Pith. sign in

REVIEW 4 major objections 6 minor 67 references

Artificial Intelligence and Civil Discourse: How LLMs Moderate Climate Change Conversations

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Large language models reply to climate-change posts with predominantly neutral emotion and lower emotional intensity than human users, a pattern the authors interpret as an inherent moderating capacity.

desk verdict The descriptive measurement is probably real, but the design cannot support the claim that LLMs inherently moderate discourse—there is no human-reply control. read the letter →

arxiv 2506.12077 v1 pith:RRS5IIRJ submitted 2025-06-07 cs.CY cs.CL

Wenlu Fan , Wentao Xu This is my paper
classification cs.CYcs.CL
keywords largelanguagemodelsclimatechangesocialmediasentimentanalysisemotionalneutralityintensityaffectivepolarizationonlinemoderation
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 asks how large language models behave emotionally when they reply to real climate-change posts on Twitter and Reddit. It reports that replies from all five models tested — Gemma, Llama 3, Llama 3.3, GPT-4o, and Claude 3.5 — are most often classified as neutral, regardless of how emotional the original post was. It also reports that LLM replies carry lower emotion intensity scores than the human posts, across nearly all emotion categories and both platforms. The authors take these two patterns as evidence that LLMs have inherent moderating capacities that could cool polarized online climate discussions. The value of the claim, if it holds, is a practical one: AI systems might be designed to dampen hostility in public discourse, though the paper itself notes persistent neutrality could be read as detachment.

What carries the argument

The central mechanism is the emotion-transition analysis: for each of five LLMs, the paper pairs a human post with the model's reply, runs both through a fine-tuned emotion classifier that returns a probability for each of seven emotion categories, and builds matrices showing how each human emotion label flows into the reply's most probable label. The complementary machinery is statistical comparison of emotion intensity scores — the classifier's category probabilities treated as intensities — across six groups (five models plus original human text) using analysis of variance and pairwise post-hoc comparisons. These two tools convert a simple observation about style into the paper's central contrast: LLM replies are both more neutral in label and lower in intensity than the human posts that prompted them.

What would settle it

Run the same emotion classifier on human replies written to the same climate posts; if human replies show neutral shares and intensity scores statistically indistinguishable from the LLM replies, the claim that LLMs have a distinctive moderating effect collapses.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that LLMs, when prompted only with a user's climate-related post, respond with text dominated by the neutral emotion category and with lower emotion probabilities than the human-authored posts they answer. This pattern is quantified with fine-grained emotion labels (joy, surprise, neutral, anger, disgust, fear, sadness) and tested with analysis of variance followed by pairwise post-hoc comparisons. The emotion-transition tables show, for example, that a large share of angry, disgusted, or sad human comments receive neutral replies from every model. The authors interpret this as an emotional cooling or emotional debiasing effect: instead of matching the emotional charge of the input, LLMs dampen it, and that dampening is the proposed mechanism by which LLMs could moderate public debate.

Load-bearing premise

The load-bearing assumption is that the emotional difference between LLM replies and original human posts is caused by the LLM's own style, not by the fact that any reply to an emotionally charged post tends to be calmer than the post itself.

Editorial extensions

If this is right

  • Across all five models and both platforms, replies to climate posts are predominantly neutral, even when the source post is angry, disgusted, or sad.
  • Emotion intensity scores for LLM replies are significantly lower than for human posts for most emotions, with platform-specific differences such as fear on Twitter and disgust on Reddit.
  • The neutral tendency holds for both open-source and commercial models, although within the same family, Llama 3.3 produces more neutral replies than Llama 3, which the authors attribute to larger training data.
  • If the moderating effect is real, LLMs could be placed in discussion threads during emotional escalation to inject balanced, low-intensity responses.
  • Because the paper only analyzes first-round replies, any real-world moderating effect on subsequent conversation remains untested.

Reading between the lines

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

  • The design does not include a human-reply control group: the paper compares LLM replies to the original posts, not to human replies to the same posts, so the observed calmness may be a general property of replies rather than a unique LLM trait; a direct human-reply control would separate the two.
  • If LLM neutrality is read by users as evasiveness rather than moderation, deploying LLMs as stabilizers could suppress the expressions of concern and moral stance that drive climate activism — a risk the authors raise but do not test.
  • A testable extension follows from the dampening claim: inserting an LLM reply into a heated thread should measurably reduce the emotional intensity of subsequent human comments compared with a thread without the LLM.
  • The emotion classifier's neutral category may absorb much of the low-intensity variation, so the distinction between neutral and low intensity needs an ablation that removes the neutral category before recomputing intensity scores.
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 manuscript examines whether LLM-generated replies to climate change posts on Twitter and Reddit are more emotionally neutral and lower in emotional intensity than human-generated content. Using a seven-class emotion classifier, the authors compare replies generated by five LLMs (Gemma, Llama 3, Llama 3.3, GPT-4o, Claude 3.5) with the original human posts that served as prompts. They report neutral-dominant emotion transitions and statistically significant lower intensity scores for LLM replies, and conclude that LLMs possess inherent moderating capabilities. The paper includes descriptive transition matrices, ANOVA tables, and post-hoc comparisons.

Significance. The descriptive finding that LLM replies differ from the original posts is plausible, and the study has useful strengths: large time-stratified corpora, multiple open-source and commercial models, and cross-platform replication. However, the central inference about LLM 'moderating capabilities' is not identified by the current design, because LLM replies are compared with original posts rather than with human replies to the same posts. The study also lacks classifier validation on LLM text and contains statistical reporting errors. If the authors can supply a human-reply baseline and correct the analysis, the result would be a meaningful contribution to understanding AI behavior in public discourse.

major comments (4)
  1. [Section III.A and IV.B] The comparison in Section IV.B between LLM-generated replies and 'human text' uses the original human posts as the human benchmark, not human replies to those posts. Because the LLM texts are replies generated from those posts, any observed difference conflates author identity (LLM vs human) with utterance position (reply vs initiating post). Since Section III.B states that both posts and comments were collected on both platforms, the authors should add a matched analysis using human comments on the same source posts. Without such a baseline, the central claim of LLM-specific moderation in the Abstract and Section VII is not identified.
  2. [Tables II and III; Section IV.B] Tables II and III are captioned 'Tukey's post-hoc test,' but their footnotes state that the stars denote p-values of the Mann-Whitney U test. These are different procedures with different assumptions and inferential targets, and the current reporting makes the quantitative support for lower emotional intensity unclear. In addition, Section IV.B says ANOVA revealed significant differences across 'nine groups,' although the design comprises six groups (five LLMs plus human text). The authors should correct the test labels and report effect sizes in addition to p-values.
  3. [Abstract and Section VII] The Abstract and Section VII claim that LLMs create 'a stabilizing effect in conversations' and possess 'inherent moderating capabilities.' Section VI explicitly states that the study cannot draw conclusions about the actual impact of LLM neutral replies on subsequent user interaction behaviors or the overall discussion atmosphere. The causal and dynamic language in the abstract and conclusion is therefore not supported by the first-turn reply design. Either the claims should be reframed as potential effects, or the study should include multi-turn interaction data.
  4. [Section III.C and IV.B] The emotion classifier from Hugging Face (DistilRoBERTa fine-tuned on human text) is used both to assign the 'neutral' label and to compute the intensity scores that drive the group comparisons. Its measurement invariance across human-authored and LLM-generated text is not established. Without validation on LLM outputs (e.g., human agreement or classification performance on generated text), the observed differences may partly reflect classifier bias rather than genuine emotional differences. The authors should report such validation or explicitly restrict their conclusions to classifier-defined emotion.
minor comments (6)
  1. [Section IV.A, first paragraph] The sentence beginning 'For other emotion categories...' says texts turned into the 'angry' label with proportions 58%, 55%, 57%, 60%, and 68%, which contradicts the surrounding claim of neutral moderation; the intended label is likely 'neutral.'
  2. [Figures 3 and 4 captions] The captions name four models (Gemma, Llama, Claude, GPT) but the panels a–e correspond to five models; please clarify the panel-to-model mapping.
  3. [Section III.B] The statement 'We used histograms to understand the basic distribution of data' is vague; Figure 2 shows daily counts, not histograms of the sampled data.
  4. [Section IV.B] The text says 'ANOVA revealed statistically significant differences ... across nine groups on Twitter'; the design has six groups and Table I correctly lists six platform-emotion combos, so the 'nine' is a typo.
  5. [Section I] 'Rational debate was replaced by artisanship' appears to be a typo; consider 'artificiality' or 'antagonism.'
  6. [Section III.C] The paper does not mention a formal citation for the emotion model; the Hugging Face URL is given but a proper reference would help reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the emotion comparisons are operationalized with an external classifier and the central 'moderation' inference is interpretive, not derived from the inputs by construction.

full rationale

The paper's central empirical finding is that LLM-generated replies receive 'neutral' labels and lower emotion-class probability scores than the original human posts, as measured by the external j-hartmann/emotion-english-distilroberta-base model. This is a measurement result, not a circular derivation: the classifier is an independent instrument, and the comparison is between two different text populations. The fact that 'neutral' is defined by the same classifier used in the comparison is a measurement-validity issue, not a self-definitional reduction; the paper does not define LLM neutrality as the classifier output and then claim the definition as a discovery. The stronger concern is inferential: the paper compares LLM replies to original human posts, not to human replies to the same posts, so the observed calmness could be attributable to reply position rather than to LLM authorship. This is a missing-control-group confound and a soundness limitation, not a circularity pattern. The paper itself acknowledges in Section VI that it is 'impossible to draw conclusions about the actual impact of LLM neutral replies on subsequent user interaction behaviors,' which further indicates that the 'stabilizing effect' claim is an interpretive extrapolation rather than a conclusion forced by the experimental construction. No load-bearing self-citation chain appears. References such as [50] and [51], which may share authors with this paper, are cited only as background examples of agent-based LLM simulations and are not used to justify the paper's emotion-measurement or moderation claims. No ansatz is smuggled in via citation, and no uniqueness theorem is invoked. The derivation chain therefore does not reduce to its own inputs; the descriptive comparison is self-contained, and the interpretive claim is clearly marked as a suggestion rather than a derived necessity.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical or conceptual entities. Its central claim rests on the validity of the emotion classifier as an intensity measure, the realism of the single-turn reply setup, and the absence of a human-reply control. The sample size quotas are design choices rather than fitted parameters.

free parameters (1)
  • Monthly sample sizes = 200 Twitter entries, 100 Reddit entries per month
    Chosen by the authors for time-stratified sampling. Not fitted to data, but determines dataset composition and the precision of all reported statistics.
assumptions (4)
  • domain assumption The DistilRoBERTa emotion classifier's output probabilities measure emotional intensity.
    Section IV.B treats normalized class probabilities as intensity scores, citing [64]; if these probabilities are miscalibrated or style-sensitive, the intensity comparison is compromised.
  • domain assumption LLM chat completion without explicit instruction simulates realistic social media replies.
    Section III.A uses the user post as the prompt; default system prompts or chat templates may inject their own style, such as helpfulness and neutrality.
  • domain assumption Single-turn replies are informative about moderation of conversations.
    Section VI admits no multi-round interactions, but moderation is a conversational dynamic that cannot be established from one reply.
  • domain assumption Sampled human text is authored by humans, not bots.
    Section VI acknowledges social bots may be present in the Twitter and Reddit data, which could bias the human baseline.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Artificial Intelligence and Civil Discourse: How LLMs Moderate Climate Change Conversations." pith.science (2026). https://pith.science/paper/RRS5IIRJ

@misc{pith2026250612077,
  author       = {Pith},
  title        = {Pith review of: Artificial Intelligence and Civil Discourse: How LLMs Moderate Climate Change Conversations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RRS5IIRJ}},
  note         = {Machine review of arXiv:2506.12077}
}
read the original abstract

As large language models (LLMs) become increasingly integrated into online platforms and digital communication spaces, their potential to influence public discourse - particularly in contentious areas like climate change - requires systematic investigation. This study examines how LLMs naturally moderate climate change conversations through their distinct communicative behaviors. We conduct a comparative analysis of conversations between LLMs and human users on social media platforms, using five advanced models: three open-source LLMs (Gemma, Llama 3, and Llama 3.3) and two commercial systems (GPT-4o by OpenAI and Claude 3.5 by Anthropic). Through sentiment analysis, we assess the emotional characteristics of responses from both LLMs and humans. The results reveal two key mechanisms through which LLMs moderate discourse: first, LLMs consistently display emotional neutrality, showing far less polarized sentiment than human users. Second, LLMs maintain lower emotional intensity across contexts, creating a stabilizing effect in conversations. These findings suggest that LLMs possess inherent moderating capacities that could improve the quality of public discourse on controversial topics. This research enhances our understanding of how AI might support more civil and constructive climate change discussions and informs the design of AI-assisted communication tools.

Figures

Figures reproduced from arXiv: 2506.12077 by the authors.

Figure 1
Figure 1. Experimental pipeline of emotion tendency for LLMs. Our experimental framework begins with human text input to five LLMs , which perform reply [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Daily data amount of Twitter and Reddit. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Emotional Transition of LLM Replies in Reddit Comments. Panels [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Emotional Transition of LLM Replies in Twitter Comments. Panels [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

67 extracted references · 57 canonical work pages

  1. [1]

    Climate Change and Society,

    T. Dietz, R. L. Shwom, and C. T. Whitley, “Climate Change and Society,” Annual Review of Sociology , vol. 46, no. V olume 46, 2020, pp. 135–158, 2020, publisher: Annual Reviews Type: Journal Article. [Online]. Available: https://www.annualreviews.org/content/journals/10. 1146/annurev-soc-121919-054614

    work page 2020

  2. [2]

    The Framing of Climate Change in Canadian, American, and International Newspapers: A Media Propaganda Model Analysis1,

    J. E. Good, “The Framing of Climate Change in Canadian, American, and International Newspapers: A Media Propaganda Model Analysis1,” Canadian Journal of Communication , Jun. 2008, publisher: University of Toronto Press. [Online]. Available: https: //cjc.utppublishing.com/doi/10.22230/cjc.2008v33n2a2017

    work page doi:10.22230/cjc.2008v33n2a2017 2008

  3. [3]

    Linking Adaptation and Mitigation in Climate Change Policy,

    S. Kane and J. F. Shogren, “Linking Adaptation and Mitigation in Climate Change Policy,” Climatic Change , vol. 45, no. 1, pp. 75–102, Apr. 2000. [Online]. Available: https://doi.org/10.1023/A: 1005688900676

    doi:10.1023/a:

  4. [4]

    Climate change 2021: the physical science basis,

    V . Masson-Delmotte, P. Zhai, A. Pirani, S. L. Connors, C. P ´ean, S. Berger, N. Caud, Y . Chen, L. Goldfarb, M. Gomis et al. , “Climate change 2021: the physical science basis,” Contribution of working group I to the sixth assessment report of the intergovernmental panel on climate change, vol. 2, no. 1, p. 2391, 2021

    work page 2021

  5. [5]

    Climate change research and action must look beyond 2100,

    C. Lyon, E. E. Saupe, C. J. Smith, D. J. Hill, A. P. Beckerman, L. C. Stringer, R. Marchant, J. McKay, A. Burke, P. O’Higginset al., “Climate change research and action must look beyond 2100,” Global Change Biology, vol. 28, no. 2, pp. 349–361, 2022

    work page 2022

  6. [6]

    Washington, Climate change denial: Heads in the sand

    H. Washington, Climate change denial: Heads in the sand . Routledge, 2013

    work page 2013

  7. [7]

    Understanding and countering the motivated roots of climate change denial,

    G. Wong-Parodi and I. Feygina, “Understanding and countering the motivated roots of climate change denial,” Current Opinion in Envi- ronmental Sustainability, vol. 42, pp. 60–64, 2020

    work page 2020

  8. [8]

    The social media life of climate change: Platforms, publics, and future imaginaries,

    W. Pearce, S. Niederer, S. M. ¨Ozkula, and N. S ´anchez Querub´ın, “The social media life of climate change: Platforms, publics, and future imaginaries,” Wiley interdisciplinary reviews: Climate change , vol. 10, no. 2, p. e569, 2019

    work page 2019

Show all 67 references
  1. [9]

    This is too much! social media integration and adults’ psychological distress: the mediating role of cyber and place-based information overload,

    I.-A. Merlici, A. Maftei, and C. Opariuc-Dan, “This is too much! social media integration and adults’ psychological distress: the mediating role of cyber and place-based information overload,” Behaviour & Information Technology, pp. 1–11, 2024

    2024

  2. [10]

    Towards understanding cyberbullying behavior in a semi-anonymous social network,

    H. Hosseinmardi, A. Ghasemianlangroodi, R. Han, Q. Lv, and S. Mishra, “Towards understanding cyberbullying behavior in a semi-anonymous social network,” in 2014 IEEE/ACM International Conference on Ad- vances in Social Networks Analysis and Mining (ASONAM 2014) . IEEE, 2014, p...

    2014

  3. [11]

    The echo chamber effect on social media,

    M. Cinelli, G. De Francisci Morales, A. Galeazzi, W. Quattrociocchi, and M. Starnini, “The echo chamber effect on social media,” Proceedings of the national academy of sciences , vol. 118, no. 9, p. e2023301118, 2021

    2021

  4. [12]

    Misinformation, disinformation, and online propaganda,

    A. M. Guess and B. A. Lyons, “Misinformation, disinformation, and online propaganda,” Social media and democracy: The state of the field, prospects for reform, vol. 10, pp. 10–33, 2020

    2020

  5. [13]

    Correcting misinformation about climate change: The impact of partisanship in an experimental setting,

    S. D. Benegal and L. A. Scruggs, “Correcting misinformation about climate change: The impact of partisanship in an experimental setting,” Climatic change, vol. 148, no. 1, pp. 61–80, 2018

    2018

  6. [14]

    Large language models (llms): survey, technical frameworks, and future challenges,

    P. Kumar, “Large language models (llms): survey, technical frameworks, and future challenges,” Artificial Intelligence Review, vol. 57, no. 10, p. 260, 2024

    2024

  7. [15]

    Foundation and large language models: fundamentals, challenges, op- portunities, and social impacts,

    D. Myers, R. Mohawesh, V . I. Chellaboina, A. L. Sathvik, P. Venkatesh, Y .-H. Ho, H. Henshaw, M. Alhawawreh, D. Berdik, and Y . Jararweh, “Foundation and large language models: fundamentals, challenges, op- portunities, and social impacts,” Cluster Computing, vol. 27, no. 1, ...

    2024

  8. [16]

    Recurrent neural networks for language understanding

    K. Yao, G. Zweig, M.-Y . Hwang, Y . Shi, and D. Yu, “Recurrent neural networks for language understanding.” in Interspeech, 2013, pp. 2524– 2528

    2013

  9. [17]

    When large language models meet personal- ization: Perspectives of challenges and opportunities,

    J. Chen, Z. Liu, X. Huang, C. Wu, Q. Liu, G. Jiang, Y . Pu, Y . Lei, X. Chen, X. Wang et al., “When large language models meet personal- ization: Perspectives of challenges and opportunities,” World Wide Web, vol. 27, no. 4, p. 42, 2024

    2024

  10. [18]

    Moderating large scale online deliberative processes with large language models (llms): Enhancing collective decision-making

    I. D. Babatunde, O. M. Nnanna, and M. Klein, “Moderating large scale online deliberative processes with large language models (llms): Enhancing collective decision-making.” in Proceedings of the 40th ACM/SIGAPP Symposium on Applied Computing , 2025, pp. 996–1003

    2025

  11. [19]

    What Is Climate Change?

    U. Nations, “What Is Climate Change?” publisher: United Nations. [Online]. Available: https://www.un.org/en/climatechange/ what-is-climate-change

  12. [20]

    From tweets to insights: A social media analysis of the emotion discourse of sustainable energy in the united states,

    J. Corbett and B. T. R. Savarimuthu, “From tweets to insights: A social media analysis of the emotion discourse of sustainable energy in the united states,” Energy Research & Social Science , vol. 89, p. 102515, 2022

    2022

  13. [21]

    Exploring public opinions on climate change policy in

    Y . Wei, P. Gong, J. Zhang, and L. Wang, “Exploring public opinions on climate change policy in” big data era”—a case study of the european union emission trading system (eu-ets) based on twitter,” Energy Policy, vol. 158, p. 112559, 2021

    2021

  14. [22]

    Climate change in the american mind: April 2019,

    A. Leiserowitz, E. W. Maibach, S. Rosenthal, J. Kotcher, P. Bergquist, M. Ballew, M. Goldberg, and A. Gustafson, “Climate change in the american mind: April 2019,” Yale University and George Mason Univer- sity. New Haven, CT: Yale Program on Climate Change Communication, 2019

    2019

  15. [23]

    Climate change denial and ideology in swedish online media: measuring ideology change using a computational approach,

    V . Yantseva, “Climate change denial and ideology in swedish online media: measuring ideology change using a computational approach,” Journal of Computational Social Science , vol. 8, no. 1, pp. 1–25, 2025

    2025

  16. [24]

    Is conservative opposition to climate change threat-based? articulating an integrated threat model of climate change attitudes,

    S. K. Stanley, K. M. Jylh ¨a, Z. Leviston, and I. Walker, “Is conservative opposition to climate change threat-based? articulating an integrated threat model of climate change attitudes,” British Journal of Social Psychology, vol. 63, no. 2, pp. 879–893, 2024

    2024

  17. [25]

    Social network platforms and climate change in China: Evidence from TikTok,

    Y . Sun, R. Jia, A. Razzaq, and Q. Bao, “Social network platforms and climate change in China: Evidence from TikTok,” Technological Forecasting and Social Change , vol. 200, p. 123197, Mar. 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S004016252300882X

    2024

  18. [26]

    The superstar social media influencer: Exploiting linguistic style and emotional contagion over content?

    M. T. Lee and C. Theokary, “The superstar social media influencer: Exploiting linguistic style and emotional contagion over content?” Journal of Business Research , vol. 132, pp. 860–871, 2021

    2021

  19. [27]

    A tale of two social influencers: A new method for the evaluation of social marketing,

    M. T. Ballestar and J. Sainz, “A tale of two social influencers: A new method for the evaluation of social marketing,” in Digital Marketing & eCommerce Conference. Springer, 2020, pp. 80–90

    2020

  20. [28]

    From ignorance to distrust: The public “discovery

    M. Mart ´ın-Llaguno, M. T. Ballestar, M. Cuerdo-Mir, and J. Sainz, “From ignorance to distrust: The public “discovery” of covid-19 around interna- tional women’s day in spain,” International Journal of Communication , vol. 16, p. 28, 2022

    2022

  21. [29]

    13. Corpus-Assisted Analysis of Internet-Based Discourses: From Patterns to Rhetoric,

    N. Koteyko, “13. Corpus-Assisted Analysis of Internet-Based Discourses: From Patterns to Rhetoric,” in Rhetoric and the Digital Humanities , J. Ridolfo and W. Hart-Davidson, Eds. Chicago: University of Chicago Press, 2015, pp. 184–196. [Online]. Available: https://doi.org/10.7...

    2015 doi

  22. [30]

    The potential of microblogs for the study of public perceptions of climate change,

    M. R. Auer, Y . Zhang, and P. Lee, “The potential of microblogs for the study of public perceptions of climate change,” Wiley Interdisciplinary Reviews: Climate Change , vol. 5, no. 3, pp. 291–296, 2014

    2014

  23. [31]

    Polarized frames on “climate change

    S. M. Jang and P. S. Hart, “Polarized frames on “climate change” and “global warming” across countries and states: Evidence from twitter big data,” Global environmental change, vol. 32, pp. 11–17, 2015

    2015

  24. [32]

    Network analysis reveals open forums and echo chambers in social media discussions of climate change,

    H. T. Williams, J. R. McMurray, T. Kurz, and F. H. Lambert, “Network analysis reveals open forums and echo chambers in social media discussions of climate change,” Global environmental change , vol. 32, pp. 126–138, 2015

    2015

  25. [33]

    Bursting filter bubbles in a digital age: Opening minds and reducing opinion polarization through digital platforms,

    G. Einav, O. Allen, T. Gur, Y . Maaravi, and D. Ravner, “Bursting filter bubbles in a digital age: Opening minds and reducing opinion polarization through digital platforms,” Technology in Society , vol. 71, p. 102136, Nov. 2022. [Online]. Available: https://www.sciencedirect....

    2022

  26. [34]

    Challenging google search fil- ter bubbles in social and political information: Disconforming evidence from a digital methods case study,

    C. Courtois, L. Slechten, and L. Coenen, “Challenging google search fil- ter bubbles in social and political information: Disconforming evidence from a digital methods case study,” Telematics and Informatics, vol. 35, no. 7, pp. 2006–2015, 2018

    2006

  27. [35]

    Breaking the filter bubble: democracy and design,

    E. Bozdag and J. Van Den Hoven, “Breaking the filter bubble: democracy and design,” Ethics and information technology , vol. 17, pp. 249–265, 2015

    2015

  28. [36]

    The virtual stages of hate: Using goffman’s work to conceptualise the motivations for online hate,

    D. Kilvington, “The virtual stages of hate: Using goffman’s work to conceptualise the motivations for online hate,”Media, Culture & Society, vol. 43, no. 2, pp. 256–272, 2021

    2021

  29. [37]

    A conflict within a conflict: Intragroup ideological polarization and intergroup intractable conflict,

    T. O. Harel, I. Maoz, and E. Halperin, “A conflict within a conflict: Intragroup ideological polarization and intergroup intractable conflict,” Current Opinion in Behavioral Sciences , vol. 34, pp. 52–57, 2020

    2020

  30. [38]

    The origins and consequences of affective polarization in the united states,

    S. Iyengar, Y . Lelkes, M. Levendusky, N. Malhotra, and S. J. Westwood, “The origins and consequences of affective polarization in the united states,” Annual review of political science , vol. 22, no. 1, pp. 129–146, 2019

    2019

  31. [39]

    Affective polarization and dynamics of information spread in online networks,

    K. Lerman, D. Feldman, Z. He, and A. Rao, “Affective polarization and dynamics of information spread in online networks,” npj Complexity, vol. 1, no. 1, Jun. 2024. [Online]. Available: http: //dx.doi.org/10.1038/s44260-024-00008-w

    2024 doi

  32. [40]

    Troll and divide: the language of online polarization,

    A. Simchon, W. J. Brady, and J. J. Van Bavel, “Troll and divide: the language of online polarization,” PNAS Nexus , vol. 1, no. 1, p. pgac019, Mar. 2022. [Online]. Available: https://doi.org/10.1093/ pnasnexus/pgac019

    2022

  33. [41]

    Cross-partisan conversation reduced affective polarization for republicans and democrats even after the contentious 2020 election,

    E. L. Rossiter and T. N. Carlson, “Cross-partisan conversation reduced affective polarization for republicans and democrats even after the contentious 2020 election,” The Journal of Politics , vol. 86, no. 4, p. 1608–1612, Oct. 2024. [Online]. Available: http://dx.doi.org/10.1...

    2020 doi

  34. [42]

    The parties in our heads: Misperceptions about party composition and their consequences,

    D. J. Ahler and G. Sood, “The parties in our heads: Misperceptions about party composition and their consequences,” The Journal of Politics , vol. 80, no. 3, pp. 964–981, 2018

    2018

  35. [43]

    (mis) perceptions of partisan polarization in the american public,

    M. S. Levendusky and N. Malhotra, “(mis) perceptions of partisan polarization in the american public,” Public Opinion Quarterly, vol. 80, no. S1, pp. 378–391, 2016

    2016

  36. [44]

    T. F. Pettigrew and L. R. Tropp, When groups meet: The dynamics of intergroup contact. psychology press, 2013

    2013

  37. [45]

    Cross-cutting social networks: Testing democratic theory in practice,

    D. C. Mutz, “Cross-cutting social networks: Testing democratic theory in practice,” American Political Science Review, vol. 96, no. 1, pp. 111– 126, 2002

    2002

  38. [46]

    A survey on large language model (llm) security and privacy: The good, the bad, and the ugly,

    Y . Yao, J. Duan, K. Xu, Y . Cai, Z. Sun, and Y . Zhang, “A survey on large language model (llm) security and privacy: The good, the bad, and the ugly,” High-Confidence Computing , vol. 4, no. 2, p. 100211, 2024. [Online]. Available: https://www.sciencedirect.com/ science/arti...

    2024

  39. [47]

    Harnessing the power of llms in practice: A survey on chatgpt and beyond,

    J. Yang, H. Jin, R. Tang, X. Han, Q. Feng, H. Jiang, S. Zhong, B. Yin, and X. Hu, “Harnessing the power of llms in practice: A survey on chatgpt and beyond,” ACM Transactions on Knowledge Discovery from Data, vol. 18, no. 6, pp. 1–32, 2024

    2024

  40. [48]

    Leveraging ai for democratic discourse: Chat interventions can improve online political conversations at scale,

    L. P. Argyle, C. A. Bail, E. C. Busby, J. R. Gubler, T. Howe, C. Rytting, T. Sorensen, and D. Wingate, “Leveraging ai for democratic discourse: Chat interventions can improve online political conversations at scale,” Proceedings of the National Academy of Sciences, vol. 120, n...

    2023 doi

  41. [49]

    Improving factuality and reasoning in language models through multiagent debate,

    Y . Du, S. Li, A. Torralba, J. B. Tenenbaum, and I. Mordatch, “Improving factuality and reasoning in language models through multiagent debate,”

  42. [50]

    Exploring large language models for communication games: An empirical study on werewolf,

    Y . Xu, S. Wang, P. Li, F. Luo, X. Wang, W. Liu, and Y . Liu, “Exploring large language models for communication games: An empirical study on werewolf,” 2024. [Online]. Available: https://arxiv.org/abs/2309.04658

    2024 arXiv

  43. [51]

    War and peace (waragent): Large language model-based multi-agent simulation of world wars,

    W. Hua, L. Fan, L. Li, K. Mei, J. Ji, Y . Ge, L. Hemphill, and Y . Zhang, “War and peace (waragent): Large language model-based multi-agent simulation of world wars,” 2024. [Online]. Available: https://arxiv.org/abs/2311.17227

    2024 arXiv

  44. [52]

    Bases: Large-scale web search user simulation with large language model based agents,

    R. Ren, P. Qiu, Y . Qu, J. Liu, W. X. Zhao, H. Wu, J.-R. Wen, and H. Wang, “Bases: Large-scale web search user simulation with large language model based agents,” 2024. [Online]. Available: https://arxiv.org/abs/2402.17505

    2024 arXiv

  45. [53]

    Large language models (llm) in compu- tational social science: prospects, current state, and challenges,

    S. Thapa, S. Shiwakoti, S. B. Shah, S. Adhikari, H. Veeramani, M. Nasim, and U. Naseem, “Large language models (llm) in compu- tational social science: prospects, current state, and challenges,” Social Network Analysis and Mining , vol. 15, no. 1, pp. 1–30, 2025

    2025

  46. [54]

    Demystifying the working, types, benefits and limitations of chatbots,

    I. Naik, D. Naik, and N. Naik, “Demystifying the working, types, benefits and limitations of chatbots,” in UK Workshop on Computational Intelligence. Springer, 2023, pp. 558–566

    2023

  47. [55]

    Applications of ai chatbots based on gen- erative ai, large language models and large multimodal models,

    D. Naik, I. Naik, and N. Naik, “Applications of ai chatbots based on gen- erative ai, large language models and large multimodal models,” in The International Conference on Computing, Communication, Cybersecurity & AI. Springer, 2024, pp. 668–690

    2024

  48. [56]

    Conflictlens: Llm-based conflict resolu- tion training in romantic relationship,

    J. Chun, G. Zhang, and M. Xia, “Conflictlens: Llm-based conflict resolu- tion training in romantic relationship,” arXiv preprint arXiv:2505.11715, 2025

    2025 arXiv

  49. [57]

    The pushshift reddit dataset,

    J. Baumgartner, S. Zannettou, B. Keegan, M. Squire, and J. Blackburn, “The pushshift reddit dataset,” CoRR, vol. abs/2001.08435, 2020. [Online]. Available: https://arxiv.org/abs/2001.08435

    2001 arXiv

  50. [58]

    Analyzing climate change discussions on reddit,

    M. S. Parsa, H. Shi, Y . Xu, A. Yim, Y . Yin, and L. Golab, “Analyzing climate change discussions on reddit,” in 2022 International Confer- ence on Computational Science and Computational Intelligence (CSCI) , 2022, pp. 826–832

    2022

  51. [59]

    Discussion of climate change on reddit: Polarized discourse or deliberative debate?

    K. Treen, H. Williams, S. O’Neill, and T. G. C. and, “Discussion of climate change on reddit: Polarized discourse or deliberative debate?” Environmental Communication , vol. 16, no. 5, pp. 680–698, 2022. [Online]. Available: https://doi.org/10.1080/17524032.2022.2050776

    2022

  52. [60]

    Cross- platform analysis of public responses to the 2019 ridgecrest earthquake sequence on twitter and reddit,

    T. Ruan, Q. Kong, S. K. McBride, A. Sethjiwala, and Q. Lv, “Cross- platform analysis of public responses to the 2019 ridgecrest earthquake sequence on twitter and reddit,” Scientific reports, vol. 12, no. 1, p. 1634, 2022

    2019

  53. [61]

    How close is chatgpt to human experts? comparison corpus, evaluation, and detection,

    B. Guo, X. Zhang, Z. Wang, M. Jiang, J. Nie, Y . Ding, J. Yue, and Y . Wu, “How close is chatgpt to human experts? comparison corpus, evaluation, and detection,” arXiv preprint arXiv:2301.07597 , 2023

    2023 arXiv

  54. [62]

    Brain function, emotional experience and personality,

    D. L. Robinson, “Brain function, emotional experience and personality,” Netherlands Journal of Psychology , vol. 64, pp. 152–168, 2008

    2008

  55. [63]

    Positive emotions, spirituality and the practice of psychiatry,

    G. E. Vaillant, “Positive emotions, spirituality and the practice of psychiatry,” Mens sana monographs , vol. 6, no. 1, p. 48, 2008

    2008

  56. [64]

    The impact of toxic trolling comments on anti-vaccine youtube videos,

    K. Miyazaki, T. Uchiba, H. Kwak, J. An, and K. Sasahara, “The impact of toxic trolling comments on anti-vaccine youtube videos,” Scientific Reports, vol. 14, no. 1, p. 5088, 2024

    2024

  57. [65]

    Effectiveness of gamified co- operation and competition strategies in a blended learning environment for developing efl business writing skills for tvet learners,

    M. Elnaggar, K. Mohamed, and S. Gehan, “Effectiveness of gamified co- operation and competition strategies in a blended learning environment for developing efl business writing skills for tvet learners,” European Scientific Journal, vol. 30, pp. 107–128, 06 2024

    2024

  58. [66]

    Engagement with social media and social media advertising: The differentiating role of platform type,

    D. G. M. Hilde A. M. V oorveld, Guda van Noort and F. Bronner, “Engagement with social media and social media advertising: The differentiating role of platform type,” Journal of Advertising , vol. 47, no. 1, pp. 38–54, 2018. [Online]. Available: https: //doi.org/10.1080/009133...

    2018

  59. [2023]

    Available: https://arxiv.org/abs/2305.14325

    [Online]. Available: https://arxiv.org/abs/2305.14325

    arXiv

Pith tools

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