Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Verifying Cross-modal Entity Consistency in News using Vision-language Models

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Large vision-language models can verify, entity by entity, whether the people, places, and events named in a news article actually appear in its photo, and adding web-crawled evidence images improves accuracy for persons and events.

desk verdict Useful task formulation and prompting ideas, but the evaluation is under-specified enough that the headline accuracy numbers are not yet interpretable. read the letter →

arxiv 2501.11403 v2 pith:EEXJWPXR submitted 2025-01-20 cs.CL cs.IRcs.MM

Sahar Tahmasebi , David Ernst , Eric Müller-Budack , Ralph Ewerth This is my paper · ORCID
classification cs.CLcs.IRcs.MM
keywords cross-modalentityconsistencyout-of-contextdisinformationdetectionlargevision-languagemodelsnewsanalyticsverificationzero-shotpromptingevidenceimagesmultimodalmisinformation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that large vision-language models (LVLMs) can automate cross-modal entity verification in news: for each person, location, or event named in an article, the model decides whether that entity is actually visible in the accompanying photo. This matters because out-of-context disinformation often pairs a real image with text that names the wrong entities, and existing methods judge whole-document consistency instead of pinpointing which entity is misleading. The authors propose a zero-shot prompting framework called LVLM4CEC, with optional web-crawled evidence images for each entity, and they release manually labeled subsets of three news datasets for evaluation. Their experiments report that evidence images improve accuracy for persons and events, and that the approach outperforms a CNN baseline on location and event verification at document level.

What carries the argument

The mechanism is a per-entity yes/no question template posed to an LVLM, combined with an evidence-image aggregation strategy. Entity extraction uses spaCy and Wikifier to find persons, locations, and events; the model's output is restricted to yes/no token classes to avoid free-form hallucination; and evidence images crawled from Google, Bing, and Wikidata are merged into a composite image (single-image LVLMs) or presented as a series (multi-image LVLMs), with majority voting producing the final decision. This machinery lets a zero-shot model ground each named entity in visual evidence rather than guessing from the news image alone.

What would settle it

Re-annotate a random sample of the TamperedNews-Ent and News400-Ent entities with at least two independent annotators and measure inter-annotator agreement; if agreement is low, or if the corrected labels move the reported accuracy gaps, the claims that evidence images help persons and events and that LVLMs beat the CNN baseline for locations and events would not stand. A simpler check is to download the released datasets and confirm that the annotated entity counts match Table 1.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that off-the-shelf LVLMs, asked simple yes/no visibility questions, can perform entity-level cross-modal verification without any task-specific training. When evidence images of an entity are added—either as a bordered composite image for single-image models or as an image series for multi-image models, with majority voting over the answers—accuracy rises for persons and events on TamperedNews-Ent and News400-Ent. For locations, evidence images do not help on average, because search-engine images for coarse place names often depict different landmarks or a different entity altogether. At the document level, the LVLM approach beats the CNN baseline for locations and events, while the baseline retains an advantage on person identification.

Load-bearing premise

The manual ground-truth labels that score every model are treated as reliable, yet they come from a single annotator with no reported inter-annotator agreement, and the paper's dataset description conflicts with its own table of statistics; if the labels are noisy, every accuracy figure and every baseline comparison rests on shaky ground.

Editorial extensions

If this is right

  • Fact-checking pipelines could use the framework to flag the specific entity a news photo contradicts, rather than only labeling an article as consistent or inconsistent.
  • Because the method works zero-shot, it can be applied to new languages and news domains without collecting training labels for each entity type.
  • Evidence images are a cheap accuracy lever for persons and events, so improving evidence retrieval is a direct route to better verification.
  • The document-level result indicates that general-purpose LVLMs can replace specialized CNN components for location and event verification in this setting.

Reading between the lines

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

  • My inference: the same prompts should transfer across news domains, so accuracy on a different-language news set would be a clean test of whether the zero-shot gains are language-dependent.
  • My inference: the location failure is likely evidence-retrieval noise, so swapping search-engine images for a curated set of canonical landmarks per city could recover location accuracy without changing the model.
  • My inference: combining LVLM votes with a face-recognition model such as FaceNet could push person verification above both methods, since the baseline's face-specific embeddings already outperform the LVLMs on persons.
  • My inference: if the dataset statistics do not match the released files, the reported accuracy differences should be re-estimated on a consistently annotated sample before relying on them.
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 / 6 minor

Summary. The paper proposes LVLM4CEC, a zero-shot pipeline for verifying cross-modal entity consistency in news, i.e., determining whether a person, location, or event mentioned in the text is visible in the accompanying news image. The system uses large vision-language models, supports verification with and without web-crawled evidence images (via composite-image and image-series strategies for single- and multi-image models), and is evaluated on three datasets extended with manual ground-truth annotations (TamperedNews-Ent, News400-Ent, MMG-Ent). The headline empirical claims are that evidence images improve accuracy for persons and events, and that the LVLM approach outperforms a CNN-based baseline for location and event verification in documents.

Significance. The task itself is well-motivated and fills a genuine gap: prior out-of-context detection operates at the whole-document level, while entity-level cross-modal verification can provide more actionable signals for fact-checking. The paper is the first to apply LVLMs to this entity-level task, and the public release of source code and extended datasets is a concrete contribution. If the reported evaluation were trustworthy, the findings would be useful to the multimedia verification community. However, the current evaluation protocol has serious gaps—prompt selection on the test set, unknown class balance and test-set sizes, single-annotator ground truth without agreement statistics, and an underspecified baseline comparison—that prevent the quantitative claims from being assessed as stated. The strengths of the method (multiple LVLMs, two evidence-integration strategies, zero-shot framing) are clear, but the evidence supporting the central claims needs to be re-derived under a sounder evaluation protocol.

major comments (4)
  1. [Section 4.2, Table 2 and Table 3]
  2. [Section 4.1 vs Table 1 and Table 3]
  3. [Section 4.1, annotation process]
  4. [Section 4.3, baseline comparison]
minor comments (6)
  1. [Section 3.3]
  2. [Figure 1 caption]
  3. [Section 4.2]
  4. [Section 4.1, implementation details]
  5. [Section 4.1, models]
  6. [Table 3 and Section 4.1]

Circularity Check

1 steps flagged · score 4.0 of 10

Test-set prompt selection makes the reported zero-shot accuracies partially self-confirming; no other load-bearing circularity.

  1. fitted input called prediction [Section 4.2 (Prompt Impact) and Section 4.1 (Implementation Details)]
    "Both InstructBLIP [7] and LLaVA[17] performed better with visibility questions for persons and events entities, while they excelled with consistency questions for location entities. For each model, we used the best-performing prompt in further experiments."

    The prompt templates are compared on TamperedNews-Ent, and the best template per model is then used in further experiments on the same test dataset. Because the prompt is the task-specific instruction for the zero-shot LVLM and its choice is decided by accuracy on the evaluation labels, the numbers reported in Tables 2 and 3 are effectively a post-hoc maximum over the tested templates on the same data. This is a fitted input (prompt selection) being reported as an independent zero-shot prediction, so the headline zero-shot performance is partly constructed by the evaluation protocol rather than derived from the model alone.

full rationale

The central derivation is empirical and not circular in the strict sense of X defined via Y: the LVLMs are externally pre-trained, the evidence images are web-crawled references, and the ground truth is manual annotation. The baseline comparison, the datasets, and the evidence-image protocol come from the same author group's earlier work, but that lineage is not load-bearing because the reported numbers do not reduce to those citations. The one genuine circular element is prompt selection: the best prompt is chosen on the test set and then the same test set is used to report zero-shot accuracy, making the figures an optimistic selected maximum. This weakens the quantitative claims but does not make the central finding equivalent to its inputs. The dataset-size inconsistency and single-annotator labels are serious data-quality and reproducibility risks, but they are correctness issues rather than circularity. Therefore the score is 4.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new theoretical constructs or entities. Its central results rest on three empirical pillars: (1) the quality of the newly created manual annotations, which are not validated for inter-annotator agreement; (2) the quality of web-sourced evidence images inherited from prior work; and (3) the test-set selection of prompt templates, which acts as an effective fitted parameter. These should be accounted for when interpreting the accuracy numbers.

free parameters (3)
  • prompt template per model and entity type = e.g., 'Is <type> <name> shown in the image?' vs 'Is the content of the image consistent with the <type> <name>?'
    Section 4.2: 'For each model, we used the best-performing prompt in further experiments', meaning the prompt was chosen based on test-set accuracy on the same datasets used for final evaluation.
  • baseline CMS threshold for yes/no classification = 0.65
    Section 4.3 caption of Figure 3: 'we classify CMS values above 0.65 as the Yes class and values below 0.65 as the No class'; a hand-chosen threshold affecting the baseline comparison.
  • number of evidence images (n) = max 20
    Section 3.3: evidence images 'maximum of 20 images per entity', inherited from the prior dataset; no sensitivity analysis is provided.
assumptions (4)
  • domain assumption spaCy and Wikifier entity extraction is correct and complete
    Section 3.1: 'we use the approach suggested by Müller-Budack et al. [22] and employ spaCy and Wikifier to identify persons, locations, and events'; errors in entity extraction propagate to all verification questions.
  • domain assumption The manually annotated ground truth is accurate
    Section 4.1: 'the cross-modal occurrence of each entity was manually verified by a human annotator. The annotator was allowed to use the web for the verification.' No inter-annotator agreement is reported.
  • domain assumption Evidence images from Google, Bing, and Wikidata depict the queried entity
    Section 3.3 and 4.2: evidence images are used for verification, but the paper itself notes failures such as 'Liverpool' retrieving Liverpool F.C. images, so this assumption is only partially true.
  • domain assumption LVLM probability-of-yes-token can serve as a cross-modal similarity score comparable to the baseline CMS
    Section 4.3: 'we use the probability of generating the token yes as the CMS for each entity' to enable document-level comparison with [22].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Verifying Cross-modal Entity Consistency in News using Vision-language Models." pith.science (2026). https://pith.science/paper/EEXJWPXR

@misc{pith2026250111403,
  author       = {Pith},
  title        = {Pith review of: Verifying Cross-modal Entity Consistency in News using Vision-language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EEXJWPXR}},
  note         = {Machine review of arXiv:2501.11403}
}
read the original abstract

The web has become a crucial source of information, but it is also used to spread disinformation, often conveyed through multiple modalities like images and text. The identification of inconsistent cross-modal information, in particular entities such as persons, locations, and events, is critical to detect disinformation. Previous works either identify out-of-context disinformation by assessing the consistency of images to the whole document, neglecting relations of individual entities, or focus on generic entities that are not relevant to news. So far, only few approaches have addressed the task of validating entity consistency between images and text in news. However, the potential of large vision-language models (LVLMs) has not been explored yet. In this paper, we propose an LVLM-based framework for verifying Cross-modal Entity Consistency~(LVLM4CEC), to assess whether persons, locations and events in news articles are consistent across both modalities. We suggest effective prompting strategies for LVLMs for entity verification that leverage reference images crawled from web. Moreover, we extend three existing datasets for the task of entity verification in news providing manual ground-truth data. Our results show the potential of LVLMs for automating cross-modal entity verification, showing improved accuracy in identifying persons and events when using evidence images. Moreover, our method outperforms a baseline for location and event verification in documents. The datasets and source code are available on GitHub at https://github.com/TIBHannover/LVLM4CEC.

Figures

Figures reproduced from arXiv: 2501.11403 by the authors.

Figure 1
Figure 1. Example of cross-modal entity verification. Image is replaced with similar one due to license restrictions. Original image is linked on the GitHub. spread deliberately), fake news, and other forms of harmful content [38]. Typi￾cally, harmful content is conveyed using different modalities such as images, text, and videos to convey information more efficiently and to attract attention. Ad￾ditionally, false or harmful … view at source ↗
Figure 2
Figure 2. Pipeline for entity consistency verification with (bottom) and without (top) using evidence images. The model assesses whether or not an entity e ∈ E is visible, i.e., shares a cross-modal relation, in the news image I. Green indicates valid relations, while red denotes invalid relations. 3.1 Entity Extraction To analyze cross-modal consistency, it is essential to first extract entities from the text. Several approa… view at source ↗
Figure 3
Figure 3. Entity verification with and without image evidence across models. Green text box indicates correct predictions; red text indicates incorrect ones. Green borders show visible entities; red borders show invisible ones. As the baseline only outputs a similarity score of Cross-modal Similarities (CMS), we classify CMS values above 0.65 as the ’Yes’ class and values below 0.65 as the ’No’ class. Images are replaced with… view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. COVE: COntext and VEracity prediction for out-of-context images

    cs.CL 2025-02 conditional novelty 5.0 of 10

    COVE predicts an image's true context before judging caption veracity, improving real-world out-of-context detection and providing a reusable context artifact for human verifiers.

Reference graph

Works this paper leans on

40 extracted references · 21 canonical work pages · cited by 1 Pith paper

  1. [1]

    In: IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022

    Abdelnabi, S., Hasan, R., Fritz, M.: Open-domain, content-based, multi-modal fact-checking of out-of-context images via online resources. In: IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022. pp. 14920–14929. IEEE (2022).https://doi.org/10. 1109/CVPR52688.2022.01452

    arXiv 2022

  2. [2]

    arXiv preprint abs/2311.07453 (2023)

    Akhtar, M., Subedi, N., Gupta, V., Tahmasebi, S., Cocarascu, O., Simperl, E.: Chartcheck: An evidence-based fact-checking dataset over real-world chart images. arXiv preprint abs/2311.07453 (2023). https://doi.org/10.48550/ ARXIV.2311.07453

    arXiv doi:10.48550/arxiv.2311.07453 2023

  3. [3]

    Ayoola, T., Tyagi, S., Fisher, J., Christodoulopoulos, C., Pierleoni, A.: Refined: An efficient zero-shot-capable approach to end-to-end entity linking. In: Conference of the North American Chapter of the Association for Computational Linguistics: Hu- man Language Technologies: Industry Track, NAACL 2022, Hybrid: Seattle, Wash- ington, USA + Online, July ...

    work page doi:10.18653/v1/2022.naacl-industry 2022

  4. [4]

    Routledge (2014)

    Bateman, J.: Text and image: A critical introduction to the visual/verbal divide. Routledge (2014)

    work page 2014

  5. [5]

    Proceedings of SiKDD472 (2017)

    Brank, J., Leban, G., Grobelnik, M.: Annotating documents with relevant wikipedia concepts. Proceedings of SiKDD472 (2017)

    work page 2017

  6. [6]

    Frontiers in Artificial Intelligence6 (2023)

    Cheema, G.S., Hakimov, S., Müller-Budack, E., Otto, C., Bateman, J.A., Ewerth, R.: Understanding image-text relations and news values for multimodal news anal- ysis. Frontiers in Artificial Intelligence6 (2023). https://doi.org/10.3389/FRAI. 2023.1125533

    arXiv 2023

  7. [7]

    Tahmasebi et al

    Dai, W., Li, J., Li, D., Tiong, A.M.H., Zhao, J., Wang, W., Li, B., Fung, P., Hoi, S.C.H.: Instructblip: Towards general-purpose vision-language models with 14 S. Tahmasebi et al. instruction tuning. In: Annual Conference on Neural Information Processing Sys- tems, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023 (2023), https://openreview.net/f...

    work page 2023

  8. [8]

    The Conversation 14, 1 (2020)

    Fazio, L.: Out-of-context photos are a powerful low-tech form of misinformation. The Conversation 14, 1 (2020)

    work page 2020

Show all 40 references
  1. [9]

    In: International Conference on Multimedia Retrieval, ICMR 2017, Bucharest, Romania, June 6-9, 2017

    Henning, C.A., Ewerth, R.: Estimating the information gap between textual and visual representations. In: International Conference on Multimedia Retrieval, ICMR 2017, Bucharest, Romania, June 6-9, 2017. pp. 14–22. ACM (2017).https: //doi.org/10.1145/3078971.3078991

    2017

  2. [10]

    To appear 7(1), 411–420 (2017)

    Honnibal, M., Montani, I.: spaCy 2: Natural language understanding with Bloom embeddings, convolutional neural networks and incremental parsing. To appear 7(1), 411–420 (2017)

    2017

  3. [11]

    In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2019, Long Beach, CA, USA, June 16-20, 2019

    Jaiswal, A., Wu, Y., AbdAlmageed, W., Masi, I., Natarajan, P.: AIRD: adversar- ial learning framework for image repurposing detection. In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2019, Long Beach, CA, USA, June 16-20, 2019. pp. 11330–11339. Computer Vi...

    2019

  4. [12]

    arXiv preprintabs/2405.01483 (2024)

    Jiang, D., He, X., Zeng, H., Wei, C., Ku, M., Liu, Q., Chen, W.: MANTIS: in- terleaved multi-image instruction tuning. arXiv preprintabs/2405.01483 (2024). https://doi.org/10.48550/ARXIV.2405.01483

    2024 doi arXiv

  5. [13]

    Kruk, J., Lubin, J., Sikka, K., Lin, X., Jurafsky, D., Divakaran, A.: Integrating text and image: Determining multimodal document intent in instagram posts. In: ConferenceonEmpiricalMethodsinNaturalLanguageProcessingandtheInterna- tional Joint Conference on Natural Language Pr...

    2019 doi

  6. [14]

    arXiv preprintabs/2307.16125 (2023)

    Li, B., Wang, R., Wang, G., Ge, Y., Ge, Y., Shan, Y.: Seed-bench: Benchmarking multimodal llms with generative comprehension. arXiv preprintabs/2307.16125 (2023). https://doi.org/10.48550/ARXIV.2307.16125

    2023 doi arXiv

  7. [15]

    In: International Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA

    Li, J., Li, D., Savarese, S., Hoi, S.C.H.: BLIP-2: bootstrapping language-image pre- training with frozen image encoders and large language models. In: International Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA. pp. 19730–19742. PMLR (2023)...

    2023

  8. [16]

    arXiv preprintabs/1908.03557 (2019)

    Li, L.H., Yatskar, M., Yin, D., Hsieh, C., Chang, K.: Visualbert: A simple and per- formant baseline for vision and language. arXiv preprintabs/1908.03557 (2019)

    2019 arXiv

  9. [17]

    In: Annual Conference on Neural Information Processing Systems, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023 (2023),https://openreview.net/forum?id=w0H2xGHlkw

    Liu, H., Li, C., Wu, Q., Lee, Y.J.: Visual instruction tuning. In: Annual Conference on Neural Information Processing Systems, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023 (2023),https://openreview.net/forum?id=w0H2xGHlkw

    2023

  10. [18]

    https://doi.org/10.48550/ARXIV.2403.05525

    Lu, H., Liu, W., Zhang, B., Wang, B., Dong, K., Liu, B., Sun, J., Ren, T., Li, Z., Yang, H., Sun, Y., Deng, C., Xu, H., Xie, Z., Ruan, C.: Deepseek-vl: Towards real-worldvision-languageunderstanding.arXivpreprint abs/2403.05525(2024). https://doi.org/10.48550/ARXIV.2403.05525

    2024 doi arXiv

  11. [19]

    In: Conference on Empirical Methods in Natural Language Processing, EMNLP 2021, Virtual Event / Punta Cana, Dominican Re- public, 7-11 November, 2021

    Luo, G., Darrell, T., Rohrbach, A.: Newsclippings: Automatic generation of out- of-context multimodal media. In: Conference on Empirical Methods in Natural Language Processing, EMNLP 2021, Virtual Event / Punta Cana, Dominican Re- public, 7-11 November, 2021. pp. 6801–6817. As...

    2021 doi

  12. [20]

    In: European Conference on Title Suppressed Due to Excessive Length 15 Computer Vision, ECCV 2018, Munich, Germany, September 8-14, 2018

    Müller-Budack, E., Pustu-Iren, K., Ewerth, R.: Geolocation estimation of photos using a hierarchical model and scene classification. In: European Conference on Title Suppressed Due to Excessive Length 15 Computer Vision, ECCV 2018, Munich, Germany, September 8-14, 2018. pp. 575–

    2018

  13. [21]

    Müller-Budack, E., Springstein, M., Hakimov, S., Mrutzek, K., Ewerth, R.: Ontology-driveneventtypeclassificationinimages.In:IEEEWinterConferenceon Applications of Computer Vision, WACV 2021, Virtual Event, January 3-8, 2021. pp. 2927–2937. IEEE (2021).https://doi.org/10.1109/W...

    2021

  14. [22]

    In: International Conference on Multimedia Retrieval, ICMR 2020, Dublin, Ireland, June 8-11, 2020

    Müller-Budack, E., Theiner, J., Diering, S., Idahl, M., Ewerth, R.: Multimodal analytics for real-world news using measures of cross-modal entity consistency. In: International Conference on Multimedia Retrieval, ICMR 2020, Dublin, Ireland, June 8-11, 2020. pp. 16–25. ACM (202...

    2020 doi

  15. [23]

    OpenAI: Gpt-4v(ision) system card (2023), https://cdn.openai.com/papers/ GPTV_System_Card.pdf

    2023

  16. [24]

    In: International Conference on Multime- dia Retrieval, ICMR 2019, Ottawa, ON, Canada, June 10-13, 2019

    Otto, C., Springstein, M., Anand, A., Ewerth, R.: Understanding, categorizing and predicting semantic image-text relations. In: International Conference on Multime- dia Retrieval, ICMR 2019, Ottawa, ON, Canada, June 10-13, 2019. pp. 168–176. ACM (2019). https://doi.org/10.1145...

    2019

  17. [25]

    Papadopoulos, S., Koutlis, C., Papadopoulos, S., Petrantonakis, P.: Synthetic mis- informers:Generatingandcombatingmultimodalmisinformation.In:International Workshop on Multimedia AI against Disinformation, MAD@ICMR 2023, Thessa- loniki, Greece, June 12-15, 2023. pp. 36–44. AC...

    2023

  18. [26]

    In: IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024, Seattle, WA, USA, June 16-22, 2024

    Qi, P., Yan, Z., Hsu, W., Lee, M.: Sniffer: Multimodal large language model for explainable out-of-context misinformation detection. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2024, Seattle, WA, USA, June 16-22, 2024. pp. 13052–13062. IEEE (2024)....

    2024

  19. [27]

    In: International Conference on Machine Learning, ICML 2021, Virtual Event, 18-24 July, 2021

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., Sutskever, I.: Learning transferable visual models from natural language supervision. In: International Conference on Machine Learning, ICML 2021...

    2021

  20. [28]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 40(5), 1072–1085 (2018)

    Ramisa, A., Yan, F., Moreno-Noguer, F., Mikolajczyk, K.: Breakingnews: Article annotation by image and text processing. IEEE Transactions on Pattern Analysis and Machine Intelligence 40(5), 1072–1085 (2018). https://doi.org/10.1109/ TPAMI.2017.2721945

    2018

  21. [29]

    In: ACM Multimedia Conference on Multimedia Confer- ence, MM 2018, Seoul, Republic of Korea, October 22-26, 2018

    Sabir, E., AbdAlmageed, W., Wu, Y., Natarajan, P.: Deep multimodal image- repurposing detection. In: ACM Multimedia Conference on Multimedia Confer- ence, MM 2018, Seoul, Republic of Korea, October 22-26, 2018. pp. 1337–1345. ACM (2018). https://doi.org/10.1145/3240508.3240707

    2018

  22. [30]

    In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2015, Boston, MA, USA, June 7-12, 2015

    Schroff, F., Kalenichenko, D., Philbin, J.: Facenet: A unified embedding for face recognition and clustering. In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2015, Boston, MA, USA, June 7-12, 2015. pp. 815–823. IEEE Computer Society (2015).https://doi.org/...

    2015

  23. [31]

    In: International Conference on Informa- tion and Knowledge Management, CIKM 2024, Boise, ID, USA, October 21 – 25

    Tahmasebi, S., Müller-Budack, E., Ewerth, R.: Multimodal misinformation detec- tion using large vision-language models. In: International Conference on Informa- tion and Knowledge Management, CIKM 2024, Boise, ID, USA, October 21 – 25

    2024

  24. [32]

    In: European Conference on Information Re- trieval, ECIR 2023, Dublin, Ireland, April 2-6, 2023, Proceedings, Part II

    Tahmasebzadeh, G., Hakimov, S., Ewerth, R., Müller-Budack, E.: Multimodal ge- olocation estimation of news photos. In: European Conference on Information Re- trieval, ECIR 2023, Dublin, Ireland, April 2-6, 2023, Proceedings, Part II. pp. 204–220. Springer (2023).https://doi.or...

    2023 doi

  25. [33]

    Communications of the ACM57(10), 78–85 (2014)

    Vrandečić, D., Krötzsch, M.: Wikidata: a free collaborative knowledgebase. Communications of the ACM57(10), 78–85 (2014). https://doi.org/10.1145/ 2629489

    2014

  26. [34]

    In: Conference on Empirical Methods in Natural Language Processing, EMNLP 2020, Online, November 16-20, 2020

    Wu, L., Petroni, F., Josifoski, M., Riedel, S., Zettlemoyer, L.: Scalable zero-shot entity linking with dense entity retrieval. In: Conference on Empirical Methods in Natural Language Processing, EMNLP 2020, Online, November 16-20, 2020. pp. 6397–6407. Association for Computat...

    2020 doi

  27. [35]

    arXiv preprintabs/2306.09265 (2023)

    Xu, P., Shao, W., Zhang, K., Gao, P., Liu, S., Lei, M., Meng, F., Huang, S., Qiao, Y., Luo, P.: Lvlm-ehub: A comprehensive evaluation benchmark for large vision- language models. arXiv preprintabs/2306.09265 (2023). https://doi.org/10. 48550/ARXIV.2306.09265

    2023 doi arXiv

  28. [36]

    arXiv preprintabs/2306.13549 (2023)

    Yin, S., Fu, C., Zhao, S., Li, K., Sun, X., Xu, T., Chen, E.: A survey on mul- timodal large language models. arXiv preprintabs/2306.13549 (2023). https: //doi.org/10.48550/ARXIV.2306.13549

    2023 doi arXiv

  29. [37]

    In: British Machine Vision Conference 2018, BMVC 2018, Newcastle, UK, September 3-6, 2018

    Zhang, M., Hwa, R., Kovashka, A.: Equal but not the same: Understanding the im- plicit relationship between persuasive images and text. In: British Machine Vision Conference 2018, BMVC 2018, Newcastle, UK, September 3-6, 2018. p. 8. BMVA Press (2018), http://bmvc2018.org/conte...

    2018

  30. [38]

    ACM Computing Surveys51(2), 32:1–32:36 (2018)

    Zubiaga, A., Aker, A., Bontcheva, K., Liakata, M., Procter, R.: Detection and resolution of rumours in social media: A survey. ACM Computing Surveys51(2), 32:1–32:36 (2018). https://doi.org/10.1145/3161603

    2018 doi

  31. [592]

    Springer (2018).https://doi.org/10.1007/978-3-030-01258-8_35

    2018 doi

  32. [2024]

    2189–2199

    p. 2189–2199. Association for Computing Machinery, New York, NY, USA (2024). https://doi.org/10.1145/3627673.3679826 16 S. Tahmasebi et al

    2024

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.