Pith. sign in

REVIEW 2 major objections 2 minor 56 references

TRENT outperforms state-of-the-art models on real-world multimodal fact-checking by triangulating evidence through three parallel cross-attention streams and relational fusion.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-07-01 03:03 UTC pith:IOXZDAXD

load-bearing objection X-POSE benchmark and TRENT's three-stream cross-attention plus relational fusion address real gaps in synthetic MFC data, but the outperformance claim rests on unverified community labels and VLM retrieval with no metrics shown. the 2 major comments →

arxiv 2606.31367 v1 pith:IOXZDAXD submitted 2026-06-30 cs.MM cs.CV

Evidence Triangulation for Multimodal Fact-Checking in the Wild

Stefanos-Iordanis Papadopoulos , Zacharias Chrysidis , Christos Koutlis , Symeon Papadopoulos , Panagiotis C. Petrantonakis This is my paper
classification cs.MM cs.CV
keywords multimodal fact-checkingevidence triangulationX-POSE benchmarkcross-attention streamsrelational fusionmisinformation detectionsocial media postsvision-language models
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper creates X-POSE, a benchmark drawn from actual X posts that carry community annotations and are paired with full-length news articles found through optimized retrieval. It introduces TRENT, which routes the post and evidence through three separate cross-attention streams before applying a relational fusion step that distinguishes entailment from contradiction. This design targets the mismatch between synthetic training sets and the mixed, external-evidence nature of in-the-wild misinformation. A sympathetic reader would care because current detectors either ignore external sources or fuse them too loosely, leaving performance gaps on authentic cases. If the triangulation approach succeeds, fact-checking systems could integrate supporting or refuting articles more reliably without defaulting to internal consistency checks alone.

Core claim

TRENT performs evidence triangulation using three parallel cross-attention streams alongside a relational fusion mechanism that explicitly models entailment and contradiction, and this allows it to consistently outperform state-of-the-art specialized models and commercial VLMs on the X-POSE benchmark of real-world multimodal posts.

What carries the argument

Evidence triangulation via three parallel cross-attention streams and relational fusion that models entailment and contradiction.

Load-bearing premise

Community annotations on X posts together with VLM-optimized retrieval of full news articles produce reliable ground-truth labels and relevant external evidence.

What would settle it

Re-labeling the X-POSE posts with independent expert annotators and finding that TRENT loses its performance lead would falsify the central claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • External news articles can be integrated into fact-checking decisions more precisely than with prior fusion methods.
  • Parallel cross-attention streams capture distinct post-evidence relations at the same time.
  • Explicit entailment and contradiction modeling improves detection of nuanced claims.
  • The X-POSE benchmark exposes weaknesses in models trained only on synthetic data.
  • The gains appear against both specialized fact-checkers and general vision-language models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The three-stream design could be tested on other tasks that combine social media with external documents, such as video claim verification.
  • VLM-based retrieval may introduce its own biases that future evidence sources would need to offset.
  • Community-scale labels enable larger benchmarks but periodic expert audits could be required to maintain quality.
  • Adding more than three streams or additional evidence types remains an open extension not addressed here.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces X-POSE, a benchmark of real-world multimodal posts from X (formerly Twitter) with community annotations and full-length news articles retrieved via VLM-optimized search, and proposes TRENT, a model that performs evidence triangulation via three parallel cross-attention streams and a relational fusion mechanism explicitly modeling entailment and contradiction. It claims that TRENT consistently outperforms state-of-the-art specialized models and commercial VLMs on this benchmark, with code, prompts, and data released publicly.

Significance. If the central empirical claims hold after addressing benchmark validation, the work would be significant for shifting MFC research toward in-the-wild data and architectures that explicitly triangulate external evidence relations rather than intra-modality checks or unconstrained fusion. The public release of code, prompt templates, and the dataset is a clear strength supporting reproducibility.

major comments (2)
  1. [§3] §3 (X-POSE construction): The benchmark relies on community annotations as ground truth and VLM-retrieved articles for entailment/contradiction signals without reported inter-annotator agreement, expert validation, or ablation on label quality. This is load-bearing for the outperformance claim because systematic noise in labels or irrelevant evidence could produce artifactual gains unrelated to the three-stream triangulation.
  2. [§5] §5 (Experiments and results): The abstract states 'extensive evaluations' and 'consistent outperformance,' yet the manuscript must include quantitative sensitivity analysis to label noise or retrieval quality to demonstrate that reported gains (e.g., over baselines) are attributable to the relational fusion rather than benchmark construction choices.
minor comments (2)
  1. [Figure 2] Figure 1 or 2 (model diagram): The three parallel cross-attention streams would be clearer with explicit arrows or labels distinguishing the entailment versus contradiction pathways in the relational fusion block.
  2. [§4.2] Notation in §4.2: The description of the relational fusion mechanism uses 'entailment score' and 'contradiction score' without a clear equation reference; adding an explicit formula (e.g., Eq. (X)) would improve precision.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback emphasizing benchmark validation and the need to isolate model contributions from data artifacts. We address each major comment below and will revise the manuscript to incorporate additional validation and analysis.

read point-by-point responses

  1. Referee: [§3] §3 (X-POSE construction): The benchmark relies on community annotations as ground truth and VLM-retrieved articles for entailment/contradiction signals without reported inter-annotator agreement, expert validation, or ablation on label quality. This is load-bearing for the outperformance claim because systematic noise in labels or irrelevant evidence could produce artifactual gains unrelated to the three-stream triangulation.

    Authors: Community annotations on X are inherently consensus-driven rather than produced by multiple independent annotators, which precludes standard IAA computation; we will explicitly note this limitation in the revised §3. We agree that expert validation and label-quality ablation are valuable and will add both: (1) expert review of a random 10% subset with agreement statistics, and (2) an ablation that injects controlled label noise and measures degradation in TRENT versus baselines. For retrieval quality we will report precision@K of the VLM-optimized search on a held-out set. revision: yes

  2. Referee: [§5] §5 (Experiments and results): The abstract states 'extensive evaluations' and 'consistent outperformance,' yet the manuscript must include quantitative sensitivity analysis to label noise or retrieval quality to demonstrate that reported gains (e.g., over baselines) are attributable to the relational fusion rather than benchmark construction choices.

    Authors: We accept that the current experiments do not isolate the contribution of relational fusion from potential benchmark artifacts. In the revision we will add a dedicated sensitivity subsection in §5 that (a) varies label-flip rates from 0–30% and plots accuracy curves for TRENT and all baselines, and (b) substitutes progressively noisier or lower-ranked evidence articles and reports the resulting performance delta. These results will be used to argue that the observed gains remain attributable to the three-stream entailment/contradiction modeling. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper presents an empirical contribution: a new benchmark (X-POSE) constructed from community annotations and VLM-optimized retrieval, plus a model (TRENT) using cross-attention streams and relational fusion. No mathematical derivations, equations, or parameter-fitting steps are described that reduce by construction to the inputs or to self-citations. The central claims rest on held-out empirical evaluations rather than tautological reductions, satisfying the default expectation of a non-circular empirical proposal.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard machine-learning assumptions plus domain-specific choices about data quality; no new physical entities or ad-hoc constants are introduced.

axioms (2)
  • domain assumption Community annotations on X posts constitute reliable ground-truth labels for multimodal factuality
    The benchmark construction and all reported performance numbers depend on these labels being accurate.
  • domain assumption VLM-optimized search retrieves articles that are relevant external evidence for the claims in the posts
    The evidence-triangulation mechanism and superiority claims presuppose that the retrieved articles supply useful entailment/contradiction signals.

pith-pipeline@v0.9.1-grok · 5748 in / 1405 out tokens · 42967 ms · 2026-07-01T03:03:22.639676+00:00 · methodology

0 comments
read the original abstract

The proliferation of multimedia content on social platforms has fueled multimodal misinformation, where images are used to reinforce false claims. Consequently, Multimodal Fact-Checking (MFC) has emerged as an increasingly important research area. However, current progress is hindered by a reliance on synthetic training data and curated benchmarks that fail to capture the complexity of in-the-wild data. Furthermore, existing detection models rely on restricted intra-modality consistency or unconstrained all-to-all fusion, failing to capture nuanced relations between posts and external evidence. To address these limitations, we introduce X-POSE, a benchmark of real-world, community-annotated multimodal posts from X (formerly Twitter), augmented with full-length news articles retrieved via VLM-optimized search. Additionally, we propose TRENT, a novel MFC model that performs evidence triangulation using three parallel cross-attention streams alongside a relational fusion mechanism that explicitly models entailment and contradiction. Extensive evaluations demonstrate that TRENT consistently outperforms state-of-the-art specialized models and commercial VLMs. The code, prompt templates, and dataset are available at https://github.com/stevejpapad/evidence-triangulation

Figures

Figures reproduced from arXiv: 2606.31367 by Christos Koutlis, Panagiotis C. Petrantonakis, Stefanos-Iordanis Papadopoulos, Symeon Papadopoulos, Zacharias Chrysidis.

Figure 1
Figure 1. Figure 1: (a) Proposed evidence collection pipeline: A VLM reformulates a multimodal post [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of TRENT: Input text (T), image (I), and evidence excerpts (E) are projected to embeddings (zt, zi , ze) and processed by three parallel cross-attention streams. The resulting contextualized representations (ct→e, ct→i , ci↔e) are integrated via relational fusion R(·) to capture entailment and contradiction for classification. First, raw embeddings are projected into a shared latent space of dimen… view at source ↗
Figure 3
Figure 3. Figure 3: (a) Performance of TRENT and baselines across helpfulness thresholds [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative examples of low consensus and evidence credibility. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

56 extracted references · 56 canonical work pages · 3 internal anchors

  1. [1]

    How non-probative photos shape belief.–Stephan Lewandowsky, Cognitive Science, page 90, 2020

    Eryn J Newman and Lynn Zhang. How non-probative photos shape belief.–Stephan Lewandowsky, Cognitive Science, page 90, 2020

    work page 2020

  2. [2]

    Is a picture worth a thousand words? an empirical study of image content and social media engagement.Journal of marketing research, 57(1):1–19, 2020

    Yiyi Li and Ying Xie. Is a picture worth a thousand words? an empirical study of image content and social media engagement.Journal of marketing research, 57(1):1–19, 2020

    work page 2020

  3. [3]

    Multimodal automated fact-checking: A survey

    Mubashara Akhtar, Michael Schlichtkrull, Zhijiang Guo, Oana Cocarascu, Elena Simperl, and Andreas Vlachos. Multimodal automated fact-checking: A survey. InFindings of the Association for Computational Linguistics: EMNLP 2023, pages 5430–5448, 2023

    work page 2023

  4. [4]

    Factify: A multi-modal fact verification dataset

    Shreyash Mishra, S Suryavardan, Amrit Bhaskar, Parul Chopra, Aishwarya N Reganti, Parth Patwa, Amitava Das, Tanmoy Chakraborty, Amit P Sheth, Asif Ekbal, et al. Factify: A multi-modal fact verification dataset. In DE-FACTIFY@ AAAI, page np, 2022

    work page 2022

  5. [5]

    Newsclippings: Automatic generation of out-of-context multi- modal media

    Grace Luo, Trevor Darrell, and Anna Rohrbach. Newsclippings: Automatic generation of out-of-context multi- modal media. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 6801–6817, 2021

    work page 2021

  6. [6]

    Cosmos: Catching out-of-context image misuse using self-supervised learning.Proceedings of the AAAI Conference on Artificial Intelligence, 37(12):14084–14092, Jun

    Shivangi Aneja, Chris Bregler, and Matthias Niessner. Cosmos: Catching out-of-context image misuse using self-supervised learning.Proceedings of the AAAI Conference on Artificial Intelligence, 37(12):14084–14092, Jun. 2023. 11 S.I. Papadopoulos et al

    work page 2023

  7. [7]

    Twitter-comms: Detecting climate, covid, and military multimodal misinformation

    Giscard Biamby, Grace Luo, Trevor Darrell, and Anna Rohrbach. Twitter-comms: Detecting climate, covid, and military multimodal misinformation. InProceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1530–1549, 2022

    work page 2022

  8. [8]

    Deep multimodal image-repurposing detection

    Ekraam Sabir, Wael AbdAlmageed, Yue Wu, and Prem Natarajan. Deep multimodal image-repurposing detection. InProceedings of the 26th ACM international conference on Multimedia, pages 1337–1345, 2018

    work page 2018

  9. [9]

    Multimodal analytics for real-world news using measures of cross-modal entity consistency

    Eric Müller-Budack, Jonas Theiner, Sebastian Diering, Maximilian Idahl, and Ralph Ewerth. Multimodal analytics for real-world news using measures of cross-modal entity consistency. InProceedings of the 2020 International Conference on Multimedia Retrieval, pages 16–25, 2020

    work page 2020

  10. [10]

    Verite: a robust benchmark for multimodal misinformation detection accounting for unimodal bias.International Journal of Multimedia Information Retrieval, 13(1):4, 2024

    Stefanos-Iordanis Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and Panagiotis C Petrantonakis. Verite: a robust benchmark for multimodal misinformation detection accounting for unimodal bias.International Journal of Multimedia Information Retrieval, 13(1):4, 2024

    work page 2024

  11. [11]

    Averimatec: A dataset for automatic verification of image-text claims with evidence from the web.Advances in Neural Information Processing Systems, 38, 2026

    Rui Cao, Zifeng Ding, Zhijiang Guo, Michael Schlichtkrull, and Andreas Vlachos. Averimatec: A dataset for automatic verification of image-text claims with evidence from the web.Advances in Neural Information Processing Systems, 38, 2026

    work page 2026

  12. [12]

    VeriTaS: The First Dynamic Benchmark for Multimodal Automated Fact-Checking

    Mark Rothermel, Marcus Kornmann, Marcus Rohrbach, and Anna Rohrbach. Veritas: The first dynamic benchmark for multimodal automated fact-checking.arXiv preprint arXiv:2601.08611, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  13. [13]

    arXiv preprint arXiv:2508.09999 , year=

    Yuzhuo Xiao, Zeyu Han, Yuhan Wang, and Huaizu Jiang. Xfacta: Contemporary, real-world dataset and evaluation for multimodal misinformation detection with multimodal llms.arXiv preprint arXiv:2508.09999, 2025

    work page arXiv 2025

  14. [14]

    Open-domain, content-based, multi-modal fact-checking of out-of-context images via online resources

    Sahar Abdelnabi, Rakibul Hasan, and Mario Fritz. Open-domain, content-based, multi-modal fact-checking of out-of-context images via online resources. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14940–14949, 2022

    work page 2022

  15. [15]

    Ecenet: explainable and context-enhanced network for muti-modal fact verification

    Fanrui Zhang, Jiawei Liu, Qiang Zhang, Esther Sun, Jingyi Xie, and Zheng-Jun Zha. Ecenet: explainable and context-enhanced network for muti-modal fact verification. InProceedings of the 31st ACM International Conference on Multimedia, pages 1231–1240, 2023

    work page 2023

  16. [16]

    Red- dot: Multimodal fact-checking via relevant evidence detection.IEEE Transactions on Computational Social Systems, 2025

    Stefanos-Iordanis Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and Panagiotis C Petrantonakis. Red- dot: Multimodal fact-checking via relevant evidence detection.IEEE Transactions on Computational Social Systems, 2025

    work page 2025

  17. [17]

    Petrantonakis

    Stefanos-Iordanis Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and Panagiotis C. Petrantonakis. Similarity over factuality: Are we making progress on multimodal out-of-context misinformation detection? In Proceedings of the Winter Conference on Applications of Computer Vision (WACV), pages 5570–5579, February 2025

    work page 2025

  18. [18]

    E2lvlm:evidence-enhanced large vision-language model for multimodal out-of-context misinformation detection, 2025

    Junjie Wu, Yumeng Fu, Nan Yu, and Guohong Fu. E2lvlm:evidence-enhanced large vision-language model for multimodal out-of-context misinformation detection, 2025

    work page 2025

  19. [19]

    Defame: Dynamic evidence-based fact-checking with multimodal experts

    Tobias Braun, Mark Rothermel, Marcus Rohrbach, and Anna Rohrbach. Defame: Dynamic evidence-based fact-checking with multimodal experts. InInternational Conference on Machine Learning, pages 5383–5417. PMLR, 2025

    work page 2025

  20. [20]

    Fact-checking meets fauxtography: Verifying claims about images

    Dimitrina Zlatkova, Preslav Nakov, and Ivan Koychev. Fact-checking meets fauxtography: Verifying claims about images. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 2099–2108, 2019

    work page 2019

  21. [21]

    Novel visual and statistical image features for microblogs news verification.IEEE transactions on multimedia, 19(3):598–608, 2016

    Zhiwei Jin, Juan Cao, Yongdong Zhang, Jianshe Zhou, and Qi Tian. Novel visual and statistical image features for microblogs news verification.IEEE transactions on multimedia, 19(3):598–608, 2016

    work page 2016

  22. [22]

    Verifying multimedia use at mediaeval 2015

    Christina Boididou, Katerina Andreadou, Symeon Papadopoulos, Duc Tien Dang Nguyen, Giulia Boato, Michael Riegler, Yiannis Kompatsiaris, et al. Verifying multimedia use at mediaeval 2015. InMediaEval 2015, volume

    work page 2015

  23. [23]

    Verifying information with multimedia content on twitter: a comparative study of automated approaches.Multimedia tools and applications, 77:15545–15571, 2018

    Christina Boididou, Stuart E Middleton, Zhiwei Jin, Symeon Papadopoulos, Duc-Tien Dang-Nguyen, Giulia Boato, and Yiannis Kompatsiaris. Verifying information with multimedia content on twitter: a comparative study of automated approaches.Multimedia tools and applications, 77:15545–15571, 2018

    work page 2018

  24. [24]

    Syn- thetic misinformers: Generating and combating multimodal misinformation

    Stefanos-Iordanis Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and Panagiotis Petrantonakis. Syn- thetic misinformers: Generating and combating multimodal misinformation. InProceedings of the 2nd ACM International Workshop on Multimedia AI against Disinformation, pages 36–44, 2023

    work page 2023

  25. [25]

    Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and Panagiotis C

    Stylianos I. Papadopoulos, Christos Koutlis, Symeon Papadopoulos, and Panagiotis C. Petrantonakis. Latent multimodal reconstruction for misinformation detection.arXiv preprint arXiv:2504.06010, 2025. 12 S.I. Papadopoulos et al

    work page arXiv 2025

  26. [26]

    Fakeddit: A new multimodal benchmark dataset for fine-grained fake news detection

    Kai Nakamura, Sharon Levy, and William Yang Wang. Fakeddit: A new multimodal benchmark dataset for fine-grained fake news detection. InProceedings of the Twelfth Language Resources and Evaluation Conference, pages 6149–6157, 2020

    work page 2020

  27. [27]

    Missing counter-evidence renders nlp fact-checking unrealistic for misinformation

    Max Glockner, Yufang Hou, and Iryna Gurevych. Missing counter-evidence renders nlp fact-checking unrealistic for misinformation. InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 5916–5936, 2022

    work page 2022

  28. [28]

    Findings of factify 2: multimodal fake news detection.arXiv preprint arXiv:2307.10475, 2023

    S Suryavardan, Shreyash Mishra, Megha Chakraborty, Parth Patwa, Anku Rani, Aman Chadha, Aishwarya Reganti, Amitava Das, Amit Sheth, Manoj Chinnakotla, et al. Findings of factify 2: multimodal fake news detection.arXiv preprint arXiv:2307.10475, 2023

    work page arXiv 2023

  29. [29]

    Credible, unreliable or leaked?: Evidence verification for enhanced automated fact-checking

    Zacharias Chrysidis, Stefanos-Iordanis Papadopoulos, Symeon Papadopoulos, and Panagiotis Petrantonakis. Credible, unreliable or leaked?: Evidence verification for enhanced automated fact-checking. InProceedings of the 3rd ACM International Workshop on Multimedia AI against Disinformation, pages 73–81, 2024

    work page 2024

  30. [30]

    M4FC: a Multimodal, Multilingual, Multicultural, Multitask Real-World Fact-Checking Dataset

    Jiahui Geng, Jonathan Tonglet, and Iryna Gurevych. M4fc: a multimodal, multilingual, multicultural, multitask real-world fact-checking dataset.arXiv preprint arXiv:2510.23508, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  31. [31]

    Mmfakebench: A mixed-source multimodal misinformation detection benchmark for lvlms

    Xuannan Liu, Zekun Li, Pei Li, Huaibo Huang, Shuhan Xia, Xing Cui, Linzhi Huang, Weihong Deng, and Zhaofeng He. Mmfakebench: A mixed-source multimodal misinformation detection benchmark for lvlms. In International Conference on Learning Representations, volume 2025, pages 86327–86352, 2025

    work page 2025

  32. [32]

    Community notes reduce engagement with and diffusion of false information online.Proceedings of the National Academy of Sciences, 122(38):e2503413122, 2025

    Isaac Slaughter, Axel Peytavin, Johan Ugander, and Martin Saveski. Community notes reduce engagement with and diffusion of false information online.Proceedings of the National Academy of Sciences, 122(38):e2503413122, 2025

    work page 2025

  33. [33]

    Can crowdchecking curb misinformation? evidence from community notes.Information Systems Research, 2025

    Yang Gao, Maggie Mengqing Zhang, and Huaxia Rui. Can crowdchecking curb misinformation? evidence from community notes.Information Systems Research, 2025

    work page 2025

  34. [34]

    Who checks the checkers? exploring source credibility in twitter’s community notes.Journal of Computational Social Science, 9(1):24, 2026

    Uku Kangur, Roshni Chakraborty, and Rajesh Sharma. Who checks the checkers? exploring source credibility in twitter’s community notes.Journal of Computational Social Science, 9(1):24, 2026

    work page 2026

  35. [35]

    On the limitations of llm-synthesized social media misinformation moderation

    Sahajpreet Singh, Jiaying Wu, Svetlana Churina, and Kokil Jaidka. On the limitations of llm-synthesized social media misinformation moderation. InI Can’t Believe It’s Not Better: Challenges in Applied Deep Learning

    work page

  36. [36]

    Hellofresh: Llm evalutions on streams of real-world human editorial actions across x community notes and wikipedia edits

    Tim Franzmeyer, Aleksandar Shtedritski, Samuel Albanie, Philip Torr, Joao F Henriques, and Jakob Foerster. Hellofresh: Llm evalutions on streams of real-world human editorial actions across x community notes and wikipedia edits. InFindings of the Association for Computational Linguistics ACL 2024, pages 12702–12716, 2024

    work page 2024

  37. [37]

    Communitynotes: A dataset for exploring the helpfulness of fact-checking explanations.arXiv preprint arXiv:2510.24810, 2025

    Rui Xing, Preslav Nakov, Timothy Baldwin, and Jey Han Lau. Communitynotes: A dataset for exploring the helpfulness of fact-checking explanations.arXiv preprint arXiv:2510.24810, 2025

    work page arXiv 2025

  38. [38]

    Beyond the Crowd: LLM-Augmented Community Notes for Governing Health Misinformation

    Jiaying Wu, Zihang Fu, Haonan Wang, Fanxiao Li, Jiafeng Guo, Preslav Nakov, and Min-Yen Kan. Beyond the crowd: Llm-augmented community notes for governing health misinformation.arXiv preprint arXiv:2510.11423, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  39. [39]

    ‘humor, art, or misinformation?”: A multimodal dataset for intent-aware synthetic image detection

    Anastasios Skoularikis, Stefanos-Iordanis Papadopoulos, Symeon Papadopoulos, and Panagiotis C Petrantonakis. ‘humor, art, or misinformation?”: A multimodal dataset for intent-aware synthetic image detection. InProceedings of the 2nd International Workshop on Diffusion of Harmful Content on Online Web, pages 95–104, 2025

    work page 2025

  40. [40]

    Support or refute: Analyzing the stance of evidence to detect out-of-context mis-and disinformation

    Xin Yuan, Jie Guo, Weidong Qiu, Zheng Huang, and Shujun Li. Support or refute: Analyzing the stance of evidence to detect out-of-context mis-and disinformation. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 4268–4280, 2023

    work page 2023

  41. [41]

    Multimodal misinformation detection using large vision-language models

    Sahar Tahmasebi, Eric Müller-Budack, and Ralph Ewerth. Multimodal misinformation detection using large vision-language models. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM ’24, page 2189–2199, New York, NY , USA, 2024. Association for Computing Machinery

    work page 2024

  42. [42]

    Sniffer: Multimodal large language model for explainable out-of-context misinformation detection

    Peng Qi, Zehong Yan, Wynne Hsu, and Mong Li Lee. Sniffer: Multimodal large language model for explainable out-of-context misinformation detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13052–13062, 2024

    work page 2024

  43. [43]

    Mad-sherlock: Multi-agent debates for out-of-context misinformation detection

    Kumud Lakara, Juil Sock, Christian Rupprecht, Philip Torr, John Collomosse, and Christian Schroeder de Witt. Mad-sherlock: Multi-agent debates for out-of-context misinformation detection. 2024

    work page 2024

  44. [44]

    Trust-vl: An explainable news assistant for general multimodal misinformation detection

    Zehong Yan, Peng Qi, Wynne Hsu, and Mong-Li Lee. Trust-vl: An explainable news assistant for general multimodal misinformation detection. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 5588–5604, 2025. 13 S.I. Papadopoulos et al

    work page 2025

  45. [45]

    The synthetic media shift: Tracking the rise, virality, and detectability of ai-generated multimodal misinformation

    Zacharias Chrysidis, Stefanos-Iordanis Papadopoulos, and Symeon Papadopoulos. The synthetic media shift: Tracking the rise, virality, and detectability of ai-generated multimodal misinformation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8626–8635, 2026

    work page 2026

  46. [46]

    A survey on deep learning-based image forgery detection.Pattern Recognition, 144:109778, 2023

    Fatemeh Zare Mehrjardi, Ali Mohammad Latif, Mohsen Sardari Zarchi, and Razieh Sheikhpour. A survey on deep learning-based image forgery detection.Pattern Recognition, 144:109778, 2023

    work page 2023

  47. [47]

    A survey of defenses against ai-generated visual media: Detection, disruption, and authentication.ACM Computing Surveys, 58(5):1–35, 2025

    Jingyi Deng, Chenhao Lin, Zhengyu Zhao, Shuai Liu, Zhe Peng, Qian Wang, and Chao Shen. A survey of defenses against ai-generated visual media: Detection, disruption, and authentication.ACM Computing Surveys, 58(5):1–35, 2025

    work page 2025

  48. [48]

    From chaos to clarity: Claim normalization to empower fact-checking

    Megha Sundriyal, Tanmoy Chakraborty, and Preslav Nakov. From chaos to clarity: Claim normalization to empower fact-checking. InFindings of the Association for Computational Linguistics: EMNLP 2023, pages 6594–6609, 2023

    work page 2023

  49. [49]

    Show me the work: Fact-checkers’ requirements for explainable automated fact-checking

    Greta Warren, Irina Shklovski, and Isabelle Augenstein. Show me the work: Fact-checkers’ requirements for explainable automated fact-checking. InProceedings of the 2025 CHI Conference on Human Factors in Computing Systems, pages 1–21, 2025

    work page 2025

  50. [50]

    True or false: Studying the work practices of professional fact-checkers.Proceedings of the ACM on Human-Computer Interaction, 6(CSCW1):1–44, 2022

    Nicholas Micallef, Vivienne Armacost, Nasir Memon, and Sameer Patil. True or false: Studying the work practices of professional fact-checkers.Proceedings of the ACM on Human-Computer Interaction, 6(CSCW1):1–44, 2022

    work page 2022

  51. [51]

    Scientific fact-checking: A survey of resources and approaches

    Juraj Vladika and Florian Matthes. Scientific fact-checking: A survey of resources and approaches. InFindings of the Association for Computational Linguistics: ACL 2023, pages 6215–6230, 2023

    work page 2023

  52. [52]

    Minicpm-v: A gpt-4v level mllm on your phone.Nat Commun 16, 5509 (2025), 2025

    Yuan Yao, Tianyu Yu, Ao Zhang, Chongyi Wang, Junbo Cui, Hongji Zhu, Tianchi Cai, Haoyu Li, Weilin Zhao, Zhihui He, et al. Minicpm-v: A gpt-4v level mllm on your phone.Nat Commun 16, 5509 (2025), 2025

    work page 2025

  53. [53]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational conference on machine learning, pages 8748–8763. PMLR, 2021

    work page 2021

  54. [54]

    Supervised learning of universal sentence representations from natural language inference data

    Alexis Conneau, Douwe Kiela, Holger Schwenk, Loïc Barrault, and Antoine Bordes. Supervised learning of universal sentence representations from natural language inference data. InProceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 670–680, 2017

    work page 2017

  55. [55]

    External reliable information-enhanced multimodal contrastive learning for fake news detection

    Biwei Cao, Qihang Wu, Jiuxin Cao, Bo Liu, and Jie Gui. External reliable information-enhanced multimodal contrastive learning for fake news detection. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 31–39, 2025

    work page 2025

  56. [56]

    A survey on automatic credibility assessment using textual credibility signals in the era of large language models.ACM Transactions on Intelligent Systems and Technology, 2025

    Ivan Srba, Olesya Razuvayevskaya, João A Leite, Robert Moro, Ipek Baris Schlicht, Sara Tonelli, Fran- cisco Moreno García, Santiago Barrio Lottmann, Denis Teyssou, Valentin Porcellini, et al. A survey on automatic credibility assessment using textual credibility signals in the era of large language models.ACM Transactions on Intelligent Systems and Techno...

    work page 2025