Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

TikTok Rewards Divisive Political Messaging During the 2025 German Federal Election

T0 review · 3 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read During the 2025 German federal election, TikTok rewarded divisive political content: anger, disgust, and outgroup animosity drew significantly more engagement, while positive and relatable messages drew less.

desk verdict Solid observational study; the divisive-content-engagement finding is plausible and likely robust, but the title outruns the design and the LLM validation is thinner than it looks. read the letter →

arxiv 2509.10336 v1 pith:7ZVPMM6R submitted 2025-09-12 cs.SI

Kirill Solovev , Chiara Drolsbach , Emma Demirel , Nicolas Pröllochs This is my paper
classification cs.SI
keywords TikTokpoliticalcommunicationengagementnegativesentimentoutgroupanimosityGermanfederalelectioncomputationalcontentanalysisLLMannotation
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 sets out to show that TikTok's engagement dynamics during the 2025 German federal election systematically rewarded divisive political communication over unifying communication. Analyzing 25,292 videos posted by German politicians, parties, and parliamentary groups, the authors find that negative sentiment, anger, disgust, and outgroup animosity were associated with significantly more likes, comments, shares, and views, whereas positive sentiment, relatability, and identity language were associated with less. They also report that far-left and far-right parties were both more likely to post this divisive content and more successful at generating engagement than centrist parties. If correct, this implies that the platform's incentive structure favors extreme actors and disincentivizes moderate, unifying messaging.

What carries the argument

The argument rests on a computational content-analysis pipeline joined with multilevel negative binomial regression. Transcripts of all 25,292 videos are labeled for sentiment, eight discrete emotions, identity language, outgroup animosity, and relatability; those labels become fixed effects in separate regression models of likes, comments, shares, and views, with random intercepts per account and monthly fixed effects. The regression is the central mechanism that converts content features into percentage differences in engagement, and the sign pattern of its coefficients carries the paper's conclusion.

What would settle it

Have native German raters manually label a random sample of about 1,000 videos from the same period for sentiment, anger, disgust, outgroup animosity, relatability, and identity language, then re-estimate the paper's multilevel negative binomial models using only the human-coded features. If the engagement advantages for negativity and outgroup animosity shrink to zero or reverse under human coding, the central result would not survive.

Watch

Extended reading notes

Core claim

The central discovery is an engagement asymmetry: on TikTok in the run-up to the 2025 German federal election, negative and group-hostile messages outperformed positive and identity-affirming ones. Videos coded as negative received roughly 8 to 12 percent more engagement than neutral videos, and outgroup animosity was associated with 37 to 79 percent more engagement across likes, comments, shares, and views. Disgust showed the largest effects among discrete emotions, while relatability showed the largest negative associations. The same models show that politicians from Die Linke and AfD, the two ideologically extreme parties, generated substantially more engagement than centrist politicians,

Load-bearing premise

The claim depends on the automated content labels being accurate across all 25,292 videos: if the models that tag sentiment, emotions, and outgroup animosity systematically misclassify videos by party or by popularity, the measured engagement differences could be an artifact of the annotation rather than of the content itself.

Editorial extensions

If this is right

  • If TikTok's dynamics reward anger, disgust, and outgroup hostility, political actors face a systematic incentive to produce more divisive content during campaigns.
  • The engagement penalty for positive, relatable, and identity-based messages means unifying communication strategies are likely to receive less visibility, potentially shaping the tone of election discourse.
  • Because far-left and far-right parties both post more divisive content and receive higher engagement than centrists, the platform's dynamics create an asymmetric benefit for ideological extremes.
  • The pattern extends earlier findings on negativity and outgroup animosity from text-based platforms to short-form video in a non-US context, suggesting a cross-platform regularity in political engagement.

Reading between the lines

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

  • Going beyond the paper: the reported gaps between views and active engagement are consistent with algorithmic amplification, but the observational design cannot separate user preference from platform ranking; a field experiment that posts matched positive and divisive videos from the same accounts under different recommendation conditions would test this separation.
  • Going beyond the paper: if the reward for divisiveness is a platform logic rather than a Germany-specific election artifact, similar analyses of quieter political periods or of other countries' TikTok feeds before elections should show a weaker or absent engagement gap; comparing those settings is a direct test.
  • Going beyond the paper: the authors' annotation-validation sample is small relative to the full dataset, so one concrete extension is a larger, stratified human-annotation study—oversampling by party and engagement level—to check whether measurement error in the labels attenuates or reverses the reported coefficients.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. This paper analyzes N=25,292 TikTok videos posted by German politicians, parties, and parliamentary groups between May 2024 and February 2025, combining Whisper transcripts with automated sentiment/emotion classification and LLM-based annotation of identity language, outgroup animosity, and relatability. Multilevel negative binomial models with account random effects and month fixed effects are used to estimate associations between content features and likes, comments, shares, and views. The main findings are that negative sentiment, anger, disgust, and outgroup animosity are associated with higher engagement, while positive sentiment, relatability, and identity language are associated with lower engagement; and that AfD and Die Linke post such content more often and receive higher engagement than centrist parties. The paper interprets these associations as evidence that TikTok's platform dynamics systematically reward divisive political communication.

Significance. If the central finding holds, it would be an important, policy-relevant contribution to the literature on political communication and platform incentives, extending prior work from text-based platforms to short-form video in a non-U.S. context. The dataset is unusually comprehensive, and the analysis includes several good practices: account-level random effects, month fixed effects, VIF checks, an implausible-placebo model, continuous-operationalization robustness, and a zero-inflated beta analysis of engagement per view. These are genuine strengths. However, the paper's headline claims are observational and depend on automated content labels whose validation is limited to a small, unstratified sample; the robustness checks do not address the measurement-error concern. The causal-sounding framing also goes beyond what the design can support.

major comments (3)
  1. [Methods, Eq. (1)] The construct validation is based on 20 raters on 100 videos, with ICCs ranging from 0.345 (Identity Language) to 0.895 (Sentiment); Outgroup Animosity is 0.715 and Disgust is 0.551. This establishes average agreement between LLM labels and human ratings, but it does not rule out differential misclassification with respect to engagement level or party. If automated labels are more accurate for high-engagement videos or for AfD/Die Linke content, the coefficients for Outgroup Animosity (0.401 likes, 0.583 comments, 0.448 shares; Table S3) and Negative Sentiment (0.110–0.116) could be biased. The robustness checks in SI E.1–E.4 vary the coding scheme, outcome, or model family; none of them stratifies annotation error by party or engagement. Please add a stratified validation analysis (e.g., misclassification rates by engagement tercile and party), a sensitivity analysis bounding the estima
  2. [Methods, Eq. (1)] The count outcomes are measured at the end of the observation window for videos posted between May 1, 2024 and February 23, 2025. A video posted in May 2024 has roughly nine months more time to accumulate likes, comments, shares, and views than a video posted in February 2025. Month fixed effects (β17 Month_ij) absorb average month differences but not within-month exposure time, and posting intensity changes sharply around the government collapse, parliament dissolution, and election (Fig. 1C). If content mix shifts over time, the content—and party—engagement coefficients are confounded by time at risk. Please include log(days since posting) as an exposure/offset or control for days-since-posting, and report whether the key coefficients survive.
  3. [Title, Abstract, Discussion] The title 'TikTok Rewards Divisive Political Messaging', the abstract's 'TikTok's platform dynamics systematically reward', and the Discussion's repeated 'reward' language make causal or platform-level claims. The analysis itself is observational, and the Discussion explicitly states 'we refrain from making causal claims.' The models identify associations conditional on controls, not a causal effect of content on engagement or an algorithmic reward mechanism. This is load-bearing because the policy implications (incentives for politicians, platform design) depend on whether the association is causal and on whether it is driven by algorithm amplification or user preferences. Please revise the title and abstract to associational wording and clearly specify the target quantity as observed engagement.
minor comments (4)
  1. [Drivers of Engagement, Disgust coefficient] The text reports '49.8% more views (coef: −0.404, p<0.001)' for Disgust, but Table S3 reports a positive coefficient (0.404). The sign in the main text appears to be a typo.
  2. [SI E.2] The comparison-to-XYZ paragraph contains a typo: 'and 0.080(p<0.315) for views (p<0.001)' should presumably read '0.080 (p<0.001)'.
  3. [SI C] The text describes ICC values as 'moderate to high', but an ICC of 0.345 for Identity Language is below the conventional threshold for good reliability (Koo & Li, 2016). Please acknowledge this directly and discuss its implication for the identity-language finding, which is currently null/negative.
  4. [Methods, Content Annotation] Please report version and release details for the sentiment model (tabularisai/multilingual-sentiment-analysis), the emotion model (tweedmann/pol_emo_mDeBERTa2), and gpt-4-turbo, along with any relevant timestamps, to improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: content features and engagement are measured independently, and the central claim is an empirical regression result.

full rationale

The paper's central derivation is an empirical association study, not a derivation from definitions. Content features (sentiment, discrete emotions, Identity Language, Relatability, Outgroup Animosity) are extracted from Whisper transcripts by external models (tabularisai/multilingual-sentiment-analysis, twedmann/pol_emo_mDeBERTa2, and the gpt-4-turbo prompt in SI B). Engagement outcomes are TikTok API counts of likes, comments, shares, and views. These are independent data streams; the multilevel negative binomial model in Eq. (1) estimates coefficients rather than computing them from construct definitions. No parameter is fitted to engagement and then 'predicted' back; the coefficients are estimated from the full sample, so there is no fitted-input-called-prediction pattern. The robustness checks (continuous scores, zero-inflated beta per-view models, XYZ placebo) vary codings and outcomes but none makes the result an identity. Self-citations ([65], [74], [78]) are methodological or literature-support citations and are not load-bearing for the claim that divisive content drives engagement; that claim is carried by the paper's own regression on independently measured variables. The 100-video human validation concerns measurement error and construct validity; even if automated labels were imperfect, that would be a bias/validity threat, not circularity, because the predictors are not defined in terms of the outcomes. No circular step can be identified from the paper's text.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The paper's empirical results rest on measurement and data-collection assumptions rather than on free parameters in a derivation. The hand-chosen annotation thresholds and the transcript-only content measure are the most consequential choices; the LLM validation study supports but does not fully secure them.

free parameters (5)
  • Sentiment discretization thresholds = +/-0.5 on mean sentence-level sentiment score
    Videos are coded negative (mean below -0.5), neutral (-0.5 to 0.5), or positive (above 0.5). Chosen by the authors; robustness with continuous scores is tested in SI E.3.
  • Emotion presence threshold = 0.65 probability
    An emotion is considered present in a sentence if the model probability exceeds 0.65, following Widmann & Wich (2022). Chosen by hand; affects all emotion indicators.
  • Emotion aggregation rule = Present if in at least one sentence
    Video-level binary emotion indicators. An alternative mean-score aggregation is tested in SI E.3 and reported as consistent.
  • Relatability coding scheme = Low/Medium/High via GPT-4-turbo prompt
    The three-level relatability variable depends on the prompt definition in SI B and the LLM's interpretation; validation on 100 videos shows moderate agreement.
  • Video-level sentiment aggregation = Arithmetic mean of sentence scores
    Sentence-level sentiment scores are averaged to obtain a video-level score before discretization. This is a modeling choice that affects the main categorical variable.
assumptions (5)
  • domain assumption TikTok Research API and the precompiled account list yield complete engagement metadata for all 727 relevant accounts and 25,292 videos.
    Invoked in Methods (Data Collection). If accounts or videos are missed, or if metrics are unreliable, the central estimates are biased.
  • domain assumption Audio transcripts produced by Whisper faithfully represent the words spoken in each video, and these transcripts are a sufficient proxy for video content.
    Invoked in Methods (Content Annotation) and Limitations. Visual and audio tone cues are ignored, which could confound the content-engagement relationship.
  • domain assumption The LLM annotations (GPT-4-turbo) and the emotion/sentiment models classify content features with validity that generalizes from the 100-video, 20-rater validation study to the full corpus.
    The user study (SI C) is the only evidence for construct validity; low ICC values for some constructs (e.g., Identity Language 0.345) make this a nontrivial assumption.
  • standard math Negative binomial regression with account random intercepts and month fixed effects identifies the conditional associations of content features with engagement.
    Standard statistical modeling assumptions; invoked in Methods (Regression Analysis).
  • domain assumption Engagement metrics (likes, comments, shares, views) reflect user and algorithmic reward rather than being dominated by unmeasured factors such as follower growth or content promotion.
    The per-view analyses partially address this, but the title claim of platform-level reward relies on this assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TikTok Rewards Divisive Political Messaging During the 2025 German Federal Election." pith.science (2026). https://pith.science/paper/7ZVPMM6R

@misc{pith2026250910336,
  author       = {Pith},
  title        = {Pith review of: TikTok Rewards Divisive Political Messaging During the 2025 German Federal Election},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7ZVPMM6R}},
  note         = {Machine review of arXiv:2509.10336}
}
read the original abstract

Short-form video platforms like TikTok reshape how politicians communicate and have become important tools for electoral campaigning. Yet it remains unclear what kinds of political messages gain traction in these fast-paced, algorithmically curated environments, which are particularly popular among younger audiences. In this study, we use computational content analysis to analyze a comprehensive dataset of N=25,292 TikTok videos posted by German politicians in the run-up to the 2025 German federal election. Our empirical analysis shows that videos expressing negative emotions (e.g., anger, disgust) and outgroup animosity were significantly more likely to generate engagement than those emphasizing positive emotion, relatability, or identity. Furthermore, ideologically extreme parties (on both sides of the political spectrum) were both more likely to post this type of content and more successful in generating engagement than centrist parties. Taken together, these findings suggest that TikTok's platform dynamics systematically reward divisive over unifying political communication, thereby potentially benefiting extreme actors more inclined to capitalize on this logic.

Figures

Figures reproduced from arXiv: 2509.10336 by the authors.

Figure 1
Figure 1. Data overview. (A) An example of a campaign post on TikTok during the 2025 German federal election. (B) Number of TikTok posts per political party and the average number of videos per account in our observation period from May 1, 2024 to February 23, 2025. (C) The daily number of videos per party. (D) The share of posts (in %) per party expressing positive and negative sentiment. (E) The share of posts (in %) per pa… view at source ↗
Figure 2
Figure 2. Estimation results. Multilevel negative binomial regression estimating the effects of (A) content characteristics (sentiment, discrete emotions, social identification), (B) political party, and (C) political alignment on the number of likes (circle), comments (triangle), shares (diamond), and views (square) on TikTok. Shown are the coefficient estimates with 95% CIs. Account-specific random effects and monthly fixed… view at source ↗

Discussion (0). Sign in 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. Algorithmic Audit of Personalisation Drift in Polarising Topics on TikTok

    cs.IR 2026-03 conditional novelty 6.0 of 10

    TikTok personalises strongly, neutralises climate/vaccine/flat-earth content toward safe neutral topics, but sustains and often stance-reinforces US politics—with a tilt toward the oppose stance when both sides are seeded.

Reference graph

Works this paper leans on

81 extracted references · 3 linked inside Pith · cited by 1 Pith paper

  1. [1]

    & López-García, X

    Vázquez-Herrero, J., Negreira-Rey, M.-C. & López-García, X. Let’s Dance the News! How the News Media Are Adapting to the Logic of Tiktok.Journalism23, 1717–1735 (2022)

    2022

  2. [2]

    How Publishers Are Learning to Create and Distribute News on Tiktok

    Newman, N. How Publishers Are Learning to Create and Distribute News on Tiktok. Tech. Rep., Reuters Institute for the Study of Journalism (2022)

    2022

  3. [3]

    & Zulli, D

    Zulli, D. & Zulli, D. J. Extending the Internet Meme: Conceptualizing Technological Mime- sis and Imitation Publics on the TikTok Platform.New Media & Society24, 1872–1890 (2020)

    2020

  4. [4]

    InProceedings of the 2024 CHI Conference on Human Factors in Computing Systems, 1–16 (2024)

    Zannettou, S.et al.Analyzing User Engagement with Tiktok’s Short Format Video Recom- mendations Using Data Donations. InProceedings of the 2024 CHI Conference on Human Factors in Computing Systems, 1–16 (2024)

    2024

  5. [5]

    & Urman, A

    Boeker, M. & Urman, A. An Empirical Investigation of Personalization Factors on TikTok. InProceedings of the ACM Web Conference 2022, 2298–2309 (2022)

    2022

  6. [6]

    & Quattrociocchi, W

    Sangiorgio, E., Cinelli, M., Cerqueti, R. & Quattrociocchi, W. Followers Do Not Dictate the Virality of News Outlets on Social Media.PNAS Nexus3, pgae257 (2024)

    2024

  7. [7]

    TikTok (2024)

    Die Bundesbeauftrage für den Datenschutz und die Informationsfreiheit. TikTok (2024). URL https://www.bfdi.bund.de/DE/Buerger/Inhalte/Telemedien/TikTok.html

    2024

  8. [8]

    TikTok Users in Germany in 2024, by Age Group (2025)

    Statista. TikTok Users in Germany in 2024, by Age Group (2025). URL https://www.statista. com/statistics/1337425/tiktok-users-age-group-germany/

    arXiv 2024

Show all 81 references
  1. [9]

    Cheng, Z. & Li, Y . Like, Comment, and Share on TikTok: Exploring the Effect of Sentiment and Second-Person View on the User Engagement with Tiktok News Videos.Social Science Computer Review42, 201–223 (2023). 22

    2023

  2. [10]

    Biswas, A., Lin, Y .-R., Tai, Y . C. & Desmarais, B. A. Political Elites in the Attention Econ- omy: Visibility over Civility and Credibility? InProceedings of the International AAAI Conference on Web and Social Media, 19, 241–258 (2025)

    2025

  3. [11]

    Seizing the Moment: The Presidential Campaigns’ Use of Twitter during the 2012 Electoral Cycle.New Media & Society18, 1473–1490 (2016)

    Kreiss, D. Seizing the Moment: The Presidential Campaigns’ Use of Twitter during the 2012 Electoral Cycle.New Media & Society18, 1473–1490 (2016)

    2012

  4. [12]

    & Van Bavel, J

    Rathje, S. & Van Bavel, J. J. The Psychology of Virality.Trends in Cognitive Sciences (2025)

    2025

  5. [13]

    & Dang-Xuan, L

    Stieglitz, S. & Dang-Xuan, L. Political Communication and Influence through Microblog- ging—An Empirical Analysis of Sentiment in Twitter Messages and Retweet Behavior. In 2012 45th Hawaii International Conference on System Sciences, 3500–3509 (IEEE, 2012)

    2012

  6. [14]

    A., Parnet, O

    Nulty, P., Theocharis, Y ., Popa, S. A., Parnet, O. & Benoit, K. Social Media and Political Communication in the 2014 Elections to the European Parliament.Electoral Studies44, 429–444 (2016)

    2014

  7. [15]

    & Enjolras, B

    Wollebæk, D., Karlsen, R., Steen-Johnsen, K. & Enjolras, B. Anger, Fear, and Echo Chambers: The Emotional Basis for Online Behavior.Social Media+ Society5, 2056305119829859 (2019)

    2019

  8. [16]

    Steffens, N. K. & Haslam, S. A. Power through ‘Us’: Leaders’ Use of We-Referencing Language Predicts Election Victory.PloS One8, e77952 (2013)

    2013

  9. [17]

    Leach, S.et al.Politicians’ Use of National Identity Rhetoric on Social Media Predicts Engagement and Electoral Success.PsyArXiv(2024)

    2024

  10. [18]

    Nettasinghe, B., Percus, A. G. & Lerman, K. How Out-Group Animosity Can Shape Partisan Divisions: A Model of Affective Polarization.PNAS Nexuspgaf082 (2025). 23

    2025

  11. [19]

    Rathje, S., Van Bavel, J. J. & van der Linden, S. Out-Group Animosity Drives Engagement on Social Media.Proceedings of the National Academy of Sciences (PNAS)118, e2024292118 (2021)

    2021

  12. [20]

    Hackenburg, K., Brady, W. J. & Tsakiris, M. Mapping Moral Language on US Presidential Primary Campaigns Reveals Rhetorical Networks of Political Division and Unity.PNAS Nexus2, pgad189 (2023)

    2023

  13. [21]

    & Chueri, J

    Törnberg, P. & Chueri, J. Elite Political Discourse Has Become More Toxic in Western Countries.arXiv preprint arXiv:2503.22411(2025)

    2025 arXiv

  14. [22]

    E.et al.Negativity Drives Online News Consumption.Nature Human Be- haviour7, 812–822 (2023)

    Robertson, C. E.et al.Negativity Drives Online News Consumption.Nature Human Be- haviour7, 812–822 (2023)

    2023

  15. [23]

    & Dang-Xuan, L

    Stieglitz, S. & Dang-Xuan, L. Emotions and Information Diffusion in Social Me- dia—Sentiment of Microblogs and Sharing Behavior.Journal of Management Information Systems29, 217–248 (2013)

    2013

  16. [24]

    & Casas, A

    Yu, X., Wojcieszak, M. & Casas, A. Partisanship on Social Media: In-Party Love Among American Politicians, Greater Engagement With Out-Party Hate Among Ordinary Users. Political Behavior46, 799–824 (2024)

    2024

  17. [25]

    & Liu, J

    Song, J. & Liu, J. A Data-Driven Short Video International Communication Model Based on Indicator System Communication Network and Attention Bilstm Neural Network.Scientific Reports14, 14532 (2024)

    2024

  18. [26]

    & Gao, L

    Gao, Y ., Liu, F. & Gao, L. Echo Chamber Effects on Short Video Platforms.Scientific Reports13, 6282 (2023)

    2023

  19. [27]

    Exploring the Promotion of Musical Intangible Cultural Heritage under TikTok Short Videos.Scientific Reports15, 21772 (2025)

    Cao, H. Exploring the Promotion of Musical Intangible Cultural Heritage under TikTok Short Videos.Scientific Reports15, 21772 (2025). 24

    2025

  20. [28]

    C., Papakyriakopoulos, O

    Medina Serrano, J. C., Papakyriakopoulos, O. & Hegelich, S. Dancing to the Partisan Beat: A First Analysis of Political Communication on TikTok. InProceedings of the 12th ACM Conference on Web Science, 257–266 (2020)

    2020

  21. [29]

    & V otta, F

    Guinaudeau, B., Munger, K. & V otta, F. Fifteen Seconds of Fame: TikTok and the Supply Side of Social Video.Computational Communication Research4, 463–485 (2022)

    2022

  22. [30]

    Ibrahim, H.et al.Tiktok’s Recommendations Skewed towards Republican Content during the 2024 US Presidential Race.arXiv preprint arXiv:2501.17831(2025)

    2024 arXiv

  23. [31]

    & Gil de Zúñiga, H

    Li, Y ., Cheng, Z. & Gil de Zúñiga, H. TikTok’s Political Landscape: Examining Echo Chambers and Political Expression Dynamics.New Media & Society14614448251339755 (2025)

    2025

  24. [32]

    & Abidin, C

    Zeng, J. & Abidin, C. ‘# okboomer, Time to Meet the Zoomers’: Studying the Memefication of Intergenerational Politics on TikTok. InThe Playful Politics of Memes, 93–115 (Routledge, 2023)

    2023

  25. [33]

    & Marín-Lladó, C

    Cervi, L. & Marín-Lladó, C. What Are Political Parties doing on TikTok? The Spanish Case. Profesional de la Información30(2021)

    2021

  26. [34]

    & Fahmy, S

    Zamora-Medina, R., Suminas, A. & Fahmy, S. S. Securing the Youth Vote: A Comparative Analysis of Digital Persuasion on TikTok among Political Actors.Media and Communication 11, 218–231 (2023)

    2023

  27. [35]

    The Use of TikTok for Political Campaigning in Canada: The Case of Jagmeet Singh.Social Media+ Society9, 20563051231157604 (2023)

    Moir, A. The Use of TikTok for Political Campaigning in Canada: The Case of Jagmeet Singh.Social Media+ Society9, 20563051231157604 (2023)

    2023

  28. [36]

    Stimmanteile bei 18-24-Jährigen im Vergleich zu 2021 (2025)

    Tagesschau. Stimmanteile bei 18-24-Jährigen im Vergleich zu 2021 (2025). URL https://www.tagesschau.de/wahl/archiv/2025-02-23-BT-DE/charts/umfrage-alter/chart_ 1862482.shtml. 25

    2021

  29. [37]

    Feuerriegel, S.et al.Using Natural Language Processing to Analyse Text Data in Behavioural Science.Nature Reviews Psychology4, 96–111 (2025)

    2025

  30. [38]

    J., Wills, J

    Brady, W. J., Wills, J. A., Jost, J. T., Tucker, J. A. & Van Bavel, J. J. Emotion Shapes the Diffusion of Moralized Content in Social Networks.Proceedings of the National Academy of Sciences (PNAS)114, 7313–7318 (2017)

    2017

  31. [39]

    & Baronchelli, A

    Falkenberg, M., Zollo, F., Quattrociocchi, W., Pfeffer, J. & Baronchelli, A. Patterns of Par- tisan Toxicity and Engagement Reveal the Common Structure of Online Political Communi- cation across Countries.Nature Communications15, 9560 (2024)

    2024

  32. [40]

    M.et al.How Do Social Media Feed Algorithms Affect Attitudes and Behavior in an Election Campaign?Science381, 398–404 (2023)

    Guess, A. M.et al.How Do Social Media Feed Algorithms Affect Attitudes and Behavior in an Election Campaign?Science381, 398–404 (2023)

    2023

  33. [41]

    & Neuberger, C

    Dang-Xuan, L., Stieglitz, S., Wladarsch, J. & Neuberger, C. An Investigation of Influentials and the Role of Sentiment in Political Communication on Twitter during Election Periods. In Social Media and Election Campaigns, 168–198 (Routledge, 2017)

    2017

  34. [42]

    Bossetta, M. The Digital Architectures of Social Media: Comparing Political Campaigning on Facebook, Twitter, Instagram, and Snapchat in the 2016 US Election.Journalism & Mass Communication Quarterly95, 471–496 (2018)

    2016

  35. [43]

    & Ijaz, M

    Atiq, M., Abid, G., Anwar, A. & Ijaz, M. F. Influencer Marketing on Instagram: A Sequential Mediation Model of Storytelling Content and Audience Engagement via Relatability and Trust.Information13, 345 (2022)

    2022

  36. [44]

    & Zheng, X

    Liu, X. & Zheng, X. The Persuasive Power of Social Media Influencers in Brand Credibility and Purchase Intention.Humanities and Social Sciences Communications11, 1–12 (2024). 26

    2024

  37. [45]

    & De Vreese, C

    Aalberg, T., Strömbäck, J. & De Vreese, C. H. The Framing of Politics as Strategy and Game: A Review of Concepts, Operationalizations and Key Findings.Journalism13, 162– 178 (2012)

    2012

  38. [46]

    Enli, G.Mediated Authenticity: How the Media Constructs Reality(Peter Lang Verlag, 2015)

    2015

  39. [47]

    Hartman, R.et al.Interventions to Reduce Partisan Animosity.Nature Human Behaviour6, 1194–1205 (2022)

    2022

  40. [48]

    A.et al.Exposure to Opposing Views on Social Media Can Increase Political Polar- ization.Proceedings of the National Academy of Sciences (PNAS)115, 9216–9221 (2018)

    Bail, C. A.et al.Exposure to Opposing Views on Social Media Can Increase Political Polar- ization.Proceedings of the National Academy of Sciences (PNAS)115, 9216–9221 (2018)

    2018

  41. [49]

    Garrett, K. N. & Bankert, A. The Moral Roots of Partisan Division: How Moral Conviction Heightens Affective Polarization.British Journal of Political Science50, 621–640 (2020)

    2020

  42. [50]

    Vaidhyanathan, S.Antisocial Media: How Facebook Disconnects Us and Undermines Democracy(Oxford University Press, 2018)

    2018

  43. [51]

    & Hertwig, R

    Lorenz-Spreen, P., Oswald, L., Lewandowsky, S. & Hertwig, R. A Systematic Review of Worldwide Causal and Correlational Evidence on Digital Media and Democracy.Nature Human Behaviour7, 74–101 (2023)

    2023

  44. [52]

    & Hall, P

    Gidron, N. & Hall, P. A. Populism as a Problem of Social Integration.Comparative Political Studies53, 1027–1059 (2020)

    2020

  45. [53]

    & Yang, J

    Munzert, S., Traunmüller, R., Barberá, P., Guess, A. & Yang, J. Citizen Preferences for Online Hate Speech Regulation.PNAS Nexus4, pgaf032 (2025)

    2025

  46. [54]

    Proceedings of the National Academy of Sciences (PNAS)121, e2308950121 (2024)

    Rathje, S.et al.GPT Is an Effective Tool for Multilingual Psychological Text Analysis. Proceedings of the National Academy of Sciences (PNAS)121, e2308950121 (2024). 27

    2024

  47. [55]

    Chen, H., Bisbee, J., Tucker, J. A. & Nagler, J. Labeling Social Media Posts: Does Show- ing Coders Multimodal Content Produce Better Human Annotation, and a Better Machine Classifier?Political Science Research and Methods1–13 (2025)

    2025

  48. [56]

    Milli, S.et al.Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media.PNAS Nexus4, pgaf062 (2025)

    2025

  49. [57]

    Combs, A.et al.Reducing Political Polarization in the United States with a Mobile Chat Platform.Nature Human Behaviour7, 1454–1461 (2023)

    2023

  50. [58]

    Junge favorisieren AfD und Linke, Alte die Union (2025)

    Statista. Junge favorisieren AfD und Linke, Alte die Union (2025). URL https://de.statista. com/infografik/33994/stimmenanteile-bei-der-bundestagswahl-2025-nach-altersgruppen/

    2025

  51. [59]

    Deutsche Politik Auf Tik Tok (2025)

    Fuchs, M. Deutsche Politik Auf Tik Tok (2025). URL https://docs.google.com/spreadsheets/ d/13yIKz_RdUe1IAExjJKiPlgzkD-T0ZXRbEnU2nIkDJFU

    2025

  52. [60]

    Tiktok Research API

    TikTok. Tiktok Research API. https://research.tiktok.com (n.d.). Accessed: 2025-06-26

    2025

  53. [61]

    yt-dlp: A Youtube-dl Fork with Additional Features and Fixes

    yt-dlp contributors. yt-dlp: A Youtube-dl Fork with Additional Features and Fixes. https: //github.com/yt-dlp/yt-dlp (2021)

    2021

  54. [62]

    Whisper: Open Source Automatic Speech Recognition (asr) System

    OpenAI. Whisper: Open Source Automatic Speech Recognition (asr) System. https://github. com/openai/whisper (2022)

    2022

  55. [63]

    W., Clark, T.et al.Robust Speech Recognition via Large-Scale Weak Supervision (2022)

    Radford, A., Gao, J. W., Clark, T.et al.Robust Speech Recognition via Large-Scale Weak Supervision (2022). URL https://cdn.openai.com/papers/whisper.pdf

    2022

  56. [64]

    Gunning, R.The Technique of Clear Writing(McGraw-Hill, 1952)

    1952

  57. [65]

    & Pröllochs, N

    Solovev, K. & Pröllochs, N. References to Unbiased Sources Increase the Helpfulness of Community Fact-Checks.arXiv preprint arXiv:2503.10560(2025). 28

    2025 arXiv

  58. [66]

    In2023 IEEE 31st International Requirements Engineering Conference (RE), 111–122 (IEEE, 2023)

    Sihag, M.et al.A Data-Driven Approach for Finding Requirements Relevant Feedback from TikTok and YouTube. In2023 IEEE 31st International Requirements Engineering Conference (RE), 111–122 (IEEE, 2023)

    2023

  59. [67]

    Welbourne, D. J. & Grant, W. J. Science Communication on YouTube: Factors that Affect Channel and Video Popularity.Public Understanding of Science25, 706–718 (2016)

    2016

  60. [68]

    J., Lelkes, Y

    Hopkins, D. J., Lelkes, Y . & Wolken, S. The Rise of and Demand for Identity-Oriented Media Coverage.American Journal of Political Science69, 483–500 (2025)

    2025

  61. [69]

    Rasmussen, J., Rathje, S., Van Bavel, J. J. & Robertson, C. Negativity and Identity Language Have Additive Effects on Online News Consumption.OSF Preprint(2025)

    2025

  62. [70]

    & Wich, M

    Widmann, T. & Wich, M. Creating and Comparing Dictionary, Word Embedding, and Transformer-Based Models to Measure Discrete Emotions in German Political Text.Po- litical Analysis31, 626–641 (2022)

    2022

  63. [71]

    Hardin, J. W. & Hilbe, J. M.Generalized Linear Models and Extensions(Stata Press, 2007)

    2007

  64. [72]

    Koo, T. K. & Li, M. Y . A Guideline of Sselecting and Reporting Intraclass Correlation Coefficients for Reliability Research.Journal of Chiropractic Medicine15, 155–163 (2016)

    2016

  65. [73]

    W., Cruz, N

    Burton, J. W., Cruz, N. & Hahn, U. Reconsidering Evidence of Moral Contagion in Online Social Networks.Nature Human Behaviour5, 1629–1635 (2021)

    2021

  66. [74]

    & Pröllochs, N

    Solovev, K. & Pröllochs, N. Moralized Language Predicts Hate Speech on Social Media. PNAS Nexus2, pgac281 (2023)

    2023

  67. [75]

    InProceedings of the 17th ACM Conference on Recommender Systems, 540–550 (2023)

    Pan, Y .et al.Understanding and Modeling Passive-Negative Feedback for Short-Video Se- quential Recommendation. InProceedings of the 17th ACM Conference on Recommender Systems, 540–550 (2023). 29

    2023

  68. [76]

    J., McLoughlin, K., Doan, T

    Brady, W. J., McLoughlin, K., Doan, T. N. & Crockett, M. J. How Social Learning Amplifies Moral Outrage Expression in Online Social Networks.Science Advances7, eabe5641 (2021)

    2021

  69. [77]

    & Gray, K

    Dillion, D., Puryear, C., Li, L., Chiquito, A. & Gray, K. National Politics Ignites more Talk of Morality and Power than Local Politics.Proceedings of the National Academy of Sciences (PNAS)3, pgae345 (2024)

    2024

  70. [78]

    & Pröllochs, N

    Solovev, K. & Pröllochs, N. Moral Emotions Shape the Virality of COVID-19 Misinforma- tion on Social Media. InProceedings of the ACM Web Conference 2022, 3706–3717 (2022). 30 Supplementary Materials Contents A Descriptive Statistics 32 B LLM-Based Annotation 33 C Validation Us...

    2022

  71. [79]

    identity_language

    Identity Language •Definition: Identify whether the caption explicitly refers to a political, social, or demo- graphic group commonly associated with German political discourse. These identities may be defined by political affiliation, ethnicity, religion, gender, age cohort, ...

  72. [80]

    outgroup_animosity

    Outgroup Animosity •Definition: Outgroup animosity refers to explicit negative characterizations of political or ideological opponents. This includes emotionally charged or polarizing language that reinforces social identity divides. •Indicators include: –Direct hostile refere...

  73. [81]

    relatability

    Relatability •Definition: Relatability refers to how likely it is that an average German voter would perceive the content as authentic, familiar, and emotionally resonant. It is a measure of connection, identification, and trust. •Key indicators: –Authenticity: Honest, direct,...

    2025

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.