Pith. sign in

REVIEW 4 major objections 4 minor 52 references

Four Shades of Life Sciences: A Dataset for Disinformation Detection in the Life Sciences

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that purely linguistic style, without fact-checking, can flag disinformative life-science full texts; on a new 2,603-document corpus, a fine-tuned biomedical language model reaches 0.9861 weighted F1.

desk verdict FSoLS is a genuinely new four-class full-text corpus for life-science disinformation, but the high F1 scores likely reflect source identity more than intrinsic disinformation style. read the letter →

arxiv 2507.03488 v1 pith:TWFUVNUB submitted 2025-07-04 cs.CL

Eva Seidlmayer , Lukas Galke , Konrad U. Förstner This is my paper
classification cs.CL
keywords disinformationdetectionlifesciencestextclassificationfull-textcorpuslinguisticstylegenrehealthmisinformationsliding-window
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 introduces Four Shades of Life Sciences (FSoLS), a corpus of 2,603 full-length health and life-science articles drawn from 17 sources and labeled into four genres: scientific texts, vernacular science writing, alternative-science texts, and disinformative texts. The central claim is that disinformation has a distinctive rhetorical fingerprint, so purely linguistic features—without checking facts against a reference corpus—can separate it from the other genres. The authors show that fine-tuning a biomedical language model on this corpus and letting it read 2,500 tokens per document through a sliding window reaches a weighted F1-score of 0.9861, a standard combined measure of precision and recall, and that a transparent linear classifier reaches about 0.97. If the claim holds, automated style-based pre-screening of full-text articles could support human evaluators in flagging suspicious health and life-science content.

What carries the argument

The load-bearing object is the FSoLS corpus: 2,603 full-text documents, balanced across four genres and 14 topics, so that genre and style—not subject matter—are the only consistent discriminative signals. The method combines this corpus with a fine-tuned biomedical transformer whose token limit is extended by a sliding window of 512 tokens with stride 256, averaging window predictions over up to 2,500 tokens. It also uses classical bag-of-words classifiers, especially a linear support-vector classifier on TF-IDF features, whose feature weights expose which words drive each genre. The same corpus supports the linguistic analysis, via TF-IDF and random-forest decision rules, that identifies the rhetorical markers typical of disinformative writing.

What would settle it

Build a test set from verified instances of disinformation that appear in neutral, mainstream-looking outlets, or from deliberately misleading articles on websites not used to build FSoLS; if a model trained on the corpus cannot separate these from ordinary scientific or vernacular writing, the claim that linguistic style alone carries the signal would be undercut.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that syntactic and semantic style is a workable proxy for intent: texts that are deliberately misleading use colloquial contractions, emotional appeals, authority claims, and other rhetorical markers that distinguish them from evidence-based science, patient-oriented summaries, and alternative-medicine writing. The authors construct a topic-balanced corpus precisely so that classifiers cannot succeed merely by recognizing topics, and they show that the four-way genre distinction is learnable from full-text documents. A fine-tuned biomedical transformer with a sliding window achieves a weighted F1-score of 0.9861, while a linear support-vector classifier with TF-IDF features reaches 0.9701. Performance on a test set built from three previously unseen topics remains close to the standard test-set performance, which the authors take as evidence that the model learns text style rather than topic content.

Load-bearing premise

The weakest premise is that labeling a text as disinformative because its website has a bad reputation, without verifying that the content is actually false and deliberately so, yields a valid ground truth; if source reputation is not a reliable proxy for disinformation, the classifier may be recognizing the source rather than the style.

Editorial extensions

If this is right

  • If the style-based claim holds, a full-text life-science article could be pre-screened automatically and ranked by resemblance to known disinformation before any fact-checking begins.
  • The four-way genre split gives downstream tools a middle ground—vernacular and alternative-science writing—between verified science and disinformation, avoiding forced binary labels.
  • The near-parity of the lightweight linear classifier, at about 0.97 F1, means screening could run on modest hardware and offer human reviewers inspectable feature weights.
  • The model's robustness to previously unseen topics suggests the corpus can grow to new subjects without a drop in accuracy, as happened when climate change, pandemics, and urine were added.

Reading between the lines

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

  • An implication the authors do not press is that the source-reputation labeling makes the reported accuracies an upper bound on pure style detection; some of the signal is likely source identity.
  • A direct test of the style thesis would be to rewrite verified disinformative claims in a neutral register and see whether the classifier still marks them as disinformative.
  • The inclusion of alternative-science texts as a separate class invites a further question: whether borderline genres can be ordered along a continuum from evidence-based science to disinformation.
  • Because the corpus construction is reproducible, the same pipeline could track how disinformation rhetoric drifts as new products, topics, and platform norms emerge.
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 / 4 minor

Summary. The paper introduces FSoLS, a corpus of 2,603 full-text life-science documents in four classes (scientific, vernacular, alternative scientific, disinformative) drawn from 17 sources on 14 topics. It reports that fine-tuned BioBERT with a 2,500-token sliding window reaches a weighted F1 of 0.9861, that a linear SVC with TF-IDF reaches 0.97, and that linguistic features such as colloquial contractions and structural markers distinguish the classes. The authors argue that purely linguistic information can serve as a proxy for disinformation detection, and they provide code for dataset reconstruction and model benchmarking.

Significance. If the central claim were established, the paper would be a useful contribution: FSoLS is one of few full-text, multi-genre life-science disinformation datasets, and the comparison of transformer-based and classical classifiers with an openly available reconstruction pipeline is valuable. The dataset itself addresses a real gap, since most prior resources are short social-media posts or binary fact-checking collections. However, the current evaluation does not support the generalized claim that the models detect intrinsic disinformation style rather than source identity. The paper's own Discussion acknowledges that evaluation on entirely different sources is still needed, and the experiments as presented leave this gap unresolved. The dataset and benchmark are worth publishing after a major revision that either adds unseen-source evaluation, demonstrates within-source discriminability, or appropriately limits the scope of the claims.

major comments (4)
  1. [§5.1.4, §5.2, Fig. 1]
  2. [§2.0.1, Table 8, §5.2]
  3. [§3.0.3, Table 4]
  4. [§3.0.2, Table 3]
minor comments (4)
  1. [Table 1, §2]
  2. [§3.0.6, Table 7]
  3. [§2.0.1]
  4. [References]

Circularity Check

1 steps flagged · score 6.0 of 10

The headline F1 is partly a source-identity score: disinformative labels are assigned by website reputation and all test splits come from the same 17 source families, so the 'style proxy' conclusion reduces, within this dataset, to separating predefined source lists.

  1. self definitional [Section 5.1.4 (Disinformative texts data source), with Table 9; evaluated in Sections 3.0.3-3.0.4 (Tables 4-5)]
    "For our disinformation category, we selected websites that have been widely recognized by subject-matter experts as sources of false or misleading information."

    The disinformative class is constructed by source reputation: every item from Mercola, NaturalNews, Health Impact News, InfoWars, or health.news is labeled 'disinformative' (Section 5.1.4, Table 9), with no content-level verification. The headline F1=0.9861 (Table 5) is measured on held-out items from these same 17 source families, and the cross-topic test deliberately uses 'familiar sources' (Section 3.0.3). The target variable is therefore, by construction, a function of the source list; a model can reach the reported score by learning source-specific lexical, structural, or formatting cues rather than intrinsic falsehood or misleading intent.

full rationale

FSoLS is a genuinely new and useful resource, and most of the paper is not circular: model selection is standard, no external 'uniqueness' theorem is imported, and the only author self-citation ([10], Galke et al.) is a model-selection review, not load-bearing for the central result. The circularity concern is narrower and located in the inference from the benchmark to the proxy claim. Because the disinformative class is populated entirely from a fixed list of websites (Section 5.1.4; Table 9) and all test splits, both topic-known and topic-unknown, are drawn from the same 17 source families, the reported 0.9861 F1 can be achieved by learning source-specific style or formatting features rather than by detecting intrinsic disinformation. The paper even states that the t-SNE shows clustering by 'designated classes and data sources' and that the unknown-topic test demonstrates the model's 'ability to handle new content from familiar sources.' The Discussion explicitly recommends 'further evaluation of texts from entirely different sources,' which acknowledges that cross-source generalization is untested. Thus the central claim that style is a reasonable proxy for disinformation is only partially supported; within FSoLS it is in part a source-recognition result. This is an evaluation-design issue, not a fabrication, and the authors' transparency about the limitation prevents a higher score.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim relies on the source-reputation labeling hypothesis and the genre taxonomy. No new theoretical entities are introduced. The free parameters listed are hyperparameters selected using test-set performance, which affect the reported accuracy and the strength of the central claim.

free parameters (2)
  • Sliding-window token count (2,500) = 2,500 tokens
    Chosen as the best token budget among 512, 2.5k, 5k, 7.5k, 10k based on test-set F1 (Table 5), which introduces test-set-based model selection.
  • Fine-tuning epoch count (3) = 3 epochs
    Best epoch selected from per-epoch test-set F1 (Table 3), a form of test-set peeking that can inflate reported performance.
assumptions (3)
  • domain assumption Texts published by sources listed as disinformative are all disinformative, and texts from the chosen scientific, vernacular, and alternative sources are not.
    Ground truth is assigned by source reputation (Section 5.1.4), not by content verification; classification performance therefore partly measures source recognition.
  • domain assumption The four classes are mutually exclusive and sufficiently exhaustive for the life-science text genres studied.
    Section 2 defines four categories but does not demonstrate that real-world texts cannot straddle categories or fall outside.
  • domain assumption Linguistic style is a reliable proxy for the intent behind disinformation.
    Sections 1 and 4 assert stylistic differences driven by intent; this is the theoretical premise of the study and is not independently validated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Four Shades of Life Sciences: A Dataset for Disinformation Detection in the Life Sciences." pith.science (2026). https://pith.science/paper/TWFUVNUB

@misc{pith2026250703488,
  author       = {Pith},
  title        = {Pith review of: Four Shades of Life Sciences: A Dataset for Disinformation Detection in the Life Sciences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TWFUVNUB}},
  note         = {Machine review of arXiv:2507.03488}
}
read the original abstract

Disseminators of disinformation often seek to attract attention or evoke emotions - typically to gain influence or generate revenue - resulting in distinctive rhetorical patterns that can be exploited by machine learning models. In this study, we explore linguistic and rhetorical features as proxies for distinguishing disinformative texts from other health and life-science text genres, applying both large language models and classical machine learning classifiers. Given the limitations of existing datasets, which mainly focus on fact checking misinformation, we introduce Four Shades of Life Sciences (FSoLS): a novel, labeled corpus of 2,603 texts on 14 life-science topics, retrieved from 17 diverse sources and classified into four categories of life science publications. The source code for replicating, and updating the dataset is available on GitHub: https://github.com/EvaSeidlmayer/FourShadesofLifeSciences

Figures

Figures reproduced from arXiv: 2507.03488 by the authors.

Figure 1
Figure 1. Vectorized t-SNE representation of FSoLS applying doc2vec with [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Sliding-window concept implemented in transformer’s forward func [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. F1-score with different token counts for fine-tuning [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Balanced compilation of dataset with regard to (a) data sources (a) [PITH_FULL_IMAGE:figures/full_fig_p020_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 39 canonical work pages

  1. [1]

    Telling a Lie: Analyzing the Language of In- formation and Misinformation during Global Health Events

    Ankit Aich and Natalie Parde. Telling a Lie: Analyzing the Language of In- formation and Misinformation during Global Health Events. In Proceedings of the Thirteenth Language Resources and Evaluation Conference , pages 4135–4141, Marseille, France, June 2022. European Language Resources Association. URL https://aclanthology.org/2022.lrec-1.439

    work page 2022

  2. [2]

    A comprehensive sur- vey on machine learning approaches for fake news detection

    Jawaher Alghamdi, Suhuai Luo, and Yuqing Lin. A comprehensive sur- vey on machine learning approaches for fake news detection. Multi- media Tools and Applications , 83(17):51009–51067, 2024. ISSN 1573-

    work page 2024

  3. [3]

    Kadan Anoop, Padmanabhan Deepak, and V. L. Lajish. Emotion cog- nizance improves health fake news identification. In Proceedings of the 24th Symposium on International Database Engineering & Applications , pages 1–10. ACM, 2020. ISBN 978-1-4503-7503-0. URL https://dl.acm.org/ doi/10.1145/3410566.3410595. https://doi.org/10.1145/3410566. 3410595

    work page arXiv

  4. [4]

    Bender, Timnit Gebru, Angelina McMillan-Major, and Shmar- garet Shmitchell

    Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmar- garet Shmitchell. On the dangers of stochastic parrots. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency ,

    work page 2021

  5. [5]

    Infographic: The best and worst sci- ence news sites, 2017

    Alex Berezow. Infographic: The best and worst sci- ence news sites, 2017. Amercian Council on Scicence and Health website https://www.acsh.org/news/2017/03/05/ infographic-best-and-worst-science-news-sites-10948

    work page 2017

  6. [6]

    Rothschild, Emily Thorson, and Duncan J

    Ceren Budak, Brendan Nyhan, David M. Rothschild, Emily Thorson, and Duncan J. Watts. Misunderstanding the harms of online misinfor- mation. Nature, 630(8015):45–53, June 2024. ISSN 0028-0836, 1476-

    work page 2024

  7. [7]

    Arman Cohan, Sergey Feldman, Iz Beltagy, Doug Downey, and Daniel S. Weld. SPECTER: Document-level Representation Learning using Citation- informed Transformers, May 2020. arXiv preprint arXiv:2004.07180 [cs]

    arXiv 2020

  8. [8]

    BERT: Pre-training of deep bidirectional transformers for language un- derstanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language un- derstanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Lan- guage Technologies, Volume 1 (Long and Short Papers) , pages 4171–4186, Minnea...

    doi:10.18653/v1/n19-1423 2019

Show all 52 references
  1. [9]

    Guidelines for safeguarding good re- search practice

    Deutsche Forschungsgemeinschaft. Guidelines for safeguarding good re- search practice. code of conduct, 2022. https://doi.org/10.5281/ zenodo.6472827

    2022

  2. [10]

    Are we really making much progress in text classification? a comparative review

    Lukas Galke, Ansgar Scherp, Andor Diera, Fabian Karl, Bao Xin Lin, Bhakti Khera, Tim Meuser, and Tushar Singhal. Are we really making much progress in text classification? a comparative review. https:// arxiv.org/abs/2204.03954, 2025. URL https://arxiv.org/abs/2204. 03954. arX...

    2025 arXiv

  3. [11]

    Fakten gegen Fake News oder Der Kampf um die Demokratie

    Patrick Gensing. Fakten gegen Fake News oder Der Kampf um die Demokratie. Bundeszentrale f¨ ur politische Bildung, Bonn, 2020. ISBN 978-3-7425-0500-2. Schriftenreihe Band 10500

    2020

  4. [12]

    How disinformation is undermining our human rights

    Global Disinformation Index. How disinformation is undermining our human rights. https://www.disinformationindex.org/blog/ 2023-07-13-how-disinformation-is-undermining-our-human-rights/ ,

    2023

  5. [13]

    Defunding Disinformation

    Global Disinformation Index. Defunding Disinformation. How we define Disinformation. https://www.disinformationindex.org/mission, 2023. Blogpost on Global Disinformation Index website

    2023

  6. [14]

    A survey on automated fact-checking

    Zhijiang Guo, Michael Schlichtkrull, and Andreas Vlachos. A survey on automated fact-checking. Transactions of the Association for Compu- tational Linguistics , 10:178–206, February 2022. ISSN 2307-387X. doi: 10.1162/tacl a 00454. URL https://doi.org/10.1162/tacl_a_00454. 23

    2022 doi

  7. [15]

    Das vermeintliche Mistel-Wunder

    Christian Honey. Das vermeintliche Mistel-Wunder. der Masterplan der Anthroposophie, 2023. https://medwatch.de/alternativmedizin/ das-vermeintliche-mistel-wunder-der-masterplan-der-anthroposophie/

    2023

  8. [16]

    Anatomy of a disinformation empire: Investigating NaturalNews

    Institute for Strategic Dialogue. Anatomy of a disinformation empire: Investigating NaturalNews. https://www.isdglobal.org/wp-content/ uploads/2020/06/20200620-ISDG-NaturalNews-Briefing-V4.pdf , 2020

    2020

  9. [17]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic op- timization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Pro- ceedings, 2015

    2015

  10. [18]

    Vinod Vydis- waran

    Alexander Kinsora, Kate Barron, Qiaozhu Mei, and V.G. Vinod Vydis- waran. Creating a labeled dataset for medical misinformation in health forums. In 2017 IEEE International Conference on Healthcare Informat- ics (ICHI) , pages 456–461. IEEE, 2017. ISBN 978-1-5090-4881-6. URL h...

    2017

  11. [19]

    Explainable Automated Fact- Checking for Public Health Claims

    Neema Kotonya and Francesca Toni. Explainable Automated Fact- Checking for Public Health Claims. http://arxiv.org/abs/2010.09926, October 2020. arXiv preprint 2010.09926

    2010 arXiv

  12. [20]

    BioBERT: a pre-trained biomedical lan- guage representation model for biomedical text mining

    Jinhyuk Lee, Wonjin Yoon, Sungdong Kim, Donghyeon Kim, Sunkyu Kim, Chan Ho So, and Jaewoo Kang. BioBERT: a pre-trained biomedical lan- guage representation model for biomedical text mining. Bioinformatics, 36 (4):1234–1240, 2020. ISSN 1367-4803. URL https://doi.org/10.1093/ bi...

    2020 doi

  13. [21]

    McGarity and Wendy Wagner

    Thomas O. McGarity and Wendy Wagner. Bending science: how special interests corrupt public health research . Harvard University Press, 2008. ISBN 978-0-674-02815-9. OCLC: ocn175174375

    2008

  14. [22]

    Naomi Oreskes and Erik M. Conway. Merchants of doubt: how a handful of scientists obscured the truth on issues from tobacco smoke to global warm- ing. Bloomsbury, paperback. ed edition, 2012. ISBN 978-1-4088-2483-2

    2012

  15. [23]

    Markatos, and Nicolas Kourtellis

    Emmanouil Papadogiannakis, Panagiotis Papadopoulos, Evangelos P. Markatos, and Nicolas Kourtellis. Who funds misinformation? a sys- tematic analysis of the ad-related profit routines of fake news sites. In Proceedings of the ACM Web Conference 2023 , WWW ’23, pages 2765– 2776,...

    2023

  16. [24]

    Hierarchical Transformers for Long Document Classi- fication

    Raghavendra Pappagari, Piotr ˙Zelasko, Jes´ us Villalba, Yishay Carmiel, and Najim Dehak. Hierarchical Transformers for Long Document Classi- fication. https://arxiv.org/abs/1910.10781, 2019. https://doi.org/ 10.48550/ARXIV.1910.10781

    1910 doi arXiv

  17. [25]

    The filter bubble: what the Internet is hiding from you

    Eli Pariser. The filter bubble: what the Internet is hiding from you . Penguin Press, New York, 2011. ISBN 978-1-59420-300-8

    2011

  18. [26]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Pas- sos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. Scikit-learn: Machine learning in Python. Journal of Machine Learni...

    2011

  19. [27]

    Information Literacy in the Digital Age: Why Critical Digital Literacy Matters for Democracy

    Gianfranco Polizzi. Information Literacy in the Digital Age: Why Critical Digital Literacy Matters for Democracy. In St´ ephane Goldstein, editor, Informed Societies, pages 1–24. St´ ephane Goldstein, 1 edition, 2019. ISBN 978-1-78330-392-2. URL https://www.cambridge.org/core/...

    2019

  20. [28]

    Automatic detection of fake news

    Ver´ onica P´ erez-Rosas, Bennett Kleinberg, Alexandra Lefevre, and Rada Mihalcea. Automatic detection of fake news. http://arxiv.org/10. 48550/ARXIV.1708.07104, 2017. URL https://arxiv.org/abs/1708. 07104. arXiv preprint arXiv:1708.07104

    2017 doi arXiv

  21. [29]

    Genna Reed, Yogi Hendlin, Anita Desikan, Taryn MacKinney, Emily Berman, and Gretchen T. Goldman. The disinformation playbook: how industry manipulates the science-policy process—and how to restore scien- tific integrity. Journal of Public Health Policy , 42(4):622–634, 2021. I...

    2021 doi

  22. [30]

    Performance, Energy Consumption and Costs: A Comparative Analysis of Automatic Text Classification Ap- proaches in the Legal Domain

    Leonardo Rigutini, Achille Globo, Marco Stefanelli, Andrea Zugarini, Sinan Gultekin, and Marco Ernandes. Performance, Energy Consumption and Costs: A Comparative Analysis of Automatic Text Classification Ap- proaches in the Legal Domain. International Journal on Natural Lan- g...

    2024

  23. [31]

    Politifact Fact Check Dataset, 2022

    Rishabh Misra. Politifact Fact Check Dataset, 2022. https://rgdoi.net/ 10.13140/RG.2.2.29923.22566

    2022

  24. [32]

    Ryan, Andrew J

    Camille D. Ryan, Andrew J. Schaul, Ryan Butner, and John T. Swarthout. Monetizing disinformation in the attention economy: The case of genet- ically modified organisms (gmos). European Management Journal , 38 (1):7–18, 2020. ISSN 0263-2373. URL https://www.sciencedirect. 25 co...

    2020

  25. [33]

    Internistenkongress: Wie fake news in der medi- zin gross werden

    Maren Schenk. Internistenkongress: Wie fake news in der medi- zin gross werden. Deutsches ¨Arzteblatt, 118(25):A–1258 / B– 1040, 2021. https://www.aerzteblatt.de/archiv/219830/ Internistenkongress-Wie-Fake-News-in-der-Medizin-gross-werden

    2021

  26. [34]

    GerDISDETECT: A German multilabel dataset for dis- information detection

    Mina Sch¨ utz, Daniela Pisoiu, Daria Liakhovets, Alexander Schindler, and Melanie Siegel. GerDISDETECT: A German multilabel dataset for dis- information detection. In Nicoletta Calzolari, Min-Yen Kan, Veronique Hoste, Alessandro Lenci, Sakriani Sakti, and Nianwen Xue, editors,...

    2024

  27. [35]

    False news of a cannabis cancer cure

    Siyu Shi, Arthur R Brant, Aaron Sabolch, and Erqi Pol- lom. False news of a cannabis cancer cure. Cureus,

  28. [36]

    Frequently asked questions about hate and antigovernment groups, 2022

    Southern Poverty Law Center. Frequently asked questions about hate and antigovernment groups, 2022. https://www.splcenter.org/20220216/ frequently-asked-questions-about-hate-and-antigovernment-groups# hate%20group

    2022

  29. [37]

    Andy S. L. Tan and Cabral A. Bigman. Misinformation about commer- cial tobacco products on social media – implications and research oppor- tunities for reducing tobacco-related health disparities. American Jour- nal of Public Health , 110:S281–S283, 2020. ISSN 1541-0048. https...

    2020

  30. [38]

    Antisocial media: how Facebook disconnects US and undermines democracy

    Siva Vaidhyanathan. Antisocial media: how Facebook disconnects US and undermines democracy. Oxford University Press, New York, 2018. ISBN 978-0-19-084117-1 978-0-19-084118-8

    2018

  31. [39]

    Of lizards and ideological entrepreneurs: Alex Jones and Infowars in the relationship between populist national- ism and the post-global media ecology

    H Van Den Bulck and A Hyzen. Of lizards and ideological entrepreneurs: Alex Jones and Infowars in the relationship between populist national- ism and the post-global media ecology. International Communication Gazette, 82(1):42–59, February 2020. ISSN 1748-0485. URL https: //jo...

    2020 doi

  32. [40]

    Visualizing data using t- sne

    Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t- sne. Journal of Machine Learning Research, 9(86):2579–2605, 2008. http: //jmlr.org/papers/v9/vandermaaten08a.html. 26

    2008

  33. [41]

    A free and pluralistic media to sustain Eu- ropean democracy

    Vaira V ¯ ıke-Freiberga, Herta D¨ aubler-Gmelin, Ben Hammersley, and Lu ´ ıs Miguel Poiares Pessoa Maduro. A free and pluralistic media to sustain Eu- ropean democracy. The Report of the High Level Group on Media Freedom and Pluralism, 2013. https://ec.europa.eu/information_so...

    2013

  34. [42]

    Fact or fiction: Veri- fying scientific claims

    David Wadden, Shanchuan Lin, Kyle Lo, Lucy Lu Wang, Madeleine van Zuylen, Arman Cohan, and Hannaneh Hajishirzi. Fact or fiction: Veri- fying scientific claims. In Bonnie Webber, Trevor Cohn, Yulan He, and Yang Liu, editors, Proceedings of the 2020 Conference on Empirical Meth-...

    2020 doi

  35. [43]

    Mitchell Waldrop

    M. Mitchell Waldrop. The genuine problem of fake news. Proceedings of the National Academy of Sciences , 114(48):12631–12634, 2017. URL https://www.pnas.org/doi/10.1073/pnas.1719005114. https://doi. org/10.1073/pnas.1719005114

    2017 doi

  36. [44]

    Using machine learning to verify scientific claims

    Lucy Lu Wang. Using machine learning to verify scientific claims. In Artificial Intelligence in Science: Challenges, Opportunities and the Future of Research. OECD Publishing, Paris, 2023. https://doi.org/10.1787/ ab9d235c-en

    2023

  37. [45]

    Multi-Granularity Hierarchical At- tention Fusion Networks for Reading Comprehension and Question An- swering

    Wei Wang, Ming Yan, and Chen Wu. Multi-Granularity Hierarchical At- tention Fusion Networks for Reading Comprehension and Question An- swering. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages 1705–1714,...

    2018

  38. [46]

    ”Liar, Liar Pants on Fire”: A New Benchmark Dataset for Fake News Detection

    William Yang Wang. ”Liar, Liar Pants on Fire”: A New Benchmark Dataset for Fake News Detection. In Proceedings of the 55th Annual Meet- ing of the Association for Computational Linguistics (Volume 2: Short Papers), pages 422–426, Vancouver, Canada, 2017. Association for Com- p...

    2017 doi

  39. [47]

    Let’s flatten the infodemic curve, 2020

    World Health Organization. Let’s flatten the infodemic curve, 2020. https://www.who.int/news-room/spotlight/ let-s-flatten-the-infodemic-curve . 27 Table 6: BioBERT model results for scientific text style class with and without consideration of citation frequency Epoch Precisi...

    2020

  40. [2019]

    https://www.cureus.com/articles/ 16818-false-news-of-a-cannabis-cancer-cure

    ISSN 2168-8184. https://www.cureus.com/articles/ 16818-false-news-of-a-cannabis-cancer-cure

  41. [2021]

    URL https://dl.acm.org/doi/10.1145/ 3442188.3445922

    ISBN 978-1-4503-8309-7. URL https://dl.acm.org/doi/10.1145/ 3442188.3445922. https://doi.org/10.1145/3442188.3445922. 22

  42. [2023]

    Blogpost on Global Disinformation Index website

  43. [4687]

    https://doi.org/10.1038/s41586-024-07417-w

    URL https://www.nature.com/articles/s41586-024-07417-w . https://doi.org/10.1038/s41586-024-07417-w

    doi

  44. [7721]

    URL https://doi.org/10.1007/ s11042-023-17470-8

    doi: 10.1007/s11042-023-17470-8. URL https://doi.org/10.1007/ s11042-023-17470-8

    doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.