree is not deep: Adaptive graph contrastive learning approach for rumor detection,

    C. Cui and C. Jia, “Propagation tree is not deep: Adaptive graph contrastive learning approach for rumor detection,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 38, no. 1, 2024, pp. 73–81

    2024

  6. [14]

    Epidemiology-informed network for robust rumor detection,

    W. Jiang, T. Chen, X. Gao, W. Zhang, L. Cui, and H. Yin, “Epidemiology-informed network for robust rumor detection,” in Pro- ceedings of the ACM on Web Conference 2025 , 2025, pp. 3618–3627

    2025

  7. [15]

    Heterogeneous hypergraph embedding for graph classification,

    X. Sun, H. Yin, B. Liu, H. Chen, J. Cao, Y . Shao, and N. Q. Viet Hung, “Heterogeneous hypergraph embedding for graph classification,” in Proceedings of the 14th ACM international conference on web search and data mining , 2021, pp. 725–733

    2021

  8. [16]

    Botrgcn: Twitter bot de- tection with relational graph convolutional networks,

    S. Feng, H. Wan, N. Wang, and M. Luo, “Botrgcn: Twitter bot de- tection with relational graph convolutional networks,” in Proceedings of the 2021 IEEE/ACM international conference on advances in social networks analysis and mining , 2021, pp. 236–239

    2021

  9. [17]

    Botmoe: Twitter bot detection with community-aware mixtures of modal-specific experts,

    Y . Liu, Z. Tan, H. Wang, S. Feng, Q. Zheng, and M. Luo, “Botmoe: Twitter bot detection with community-aware mixtures of modal-specific experts,” in Proceedings of the 46th International ACM SIGIR Confer- ence on Research and Development in Information Retrieval , 2023, pp. 485–495

    2023

  10. [18]

    Sebot: Structural entropy guided multi-view contrastive learning for social bot detection,

    Y . Yang, Q. Wu, B. He, H. Peng, R. Yang, Z. Hao, and Y . Liao, “Sebot: Structural entropy guided multi-view contrastive learning for social bot detection,” in Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , 2024, pp. 3841–3852

    2024

  11. [19]

    Imbalanced node classification beyond homophilic assumption,

    J. Liu, M. He, G. Wang, N. Q. V . Hung, X. Shang, and H. Yin, “Imbalanced node classification beyond homophilic assumption,” arXiv preprint arXiv:2304.14635, 2023

    2023 arXiv

  12. [20]

    Ms-hgat: memory- enhanced sequential hypergraph attention network for information dif- fusion prediction,

    L. Sun, Y . Rao, X. Zhang, Y . Lan, and S. Yu, “Ms-hgat: memory- enhanced sequential hypergraph attention network for information dif- fusion prediction,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 36, no. 4, 2022, pp. 4156–4164

    2022

  13. [21]

    Rotdiff: A hyperbolic rotation representation model for information diffusion prediction,

    H. Qiao, S. Feng, X. Li, H. Lin, H. Hu, W. Wei, and Y . Ye, “Rotdiff: A hyperbolic rotation representation model for information diffusion prediction,” in Proceedings of the 32nd ACM International Conference on Information and Knowledge Management , 2023, pp. 2065–2074

    2023

  14. [22]

    Enhancing multi-scale diffusion prediction via sequential hypergraphs and adversarial learning,

    P. Jiao, H. Chen, Q. Bao, W. Zhang, and H. Wu, “Enhancing multi-scale diffusion prediction via sequential hypergraphs and adversarial learning,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 8, 2024, pp. 8571–8581

    2024

  15. [23]

    Social noise: the influence of observers on social media information behavior,

    T. Zimmerman, “Social noise: the influence of observers on social media information behavior,” Journal of Documentation , vol. 78, no. 6, pp. 1228–1248, 2022

    2022

  16. [24]

    When ‘friends’ collide: Social heterogeneity and user vulnerability on social network sites,

    S. L. Buglass, J. F. Binder, L. R. Betts, and J. D. Underwood, “When ‘friends’ collide: Social heterogeneity and user vulnerability on social network sites,” Computers in Human Behavior, vol. 54, pp. 62–72, 2016

    2016

  17. [25]

    Government access to and manipulation of social media: Legal and policy challenges,

    R. Levinson-Waldman, “Government access to and manipulation of social media: Legal and policy challenges,” Howard LJ, vol. 61, p. 523, 2017

    2017

  18. [26]

    Social media use by government: From the routine to the critical,

    A. Kavanaugh, E. A. Fox, S. Sheetz, S. Yang, L. T. Li, T. Whalen, D. Shoemaker, P. Natsev, and L. Xie, “Social media use by government: From the routine to the critical,” in Proceedings of the 12th annual in- ternational digital government research conference: digital governme...

    2011

  19. [27]

    The emergence of social media data and sentiment analysis in election prediction,

    P. Chauhan, N. Sharma, and G. Sikka, “The emergence of social media data and sentiment analysis in election prediction,” Journal of Ambient Intelligence and Humanized Computing , vol. 12, pp. 2601–2627, 2021

    2021

  20. [28]

    Social media big data and capital markets—an overview,

    J. Bukovina, “Social media big data and capital markets—an overview,” Journal of Behavioral and Experimental Finance , vol. 11, pp. 18–26, 2016

    2016

  21. [29]

    Physics-guided active sample reweighting for urban flow prediction,

    W. Jiang, T. Chen, G. Ye, W. Zhang, L. Cui, Z. Huang, and H. Yin, “Physics-guided active sample reweighting for urban flow prediction,” arXiv preprint arXiv:2407.13605 , 2024

    2024 arXiv

  22. [30]

    Airphynet: Harnessing physics-guided neural networks for air quality prediction,

    K. H. Hettige, J. Ji, S. Xiang, C. Long, G. Cong, and J. Wang, “Airphynet: Harnessing physics-guided neural networks for air quality prediction,” arXiv preprint arXiv:2402.03784 , 2024

    2024 arXiv

  23. [31]

    Virus spread in networks,

    P. Van Mieghem, J. Omic, and R. Kooij, “Virus spread in networks,” IEEE/ACM Transactions On Networking, vol. 17, no. 1, pp. 1–14, 2008

    2008

  24. [32]

    Dynamics of information diffusion and its applications on complex networks,

    Z.-K. Zhang, C. Liu, X.-X. Zhan, X. Lu, C.-X. Zhang, and Y .-C. Zhang, “Dynamics of information diffusion and its applications on complex networks,” Physics Reports, vol. 651, pp. 1–34, 2016

    2016

  25. [33]

    An individual-based approach to sir epidemics in contact networks,

    M. Youssef and C. Scoglio, “An individual-based approach to sir epidemics in contact networks,” Journal of theoretical biology, vol. 283, no. 1, pp. 136–144, 2011

    2011

  26. [34]

    Rumor detection on social media with bi-directional graph convolu- tional networks,

    T. Bian, X. Xiao, T. Xu, P. Zhao, W. Huang, Y . Rong, and J. Huang, “Rumor detection on social media with bi-directional graph convolu- tional networks,” in Proceedings of the AAAI conference on artificial intelligence, vol. 34, no. 01, 2020, pp. 549–556

    2020

  27. [35]

    Heterogeneity-aware twitter bot detection with relational graph transformers,

    S. Feng, Z. Tan, R. Li, and M. Luo, “Heterogeneity-aware twitter bot detection with relational graph transformers,” inProceedings of the AAAI Conference on Artificial Intelligence , vol. 36, no. 4, 2022, pp. 3977– 3985

    2022

  28. [36]

    Dyhgcn: A dynamic heterogeneous graph convolutional network to learn users’ dynamic preferences for information diffusion prediction,

    C. Yuan, J. Li, W. Zhou, Y . Lu, X. Zhang, and S. Hu, “Dyhgcn: A dynamic heterogeneous graph convolutional network to learn users’ dynamic preferences for information diffusion prediction,” in Machine Learning and Knowledge Discovery in Databases: European Confer- ence, ECML P...

    2020

  29. [37]

    Social bots and the spread of disinformation in social media: the challenges of artificial intelligence,

    N. Hajli, U. Saeed, M. Tajvidi, and F. Shirazi, “Social bots and the spread of disinformation in social media: the challenges of artificial intelligence,” British Journal of Management , vol. 33, no. 3, pp. 1238– 1253, 2022

    2022

  30. [38]

    Bots and misinformation spread on social media: implications for covid-19,

    M. Himelein-Wachowiak, S. Giorgi, A. Devoto, M. Rahman, L. Ungar, H. A. Schwartz, D. H. Epstein, L. Leggio, and B. Curtis, “Bots and misinformation spread on social media: implications for covid-19,” Journal of medical Internet research , vol. 23, no. 5, p. e26933, 2021

    2021

  31. [39]

    Graph con- trastive learning with augmentations,

    Y . You, T. Chen, Y . Sui, T. Chen, Z. Wang, and Y . Shen, “Graph con- trastive learning with augmentations,” Advances in neural information processing systems, vol. 33, pp. 5812–5823, 2020

    2020

  32. [40]

    Detecting rumors from microblogs with recurrent neural networks,

    J. Ma, W. Gao, P. Mitra, S. Kwon, B. J. Jansen, K.-F. Wong, and M. Cha, “Detecting rumors from microblogs with recurrent neural networks,” 2016

    2016

  33. [41]

    Twibot-22: Towards graph-based twitter bot detection,

    S. Feng, Z. Tan, H. Wan, N. Wang, Z. Chen, B. Zhang, Q. Zheng, W. Zhang, Z. Lei, S. Yang et al. , “Twibot-22: Towards graph-based twitter bot detection,” Advances in Neural Information Processing Systems, vol. 35, pp. 35 254–35 269, 2022. 10

    2022

  34. [42]

    Mgtab: A multi-relational graph-based twitter account detection benchmark,

    S. Shi, K. Qiao, J. Chen, S. Yang, J. Yang, B. Song, L. Wang, and B. Yan, “Mgtab: A multi-relational graph-based twitter account detection benchmark,” arXiv preprint arXiv:2301.01123 , 2023

    2023 arXiv

  35. [43]

    Inf-vae: A variational autoencoder framework to integrate homophily and influence in diffusion prediction,

    A. Sankar, X. Zhang, A. Krishnan, and J. Han, “Inf-vae: A variational autoencoder framework to integrate homophily and influence in diffusion prediction,” in Proceedings of the 13th international conference on web search and data mining , 2020, pp. 510–518

    2020

  36. [44]

    The simple rules of social contagion,

    N. O. Hodas and K. Lerman, “The simple rules of social contagion,” Scientific reports, vol. 4, no. 1, p. 4343, 2014

    2014

  37. [45]

    Detecting social bot on the fly using contrastive learning,

    M. Zhou, D. Zhang, Y . Wang, Y .-A. Geng, and J. Tang, “Detecting social bot on the fly using contrastive learning,” in Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, 2023, pp. 4995–5001

    2023

  38. [46]

    Information diffusion prediction with graph neural ordinary differential equation network,

    D. Wang, W. Zhou, and S. Hu, “Information diffusion prediction with graph neural ordinary differential equation network,” in Proceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 9699– 9708

    2024

  39. [47]

    Kagnns: Kolmogorov-arnold networks meet graph learning,

    R. Bresson, G. Nikolentzos, G. Panagopoulos, M. Chatzianastasis, J. Pang, and M. Vazirgiannis, “Kagnns: Kolmogorov-arnold networks meet graph learning,” arXiv preprint arXiv:2406.18380 , 2024

    2024 arXiv

  40. [48]

    Rumor detection on social media with graph adversarial contrastive learning,

    T. Sun, Z. Qian, S. Dong, P. Li, and Q. Zhu, “Rumor detection on social media with graph adversarial contrastive learning,” in Proceedings of the ACM Web Conference 2022 , 2022, pp. 2789–2797

    2022

  41. [49]

    Semantic evolvement enhanced graph autoencoder for rumor detection,

    X. Tao, L. Wang, Q. Liu, S. Wu, and L. Wang, “Semantic evolvement enhanced graph autoencoder for rumor detection,” in Proceedings of the ACM on Web Conference 2024 , 2024, pp. 4150–4159

    2024

  42. [50]

    Are we really making much progress? revisiting, benchmarking and refining heterogeneous graph neural networks,

    Q. Lv, M. Ding, Q. Liu, Y . Chen, W. Feng, S. He, C. Zhou, J. Jiang, Y . Dong, and J. Tang, “Are we really making much progress? revisiting, benchmarking and refining heterogeneous graph neural networks,” in Proceedings of the 27th ACM SIGKDD conference on knowledge discovery ...

    2021

  43. [51]

    Full-scale information diffusion prediction with reinforced recurrent networks,

    C. Yang, H. Wang, J. Tang, C. Shi, M. Sun, G. Cui, and Z. Liu, “Full-scale information diffusion prediction with reinforced recurrent networks,” IEEE Transactions on Neural Networks and Learning Sys- tems, vol. 34, no. 5, pp. 2271–2283, 2021. 11

    2021

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.