Pith. sign in

REVIEW 3 major objections 6 minor 55 references

ISMAF: Intrinsic-Social Modality Alignment and Fusion for Multimodal Rumor Detection

T0 review · 3 major / 6 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 This paper claims that explicitly aligning social-context features with text and image features, then fusing them adaptively, makes a rumor detector outperform existing top models on both English and Chinese datasets.

desk verdict Competent incremental multimodal rumor detector with a strong ablation but an unverifiable SOTA comparison until code and data splits are released. read the letter →

arxiv 2505.24176 v1 pith:ILQAFG65 submitted 2025-05-30 cs.MM

Zihao Yu , Xiang Li , Jing Zhang This is my paper
classification cs.MM
keywords rumordetectionmultimodalfusioncross-modalalignmentsupervisedcontrastivelearningadaptivesocialcontextintrinsicmodality
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a rumor detector improves when it treats text and image as one intrinsic modality and propagation/user-reaction features as a separate social modality, and then explicitly aligns the two before fusing all three signals. On the English PHEME and Chinese Weibo datasets, the proposed ISMAF framework reports accuracy and F1 of 91.01/89.09 and 93.42/92.95, respectively, beating the strongest prior model CLFFRD on every metric. A sympathetic reader would care because earlier multimodal detectors align only text with images or fuse social context in a fixed way, leaving the tension between what a post says and how it spreads unexploited. The paper's point is that this tension is itself informative and can be captured with three complementary loss terms plus an adaptive fusion step.

What carries the argument

The load-bearing object is the unified bridging modalities scheme that connects the intrinsic modality (text plus image, collapsed into a co-attended vector $Z^i$) with the social modality (a graph-attention embedding $R^i_G$ of posts, comments, and users). It works through two complementary losses: a cross-modal consistency alignment loss $L_{cmca}$ that contrastively binds $Z^i$ to $R^i_G$ for the same post, and a mutual learning loss $L_{ml}$ that symmetrizes the KL divergence between the two modality-specific classifiers. These are combined with supervised contrastive learning, an encoder-decoder reconstruction loss for adaptive fusion, and cross-entropy in a weighted total loss $L_{overall} = L_{ce} + \lambda_1 L_{scl} + \lambda_2 L_{cmca} + \lambda_3 L_{ml} + \lambda_4 L_{af}$.

What would settle it

Re-run the comparison of ISMAF with CLFFRD and MFAN on the canonical public splits and original labels of PHEME and Weibo; if ISMAF's accuracy and F1 edges vanish or flip under matched preprocessing, the central claim is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that modeling the correspondence between intrinsic content (text and images) and social context (comments, users, propagation structure) is what pushes multimodal rumor detection past current top results. ISMAF first refines all unimodal features with supervised contrastive learning, then builds an intrinsic representation by multi-head self- and co-attention over text and image, and pulls that representation toward the social-graph representation of the same post with a contrastive consistency loss. A mutual-learning loss based on symmetric KL divergence between the two modality-specific classifiers further aligns their decisions, and an encoder-decoder adaptive fusion mechanism weighs the three modalities dynamically. With all components active, the model reaches 91.01% ACC/89.09% F1 on PHEME and 93.42% ACC/92.95% F1 on Weibo, exceeding CLFFRD by roughly 1.06/0.96 and 2.16/3.13 points, and the ablations show each component contributes.

Load-bearing premise

The experimental claim stands on the assumption that the re-annotated, filtered PHEME and Weibo datasets used for ISMAF are comparable to the versions used to produce the baseline numbers, so the reported gains reflect the method rather than the preprocessing.

Editorial extensions

If this is right

  • If the central claim holds, multimodal rumor detectors should treat intrinsic-social inconsistency as a first-class signal rather than only aligning text with images.
  • The ablations show that removing either the consistency alignment or the mutual learning costs roughly 1.2 to 1.6 accuracy points, while removing both costs 2.06 points on PHEME and 2.57 points on Weibo, indicating the two modules work synergistically.
  • Replacing adaptive fusion with plain concatenation drops accuracy by 1.71 points on PHEME and 1.56 points on Weibo, so the dynamic weighting is doing real work beyond simple feature combination.
  • The reported tuned loss weights, with the consistency alignment term weighted higher than the supervised contrastive term, suggest that cross-modal binding is the more influential contrastive component in this design.

Reading between the lines

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

  • The paper leaves implicit that the intrinsic-social distinction could serve as a weak supervision signal: posts whose content looks credible but whose propagation network is suspicious, or vice versa, are exactly the hard cases these alignment losses shape, so the model may be learning to flag inconsistency rather than just content.
  • A natural extension the paper does not test is whether the same alignment scheme helps early rumor detection, where only partial propagation data is available; the contrastive binding to incomplete social graphs may need confidence weighting to remain effective.
  • Because the authors re-annotate and filter both datasets and do not release splits or code, an independent re-implementation on the canonical public PHEME and Weibo splits would test whether the reported margins persist outside the filtered setting.
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 / 6 minor

Summary. The paper proposes ISMAF, a multimodal rumor detection framework that combines three-modality features (text, image, social graph) with supervised contrastive learning, a cross-modal consistency alignment loss between intrinsic and social representations, a mutual-learning KL divergence between the two modality-specific classifiers, and an encoder-decoder adaptive fusion mechanism. The method is evaluated on PHEME and Weibo datasets against eight baselines. The central empirical claim is that ISMAF consistently outperforms all baselines on both datasets, with reported ACC/F1 of 91.01/89.09 on PHEME and 93.42/92.95 on Weibo, outperforming the strongest baseline CLFFRD by roughly 1.06/0.96 and 2.16/3.13 points respectively.

Significance. If the reported results are reproducible, ISMAF is a reasonable architectural contribution to multimodal rumor detection: it explicitly models intrinsic-social alignment, provides component-level ablations, and reports standard deviations over five runs. However, the paper's value is almost entirely empirical, and the experimental protocol currently does not establish apples-to-apples comparability with the baselines because the dataset sizes are non-standard, the preprocessing is not fully specified, and no code, splits, or preprocessing scripts are released. The central claim therefore rests on an unverified comparability assumption rather than on a theoretical or formal result.

major comments (3)
  1. [Section V-A, Table II] The dataset protocol is not shown to be comparable to the published baselines. PHEME is reduced to 2,018 posts and Weibo to 1,467 posts, and the text states that the authors 'annotated both datasets with binary labels' and 'excluded instances missing either text or image data during preprocessing.' These filtered sizes and the labeling wording indicate a custom preprocessing pipeline, whereas MFAN and CLFFRD report results on their own splits of these datasets. Since no code, preprocessing scripts, or instance IDs are released, the Table III gains over CLFFRD (PHEME ACC/F1 +1.06/+0.96, Weibo +2.16/+3.13) could be an artifact of filtering, relabeling, or split differences rather than of the proposed method. The authors should specify the exact source and derivation of the data, provide the exact train/validation/test splits or instance lists, and rerun all baselines under the identical protocol; alternatively, they should use canonical public splits.
  2. [Section V-E] The hyperparameters λ1–λ4 are selected by 'preliminary experiments' on the same datasets, and the reported sensitivity analysis only shows five parameter sets around the selected point. Because the method combines several interacting loss terms, selection of these weights on the same corpus used for the final comparison can inflate the reported performance. The authors should either adopt a nested validation procedure or report results across a broader grid of λ values, and demonstrate that the consistent improvement over CLFFRD in Table III is not an artifact of tuning near the reported operating point.
  3. [Section IV-B1 and Section V-A] The social graph construction is underspecified and its sensitivity is not evaluated. The edges are formed by cosine similarity with an unspecified threshold, and the number of GAT layers, the mini-batch size B, and the exact training epochs are not reported in the parameter settings. Since the social graph is the sole source of the social modality, the reported results are not reproducible without these details. The authors should report the edge threshold and other graph-construction choices, and include a sensitivity analysis for the threshold.
minor comments (6)
  1. [Section IV-A and Fig. 2] The phrase 'Adptive Fusion and Detection' contains a typo; it should read 'Adaptive Fusion and Detection.'
  2. [Equation (2)] In the attention formula, the value matrix is written as V_i^T but it should presumably be V_i^m for the corresponding modality; please correct this notation.
  3. [Equations (12)-(13)] The KL divergence is written as a sum over samples without an explicit category index or averaging over the batch; please provide the exact per-sample and per-batch formulation used in the implementation.
  4. [Equations (15)-(16)] The classification head is described as a softmax followed by a binary cross-entropy-style loss; please clarify whether the output layer has one or two units and how the softmax probability is mapped to the scalar ŷ_i in Eq. (16).
  5. [Section V-E and Fig. 4] The text says a 'representative subset of five distinct parameter sets' is visualized, but Fig. 4 contains eight panels; this inconsistency should be fixed.
  6. [Section V-A] Several training details are missing: the value of the mini-batch size B, the number of epochs, the learning-rate decay schedule, and the cosine-similarity threshold for the social graph edges. These should be reported for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ISMAF's central claim is an empirical benchmark comparison, not a derivation that reduces to its own inputs.

full rationale

The paper's central claim is that ISMAF outperforms prior methods on PHEME and Weibo (Section V-C, Table III). This is an empirical result obtained by training the proposed model and comparing it against external baselines; no prediction or derived quantity is defined in terms of the claimed outcome. The loss components (Lscl, Lcmca, Lml, Laf) are combined in Eq. (17) with weights selected by preliminary experiments, which is ordinary hyperparameter tuning rather than fitting the reported test numbers. The ablation study (Section V-D) compares model variants empirically, and the sensitivity analysis (Section V-E) examines robustness around the chosen weights; neither renames fitted parameters as predictions. The paper cites prior work for components such as GAT and supervised contrastive learning, but those citations provide standard building blocks and do not carry a load-bearing claim that the authors' own method is forced by an imported uniqueness theorem. The main validity concern is dataset comparability: Section V-A states that the authors annotated both datasets with binary labels and excluded instances missing text or image data, and Table II reports PHEME with 2,018 posts and Weibo with 1,467 posts, which may differ from the splits used by the baselines. However, that is an experimental-reproducibility and apples-to-apples benchmark concern, not a circularity of the derivation. Since no equation, fitted parameter, or self-citation is shown to be equivalent to the paper's reported conclusion, the appropriate circularity finding is none.

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

The central claim rests on standard machine-learning components and a set of hand-chosen hyperparameters rather than a theoretical derivation. The most consequential free parameters are the loss weights lambda, selected by preliminary experiments, and the unreported social graph threshold and temperature. The paper's conceptual split of text and image as intrinsic versus social graph as social is a framing choice, not a new entity.

free parameters (6)
  • Loss weights lambda1, lambda2, lambda3, lambda4 = (0.3, 0.7, 0.4, 0.4)
    Selected by 'preliminary experiments' in Section V-E; these weights control the contribution of each loss to the final objective and are tuned on the same datasets used for evaluation.
  • Contrastive temperature tau = not reported
    Used in supervised contrastive loss Eq. (1) and cross-modal consistency loss Eq. (6); no value or schedule is given, so reimplementation cannot match without guessing.
  • Social graph edge threshold = not reported
    Section IV-B1 says edges are formed by cosine similarity 'where a threshold determines the connections'; the threshold is never specified, and it directly controls the social context features.
  • Mini-batch size B and number of epochs = not reported
    Algorithm 1 depends on mini-batch size B and epochs epsilon, but neither is listed in the parameter settings.
  • PGD perturbation parameters = not reported
    Section V-B applies Projected Gradient Descent following [21], but the perturbation bound and number of steps are not given.
  • Autoencoder layer sizes and activations = not reported
    Section IV-D uses an encoder-decoder to compress 3d to d, but layer widths, depth, and activation details are unspecified.
assumptions (4)
  • domain assumption The filtered PHEME and Weibo datasets carry reliable binary rumor labels and are comparable to the splits used by baseline methods.
    Section V-A describes annotation and filtering, but does not show that the resulting 2018/1467 posts match baseline preprocessing.
  • domain assumption A social graph built from cosine similarity of post, comment, and user text embeddings captures meaningful propagation and interaction context.
    Section IV-B1 constructs the social graph this way; the rumor signal must be present in these content-similarity edges.
  • ad hoc to paper Symmetric KL divergence between intrinsic and social classifiers transfers useful knowledge without degrading either classifier.
    Section IV-C2 introduces mutual learning as a design choice; no analysis is given for when this assumption holds.
  • ad hoc to paper The reconstruction objective in adaptive fusion preserves the information needed for classification.
    Section IV-D uses an encoder-decoder with L2 reconstruction loss, but task relevance of the reconstructed representation is not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ISMAF: Intrinsic-Social Modality Alignment and Fusion for Multimodal Rumor Detection." pith.science (2026). https://pith.science/paper/ILQAFG65

@misc{pith2026250524176,
  author       = {Pith},
  title        = {Pith review of: ISMAF: Intrinsic-Social Modality Alignment and Fusion for Multimodal Rumor Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ILQAFG65}},
  note         = {Machine review of arXiv:2505.24176}
}
read the original abstract

The rapid dissemination of rumors on social media highlights the urgent need for automatic detection methods to safeguard societal trust and stability. While existing multimodal rumor detection models primarily emphasize capturing consistency between intrinsic modalities (e.g., news text and images), they often overlook the intricate interplay between intrinsic and social modalities. This limitation hampers the ability to fully capture nuanced relationships that are crucial for a comprehensive understanding. Additionally, current methods struggle with effectively fusing social context with textual and visual information, resulting in fragmented interpretations. To address these challenges, this paper proposes a novel Intrinsic-Social Modality Alignment and Fusion (ISMAF) framework for multimodal rumor detection. ISMAF first employs a cross-modal consistency alignment strategy to align complex interactions between intrinsic and social modalities. It then leverages a mutual learning approach to facilitate collaborative refinement and integration of complementary information across modalities. Finally, an adaptive fusion mechanism is incorporated to dynamically adjust the contribution of each modality, tackling the complexities of three-modality fusion. Extensive experiments on both English and Chinese real-world multimedia datasets demonstrate that ISMAF consistently outperforms state-of-the-art models.

Figures

Figures reproduced from arXiv: 2505.24176 by the authors.

Figure 1
Figure 1. Intrinsic modality and social modality of multimodal posts. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The proposed ISMAF framework. The above is an overview structure and the internal structures of three modules are below. The intrinsic and social [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Performance comparison of fusion strategies on the PHEME and [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Parametric analysis of the hyperparameter set [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

55 extracted references · 45 canonical work pages

  1. [1]

    Combating fake news: A survey on identification and mitigation techniques,

    K. Sharma, F. Qian, H. Jiang, N. Ruchansky, M. Zhang, and Y . Liu, “Combating fake news: A survey on identification and mitigation techniques,” ACM Transactions on Intelligent Systems and Technology (TIST), vol. 10, no. 3, pp. 1–42, 2019

    work page 2019

  2. [2]

    Fake news detection on social media: A data mining perspective,

    K. Shu, A. Sliva, S. Wang, J. Tang, and H. Liu, “Fake news detection on social media: A data mining perspective,” ACM SIGKDD explorations newsletter, vol. 19, no. 1, pp. 22–36, 2017

    work page 2017

  3. [3]

    The spread of covid- 19 fake news on social media and its impact among malaysians,

    S. Ngadiron, A. Abd Aziz, and S. S. Mohamed, “The spread of covid- 19 fake news on social media and its impact among malaysians,” Multidisciplinary Approaches in Social Sciences, Islamic & Technology (ICMASIT 2020), vol. 13, p. 222, 2020

    work page 2020

  4. [4]

    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

  5. [5]

    A convolutional approach for misinformation identification

    F. Yu, Q. Liu, S. Wu, L. Wang, T. Tan et al., “A convolutional approach for misinformation identification.” in IJCAI, 2017, pp. 3901–3907

    work page 2017

  6. [6]

    Eann: Event adversarial neural networks for multi-modal fake news detection,

    Y . Wang, F. Ma, Z. Jin, Y . Yuan, G. Xun, K. Jha, L. Su, and J. Gao, “Eann: Event adversarial neural networks for multi-modal fake news detection,” in Proceedings of the 24th acm sigkdd international conference on knowledge discovery & data mining , 2018, pp. 849–857

    work page 2018

  7. [7]

    Mvae: Multimodal variational autoencoder for fake news detection,

    D. Khattar, J. S. Goud, M. Gupta, and V . Varma, “Mvae: Multimodal variational autoencoder for fake news detection,” in The world wide web conference, 2019, pp. 2915–2921

    work page 2019

  8. [8]

    Spotfake: A multi-modal framework for fake news detection,

    S. Singhal, R. R. Shah, T. Chakraborty, P. Kumaraguru, and S. Satoh, “Spotfake: A multi-modal framework for fake news detection,” in 2019 IEEE fifth international conference on multimedia big data (BigMM) . IEEE, 2019, pp. 39–47

    work page 2019

Show all 55 references
  1. [9]

    Safe: similarity-aware multi-modal fake news detection (2020),

    X. Zhou, J. Wu, and R. Zafarani, “Safe: similarity-aware multi-modal fake news detection (2020),” Preprint. arXiv , vol. 200304981, no. 2, 2020

    2020

  2. [10]

    Novel visual and statisti- cal image features for microblogs news verification,

    Z. Jin, J. Cao, Y . Zhang, J. Zhou, and Q. Tian, “Novel visual and statisti- cal image features for microblogs news verification,” IEEE transactions on multimedia, vol. 19, no. 3, pp. 598–608, 2016

    2016

  3. [11]

    Exploiting multi-domain visual information for fake news detection,

    P. Qi, J. Cao, T. Yang, J. Guo, and J. Li, “Exploiting multi-domain visual information for fake news detection,” in 2019 IEEE international conference on data mining (ICDM) . IEEE, 2019, pp. 518–527

    2019

  4. [12]

    Detect rumors using time series of social context information on microblogging websites,

    J. Ma, W. Gao, Z. Wei, Y . Lu, and K.-F. Wong, “Detect rumors using time series of social context information on microblogging websites,” in Proceedings of the 24th ACM international on conference on information and knowledge management , 2015, pp. 1751–1754

    2015

  5. [13]

    Rumor detection on social media with graph structured adversarial learning,

    X. Yang, Y . Lyu, T. Tian, Y . Liu, Y . Liu, and X. Zhang, “Rumor detection on social media with graph structured adversarial learning,” in Proceed- ings of the twenty-ninth international conference on international joint conferences on artificial intelligence , 2021, pp. 1417–1423

    2021

  6. [14]

    User preference-aware fake news detection,

    Y . Dou, K. Shu, C. Xia, P. S. Yu, and L. Sun, “User preference-aware fake news detection,” in Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval, 2021, pp. 2051–2055

    2021

  7. [15]

    Rumor detection on social media with event augmentations,

    Z. He, C. Li, F. Zhou, and Y . Yang, “Rumor detection on social media with event augmentations,” in Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval, 2021, pp. 2020–2024

    2021

  8. [16]

    Early detection of fake news on social media through propagation path classification with recurrent and convolutional networks,

    Y . Liu and Y .-F. Wu, “Early detection of fake news on social media through propagation path classification with recurrent and convolutional networks,” in Proceedings of the AAAI conference on artificial intelli- gence, vol. 32, no. 1, 2018

    2018

  9. [17]

    Rumor detection with hierarchical social attention network,

    H. Guo, J. Cao, Y . Zhang, J. Guo, and J. Li, “Rumor detection with hierarchical social attention network,” in Proceedings of the 27th ACM international conference on information and knowledge management , 2018, pp. 943–951

    2018

  10. [18]

    Jointly embedding the local and global relations of heterogeneous graph for rumor detection,

    C. Yuan, Q. Ma, W. Zhou, J. Han, and S. Hu, “Jointly embedding the local and global relations of heterogeneous graph for rumor detection,” in 2019 IEEE international conference on data mining (ICDM) . IEEE, 2019, pp. 796–805

    2019

  11. [19]

    Unsupervised rumor detection based on propagation tree vae,

    L. Fang, K. Feng, K. Zhao, A. Hu, and T. Li, “Unsupervised rumor detection based on propagation tree vae,” IEEE Transactions on Knowl- edge and Data Engineering , vol. 35, no. 10, pp. 10 309–10 323, 2023

    2023

  12. [20]

    Multimodal fusion with recurrent neural networks for rumor detection on microblogs,

    Z. Jin, J. Cao, H. Guo, Y . Zhang, and J. Luo, “Multimodal fusion with recurrent neural networks for rumor detection on microblogs,” in Proceedings of the 25th ACM international conference on Multimedia , 2017, pp. 795–816

    2017

  13. [21]

    Mfan: Multi-modal feature-enhanced attention networks for rumor detection

    J. Zheng, X. Zhang, S. Guo, Q. Wang, W. Zang, and Y . Zhang, “Mfan: Multi-modal feature-enhanced attention networks for rumor detection.” in IJCAI, vol. 2022, 2022, pp. 2413–2419. JOURNAL OF LATEX CLASS FILES, VOL. 99, NO. 9, MAY 2025 11

    2022

  14. [22]

    Clffrd: Curriculum learning and fine-grained fusion for multimodal rumor detection,

    F. Xu, L. Zeng, B. Zou, A. Aw, and H. Rong, “Clffrd: Curriculum learning and fine-grained fusion for multimodal rumor detection,” in Proceedings of the 2024 Joint International Conference on Computa- tional Linguistics, Language Resources and Evaluation (LREC-COLING 2024), 202...

    2024

  15. [23]

    Hierarchical multi-modal contextual attention network for fake news detection,

    S. Qian, J. Wang, J. Hu, Q. Fang, and C. Xu, “Hierarchical multi-modal contextual attention network for fake news detection,” in Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval , 2021, pp. 153–162

    2021

  16. [24]

    Hierarchical graph attention networks for multi-modal rumor detection on social media,

    F. Xu, L. Zeng, Q. Huang, K. Yan, M. Wang, and V . S. Sheng, “Hierarchical graph attention networks for multi-modal rumor detection on social media,” Neurocomputing, vol. 569, p. 127112, 2024

    2024

  17. [25]

    Leveraging intra and inter modality relationship for multimodal fake news detection,

    S. Singhal, T. Pandey, S. Mrig, R. R. Shah, and P. Kumaraguru, “Leveraging intra and inter modality relationship for multimodal fake news detection,” in Companion Proceedings of the Web Conference 2022, 2022, pp. 726–734

    2022

  18. [26]

    Multimodal fusion with co-attention networks for fake news detection,

    Y . Wu, P. Zhan, Y . Zhang, L. Wang, and Z. Xu, “Multimodal fusion with co-attention networks for fake news detection,” in Findings of the association for computational linguistics: ACL-IJCNLP 2021 , 2021, pp. 2560–2569

    2021

  19. [27]

    Automatic detection of fake news,

    V . P´erez-Rosas, B. Kleinberg, A. Lefevre, and R. Mihalcea, “Automatic detection of fake news,” arXiv preprint arXiv:1708.07104 , 2017

    2017 arXiv

  20. [28]

    Survey of fake news detection with multi-model learning,

    L. Hualing, C. Shanghui, C. Shijie, Z. Jianliang, and R. Qingqing, “Survey of fake news detection with multi-model learning,” Journal of Frontiers of Computer Science & Technology , vol. 17, no. 9, p. 2015, 2023

    2015

  21. [29]

    Towards news verification: De- ception detection methods for news discourse,

    V . L. Rubin, N. J. Conroy, and Y . Chen, “Towards news verification: De- ception detection methods for news discourse,” in Hawaii international conference on system sciences , 2015, pp. 5–8

    2015

  22. [30]

    A stylometric inquiry into hyperpartisan and fake news,

    M. Potthast, J. Kiesel, K. Reinartz, J. Bevendorff, and B. Stein, “A stylometric inquiry into hyperpartisan and fake news,” arXiv preprint arXiv:1702.05638, 2017

    2017 arXiv

  23. [31]

    defend: Explainable fake news detection,

    K. Shu, L. Cui, S. Wang, D. Lee, and H. Liu, “defend: Explainable fake news detection,” in Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining , 2019, pp. 395–405

    2019

  24. [32]

    Stanker: Stacking network based on level-grained attention-masked bert for rumor detection on social media,

    D. Rao, X. Miao, Z. Jiang, and R. Li, “Stanker: Stacking network based on level-grained attention-masked bert for rumor detection on social media,” in Proceedings of the 2021 conference on empirical methods in natural language processing, 2021, pp. 3347–3363

    2021

  25. [33]

    Leveraging joint interactions for credi- bility analysis in news communities,

    S. Mukherjee and G. Weikum, “Leveraging joint interactions for credi- bility analysis in news communities,” in Proceedings of the 24th ACM International on Conference on Information and Knowledge Manage- ment, 2015, pp. 353–362

    2015

  26. [34]

    Early detection of fake news by utilizing the credibility of news, publishers, and users based on weakly supervised learning,

    C. Yuan, Q. Ma, W. Zhou, J. Han, and S. Hu, “Early detection of fake news by utilizing the credibility of news, publishers, and users based on weakly supervised learning,” arXiv preprint arXiv:2012.04233 , 2020

    2012 arXiv

  27. [35]

    Multi-level attention map network for multimodal sentiment analysis,

    X. Xue, C. Zhang, Z. Niu, and X. Wu, “Multi-level attention map network for multimodal sentiment analysis,” IEEE Transactions on Knowledge and Data Engineering , vol. 35, no. 5, pp. 5105–5118, 2022

    2022

  28. [36]

    Mining dual emotion for fake news detection,

    X. Zhang, J. Cao, X. Li, Q. Sheng, L. Zhong, and K. Shu, “Mining dual emotion for fake news detection,” in Proceedings of the web conference 2021, 2021, pp. 3465–3476

    2021

  29. [37]

    Rumor detection by exploiting user credi- bility information, attention and multi-task learning,

    Q. Li, Q. Zhang, and L. Si, “Rumor detection by exploiting user credi- bility information, attention and multi-task learning,” in Proceedings of the 57th annual meeting of the association for computational linguistics , 2019, pp. 1173–1179

    2019

  30. [38]

    Zoom out and observe: News environment perception for fake news detection,

    Q. Sheng, J. Cao, X. Zhang, R. Li, D. Wang, and Y . Zhu, “Zoom out and observe: News environment perception for fake news detection,” arXiv preprint arXiv:2203.10885, 2022

    2022 arXiv

  31. [39]

    Rumor detection on twitter with tree-structured recursive neural networks,

    J. Ma, W. Gao, and K.-F. Wong, “Rumor detection on twitter with tree-structured recursive neural networks,” in Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (ACL). Association for Computational Linguistics, 2018

    2018

  32. [40]

    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

  33. [41]

    Towards propagation uncertainty: Edge-enhanced bayesian graph convolutional networks for rumor detection,

    L. Wei, D. Hu, W. Zhou, Z. Yue, and S. Hu, “Towards propagation uncertainty: Edge-enhanced bayesian graph convolutional networks for rumor detection,” arXiv preprint arXiv:2107.11934 , 2021

    2021 arXiv

  34. [42]

    A survey on multimodal disinformation detection,

    F. Alam, S. Cresci, T. Chakraborty, F. Silvestri, D. Dimitrov, G. D. S. Martino, S. Shaar, H. Firooz, and P. Nakov, “A survey on multimodal disinformation detection,” arXiv preprint arXiv:2103.12541 , 2021

    2021 arXiv

  35. [43]

    Hgmf: heterogeneous graph-based fusion for multimodal data with incompleteness,

    J. Chen and A. Zhang, “Hgmf: heterogeneous graph-based fusion for multimodal data with incompleteness,” in Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, 2020, pp. 1295–1305

    2020

  36. [44]

    Detecting fake news by exploring the consistency of multimodal data,

    J. Xue, Y . Wang, Y . Tian, Y . Li, L. Shi, and L. Wei, “Detecting fake news by exploring the consistency of multimodal data,” Information Processing & Management , vol. 58, no. 5, p. 102610, 2021

    2021

  37. [45]

    Game-on: Graph attention network based multimodal fusion for fake news detection,

    M. Dhawan, S. Sharma, A. Kadam, R. Sharma, and P. Kumaraguru, “Game-on: Graph attention network based multimodal fusion for fake news detection,” Social Network Analysis and Mining , vol. 14, no. 1, p. 114, 2024

    2024

  38. [46]

    Rumor detection based on cross- modal information-enhanced fusion network,

    Z. Guo, Z. Yang, and D. Liu, “Rumor detection based on cross- modal information-enhanced fusion network,” in2024 16th International Conference on Advanced Computational Intelligence (ICACI) . IEEE, 2024, pp. 158–164

    2024

  39. [47]

    Human cognition-based consistency inference networks for multi-modal fake news detection,

    L. Wu, P. Liu, Y . Zhao, P. Wang, and Y . Zhang, “Human cognition-based consistency inference networks for multi-modal fake news detection,” IEEE Transactions on Knowledge and Data Engineering , vol. 36, no. 1, pp. 211–225, 2023

    2023

  40. [48]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778

    2016

  41. [49]

    Supervised contrastive learn- ing,

    P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, A. Maschinot, C. Liu, and D. Krishnan, “Supervised contrastive learn- ing,” Advances in neural information processing systems , vol. 33, pp. 18 661–18 673, 2020

    2020

  42. [50]

    A simple framework for contrastive learning of visual representations,

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” in International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, H. Daume and A. Singh, Eds., vol. 119, 2020, international...

    2020

  43. [51]

    Adaptive fusion techniques for multi- modal data,

    G. Sahu and O. Vechtomova, “Adaptive fusion techniques for multi- modal data,” arXiv preprint arXiv:1911.03821 , 2019

    1911 arXiv

  44. [52]

    Ced: Credible early detection of social media rumors,

    C. Song, C. Yang, H. Chen, C. Tu, Z. Liu, and M. Sun, “Ced: Credible early detection of social media rumors,” IEEE Transactions on Knowledge and Data Engineering , vol. 33, no. 8, pp. 3035–3047, 2019

    2019

  45. [53]

    Towards deep learning models resistant to adversarial attacks,

    A. Madry, A. Makelov, L. Schmidt, D. Tsipras, and A. Vladu, “Towards deep learning models resistant to adversarial attacks,” 2017

    2017

  46. [54]

    Adam: A method for stochastic optimization,

    D. Kingma and J. Ba, “Adam: A method for stochastic optimization,” Computer Science, 2014

    2014

  47. [55]

    Qsan: A quantum-probability based signed attention network for explainable false information detection,

    T. Tian, Y . Liu, X. Yang, Y . Lyu, X. Zhang, and B. Fang, “Qsan: A quantum-probability based signed attention network for explainable false information detection,” in Proceedings of the 29th ACM international conference on information & knowledge management , 2020, pp. 1445– 1454

    2020

Pith tools

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