Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Communication is All You Need: Persuasion Dataset Construction via Multi-LLM Communication

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 The paper claims that a multi-LLM communication framework can automatically generate persuasive dialogues that human annotators struggle to distinguish from human-written text, while labeling each turn with a continuous perspective-change…

desk verdict A genuinely useful dialogue-generation pipeline whose headline capability holds up, but the continuous persuasion labels—the paper's main novelty—are unvalidated and prompt-sensitive. read the letter →

arxiv 2502.08896 v1 pith:GNSU3A3A submitted 2025-02-13 cs.CL cs.AI

Weicheng Ma , Hefan Zhang , Ivory Yang , Shiyu Ji , Joice Chen , Farnoosh Hashemi , Shubham Mohole , Ethan Gearey
show 3 more authors
Michael Macy Saeed Hassanpour Soroush Vosoughi
This is my paper · ORCID
classification cs.CLcs.AI
keywords persuasiondatasetconstructionmulti-agentLLMframeworkpersuasivedialoguegenerationcontinuousperspective-changelabelshumanevaluationofgeneratedtextsocialnormsandtaboosLLM-generateddetectionmulti-party
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that persuasion data can be produced at scale by several large language models talking to one another, with almost no human supervision, and that the resulting dialogues are natural, diverse, and persuasive enough to support research. It proposes a six-role pipeline: a persuader and a persuadee speak, while dedicated agents check utterance quality, remove polite filler, score how much each speaker's position shifts, regulate the debate, and polish the final transcript. The headline evidence is a forced-choice test on 400 utterance pairs in which two human annotators together identified the model-generated sentence only 29.25% of the time, just above the 25% chance baseline. If the claim holds, social and computational scientists gain a cheap and adjustable source of persuasion data, including scenarios that involve social taboos or multiple speakers.

What carries the argument

The central mechanism is a role-specialized multi-agent communication loop: each language-model agent handles one narrow job and its output is fed to the next agent, so no single model must hold the entire dialogue task in one prompt. The load-bearing component for the dataset's scientific value is the persuasiveness annotation agent, which assigns every utterance a cumulative score in $[0,1]$ measuring how far a speaker's current stance has shifted from their original position.

What would settle it

Have independent human annotators assign perspective-change scores to a sample of the generated dialogues and compute agreement with the annotation agent's labels; if agreement is low, or if the agent gives a non-zero score to a round in which no speaker changed position, the continuous-label contribution fails even though the dialogues may still read fluently.

Watch

Extended reading notes

Core claim

The authors claim that assigning distinct roles to multiple communicating language models solves the two problems that plague single-model persuasion generation: refusals in sensitive scenarios and shallow exchanges. The framework produces multi-round dialogues between a persuader and a persuadee, while dedicated agents monitor utterance quality, strip polite filler, assign a cumulative 0-to-1 perspective-change score to every utterance, regulate logical flow and strategy novelty, and post-process the final text. The quantitative centerpiece is a forced-choice test on 400 utterance pairs: two human annotators jointly identified the machine-written sentence only 29.25% of the time, close to the 25% random baseline, with individual accuracies of 0.546, 0.558, and 0.508. The authors also report high dialogue-level ratings for coherence, role consistency, and topic consistency, and show that persuasion-strategy distributions shift with topic.

Load-bearing premise

The framework's value depends on the language-model annotation agent producing trustworthy 0-to-1 perspective-change scores for every utterance, and that trust rests on a handful of scoring examples and spot-checks rather than measured human agreement.

Editorial extensions

If this is right

  • Researchers could generate persuasive-dialogue corpora on demand for topics that would be costly or ethically sensitive to collect from humans, including social taboos.
  • The framework accepts a requested persuasion strategy as an optional input, so studies can isolate how logical versus emotional appeals behave.
  • It extends to multi-party settings, such as two persuaders working on one persuadee, with only minor changes to agent initialization.
  • The continuous perspective-change labels allow analysis of gradual persuasion rather than binary stance flips.

Reading between the lines

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

  • Inference: The indistinguishability result is reported for one model combination and one rewriting protocol; it is an open question whether other model families or annotators trained to spot machine tells would see the same near-chance accuracy.
  • Inference: The continuous persuasion labels are validated by spot-checking rather than measured human agreement, so a natural next step is an inter-rater study that quantifies label reliability before the labels are used in downstream persuasion research.
  • Inference: If the framework is as flexible as shown, it could generate on-demand persuasion corpora for arbitrary norms, a capability that would accelerate persuasion research while sharpening the dual-use concerns the authors acknowledge regarding misinformation and propaganda.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes a multi-LLM communication framework for automatically generating persuasive dialogue data. Six agent groups (persuader, persuadee, utterance quality monitor, language refinement, persuasiveness annotation, global regulation, and postprocessing) cooperate to produce dialogues on arbitrary topics, including ethically sensitive scenarios sampled from NormBank. The authors evaluate 200 generated dialogues via human differentiation of model-generated versus human-rewritten utterances (Section 3.1), dialogue-level quality ratings (Section 3.2), and strategy diversity annotations (Section 3.3). They also demonstrate strategy-controlled generation and multi-party extensions (Section 4). The central capability claim is that the framework produces fluent, coherent, and strategically diverse persuasive dialogues that humans cannot reliably distinguish from human writing, with minimal human oversight, along with continuous per-utterance perspective-change scores.

Significance. If the claimed results hold, this is a useful contribution: a scalable pipeline for producing persuasive-dialogue data with a continuous annotation of perspective shift, with potential value for NLP research on persuasion and for social-science studies of influence. The paper includes several concrete strengths: a detailed multi-agent architecture, human evaluation of dialogue quality on established dimensions, a near-chance human detection result that supports the naturalness of the generated utterances, and an explicit strategy-diversity analysis with cross-topic comparisons. The framework's flexibility is demonstrated through strategy-controlled and multi-party settings. However, the load-bearing continuous perspective-shift labels—the most distinctive asset of the dataset—rest on a single unvalidated LLM annotation agent, and at least one central quantitative claim contains an arithmetic inconsistency. These issues need to be resolved before the paper's central claims are fully supported.

major comments (4)
  1. [Section 2.4, Appendix G, Figure I1] The continuous perspective-shift labels are a core contribution, but the paper does not quantitatively validate the annotation agent that produces them. Section 1 states that the labels 'align closely with human judgments,' yet Appendix G reports only that the authors 'manually checked' a few dialogues; no inter-rater agreement or correlation with human scores is provided. Moreover, the ablation in Figure I1 shows that removing the two scoring examples makes the agent assign a score of 1.0 to a round in which neither speaker changes position, demonstrating high prompt sensitivity. The scoring example in Figure B4 is also under-calibrated: a persuadee moving from 'should not pick flowers' to 'it is OK to pick flowers' is assigned 0.5, not 1.0, leaving the meaning of the upper endpoint unclear. Please report measured agreement between the annotation agent and human raters on a stratified sample (e.g., Cohen's kappa or ICC) and clarify the scale's anchoring.
  2. [Section 3.1.1] The sentence 'In 49% cases (98 pairs)' is arithmetically inconsistent: 49% of 400 is 196 pairs, not 98. The counts 117 (both correct), 98 (disagreement), and 87 (both incorrect, 21.75% of 400) sum to 302, not 400. This error appears in the main quantitative result and must be corrected. In addition, the headline result that both annotators identified the model-generated utterance in only 29.25% of pairs is reported without a significance test or confidence interval; a one-sample binomial test against the 25% baseline would establish whether the result differs from chance, and the paper should report it.
  3. [Section 3.2.1, Table 2] The dialogue quality ratings are based on only two annotators for 50 dialogues, and the unweighted Cohen's kappa values for some items are low (e.g., 0.339 for 'The utterances build on prior information in near context' and 0.397 for 'There are no sudden shifts in a speaker's objectives'). These low agreement values bound the reliability of the high average scores (e.g., 2.969 for coherence and 3.0 for clarity). The paper should report per-item confidence intervals, discuss how the low agreement affects the interpretation, and consider whether the two annotators' scores are sufficiently reliable for the claims made.
  4. [Section 3.3] The strategy diversity evaluation relies on human annotators identifying all persuasion strategies used in 25 dialogues, but no inter-annotator agreement is reported for this task. Without a measure such as Cohen's kappa, the claims of 'significant variations in strategy usage across different contexts' (Section 3.3) and the heatmap analysis in Figure 3 are difficult to interpret. Please report agreement statistics for the strategy annotations.
minor comments (6)
  1. [Section 3, first paragraph] There is a typo: 'the the topic' should be 'the topic'.
  2. [Table 1] The example utterances contain several typos and informal errors: 'comprimized' should be 'compromised', 'staboe' should be 'stable', and 'its a opportunity' should be 'it's an opportunity'. Since the table is intended to illustrate naturalness, these errors are distracting and should be corrected.
  3. [Section 2.3] The phrase '2 examples are also provided' should be 'Two examples are also provided' for formal style.
  4. [Figure 2] The text references 'Figure2' without a space; please fix the citation to 'Figure 2'.
  5. [Section 4.2] The multi-party generation is demonstrated with a single example and the claim that it is 'validated as high in quality' is not supported by quantitative evaluation. Please either add a brief evaluation or temper the claim.
  6. [Data availability] The paper states that the full code and dialogues 'will be made publicly available' but provides no repository link; consider including a data/code availability statement with the actual link or an anonymized version.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's headline quality claims rest on external human evaluation and literature-based rubrics; the LLM self-annotation layers are validation gaps, not circular reductions.

full rationale

The paper's main quality claims are supported by human evaluation that is not defined in terms of the model outputs being tested. Naturalness is assessed through a forced-choice differentiation task in which human annotators compare model-generated utterances against human rewrites (Section 3.1.1); dialogue-level coherence and fluency are rated by trained annotators using rubric dimensions drawn from prior literature (Section 3.2.1, Table 2); and strategy diversity is judged by human annotators coding utterances against an externally sourced taxonomy (Section 3.3, Table 3). None of these evaluations is a fitted parameter renamed as a prediction, and none is justified by a self-citation chain. The continuous perspective-change labels are produced by a GPT-4 annotation agent and only manually spot-checked in Appendix G, and the error analysis in Section 3.1.2 is assisted by o1-preview; these are same-family-LLM self-referential validation weaknesses, not circular reductions, because the labels are not fit to the claim being made and the claim is not equivalent to the annotation prompt. The two scoring examples in the annotation prompt (Figure B4) are hand-engineered inputs rather than fitted outputs, and the ablation in Figure I1 is presented as a prompt-sensitivity limitation rather than as evidence for label validity. The self-citations in the introduction (Wang et al. 2024; Yang et al. 2024) are contextual related-work references and are not load-bearing. Therefore, while the persuasiveness-label validation is weaker than the paper claims, the derivation chain is not circular.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

Most quantities in this paper are design choices rather than fitted parameters, so the ledger is short: the 0-1 perspective-change scale, the two scoring examples, and the taboo-weighted NormBank sample are hand-set and shape the reported numbers. The load-bearing assumptions are that human rewrites are a valid gold standard for naturalness, that near-chance detection implies quality, that the persuasion taxonomy transfers to LLM text, that NormBank covers the relevant norms, and that the annotation agent's scores are valid outside the two examples and one spot-checked sample. If any of these fails, the corresponding claim weakens.

free parameters (3)
  • Persuasiveness score scale = 0 to 1 continuous
    Hand-chosen continuous scale for perspective change, generated by GPT-4 with two few-shot examples (Section 2.4, Figure B4) and validated only by manual spot checks (Appendix G).
  • Scoring examples provided to annotation agent = 2
    Appendix I (Figure I1) shows the annotation agent without examples miscodes a no-change round as full persuasion, so the number of examples is an ad hoc choice that the dataset's labels depend on.
  • NormBank sample composition = 98 taboos, 76 normal, 26 expected
    Deliberately weighted toward taboos (Section 3), not a random or fitted sample; this composition shapes all reported quality statistics and the claims about taboo scenarios.
assumptions (5)
  • domain assumption Human rewrites of model utterances are a valid gold standard for naturalness in the differentiation task.
    Section 3.1.1 and Figure D1: the headline near-chance detection result only supports 'natural' if a human rewrite is the right reference text; rewriters were instructed to rephrase even satisfactory sentences, so the reference may be a lowered bar.
  • domain assumption Near-chance forced-choice detection by a small annotator pool implies high utterance quality.
    Section 3.1.1: the paper interprets 29.25% both-correct (chance 25%) as 'challenge of distinguishing' without a significance test or CI, and only 3 annotators participated.
  • domain assumption The persuasion strategy taxonomy from Anand et al. (2011) transfers to LLM-generated dialogue and is identifiable by human annotators.
    Section 3.3: strategy frequency claims rest on human annotation of 25 dialogues against a 9-strategy taxonomy; the taxonomy's coverage and transfer to model-generated text are assumed.
  • domain assumption NormBank norms are a representative source of social norms and taboos for persuasion scenarios.
    Sections 2 and 3: taboo scenarios are sampled from NormBank; the framework's claimed coverage of ethically challenging persuasion is bounded by NormBank's contents.
  • domain assumption The GPT-4 annotation agent assigns valid 0-1 perspective-change scores when given two scoring examples.
    Section 2.4, Figure B4, Appendix I: Figure I1 shows the same agent, without examples, assigns 1 (full persuasion) to a no-change round; Appendix G validates scores only by manual spot checks, so dataset-wide label validity is weakly tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Communication is All You Need: Persuasion Dataset Construction via Multi-LLM Communication." pith.science (2026). https://pith.science/paper/GNSU3A3A

@misc{pith2026250208896,
  author       = {Pith},
  title        = {Pith review of: Communication is All You Need: Persuasion Dataset Construction via Multi-LLM Communication},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GNSU3A3A}},
  note         = {Machine review of arXiv:2502.08896}
}
read the original abstract

Large Language Models (LLMs) have shown proficiency in generating persuasive dialogue, yet concerns about the fluency and sophistication of their outputs persist. This paper presents a multi-LLM communication framework designed to enhance the generation of persuasive data automatically. This framework facilitates the efficient production of high-quality, diverse linguistic content with minimal human oversight. Through extensive evaluations, we demonstrate that the generated data excels in naturalness, linguistic diversity, and the strategic use of persuasion, even in complex scenarios involving social taboos. The framework also proves adept at generalizing across novel contexts. Our results highlight the framework's potential to significantly advance research in both computational and social science domains concerning persuasive communication.

Figures

Figures reproduced from arXiv: 2502.08896 by the authors.

Figure 1
Figure 1. Overview of our data generation and anno [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Frequency Distribution of Persuasion Strate [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Heatmap displaying the cosine similarity be [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. CLAIM: An Intent-Driven Multi-Agent Framework for Analyzing Manipulation in Courtroom Dialogues

    cs.CL 2025-06 conditional novelty 6.0 of 10

    LegalCon and CLAIM introduce a courtroom dialogue dataset and a two-stage multi-agent framework that reports improved manipulation detection and manipulator identification over zero-shot and few-shot LLM baselines.

Reference graph

Works this paper leans on

49 extracted references · 32 canonical work pages · cited by 1 Pith paper

  1. [1]

    Pranav Anand, Joseph King, Jordan Boyd-Graber, Earl Wagner, Craig Martell, Doug Oard, and Philip Resnik. 2011. Believe me: we can do this! annotating persuasive acts in blog text. In Proceedings of the 10th AAAI Conference on Computational Models of Natural Argument, AAAIWS'11-10, page 11–15. AAAI Press

    work page 2011

  2. [2]

    Lisa P Argyle, Christopher A Bail, Ethan C Busby, Joshua R Gubler, Thomas Howe, Christopher Rytting, Taylor Sorensen, and David Wingate. 2023. Leveraging ai for democratic discourse: Chat interventions can improve online political conversations at scale. Proceedings of the National Academy of Sciences, 120(41):e2311627120

    work page 2023

  3. [3]

    Chongyang Bai, Haipeng Chen, Srijan Kumar, Jure Leskovec, and VS Subrahmanian. 2021. M2p2: Multimodal persuasion prediction using adaptive fusion. IEEE Transactions on Multimedia, 25:942--952

    work page 2021

  4. [4]

    Hui Bai, Jan Voelkel, Johannes Eichstaedt, and Robb Willer. 2023. Artificial intelligence can persuade humans on political issues

    work page 2023

  5. [5]

    Annye Braca and Pierpaolo Dondio. 2023. Developing persuasive systems for marketing: the interplay of persuasion techniques, customer traits and persuasive message design. Italian Journal of Marketing, 2023(3):369--412

    work page 2023

  6. [6]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert - Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litw...

    2020

  7. [7]

    Canyu Chen and Kai Shu. 2023. Combating misinformation in the age of llms: Opportunities and challenges. arXiv preprint arXiv:2311.05656

    arXiv 2023

  8. [8]

    Jiaao Chen and Diyi Yang. 2021. https://doi.org/10.1609/aaai.v35i14.17498 Weakly-supervised hierarchical models for predicting persuasive strategies in good-faith textual requests . Proceedings of the AAAI Conference on Artificial Intelligence, 35(14):12648--12656

    work page doi:10.1609/aaai.v35i14.17498 2021

Show all 49 references
  1. [9]

    Laura Espinosa and Marcel Salath \'e . 2024. Use of large language models as a scalable approach to understanding public health discourse. medRxiv, pages 2024--02

    2024

  2. [10]

    Brian J Fogg. 2009. A behavior model for persuasive design. In Proceedings of the 4th international Conference on Persuasive Technology, pages 1--7

    2009

  3. [11]

    Laura French. 2024. https://www.scmagazine.com/news/openai-report-reveals-threat-actors-using-chatgpt-in-influence-operations Openai report reveals threat actors using chatgpt in influence operations . Accessed: 2024-06-12

    2024

  4. [12]

    Sebastian Gehrmann, Hendrik Strobelt, and Alexander Rush. 2019. https://doi.org/10.18653/v1/P19-3019 GLTR : Statistical detection and visualization of generated text . In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: System Demonstrat...

    2019 doi

  5. [13]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, et al. 2023. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. arXiv preprint arXiv:2311.05232

    2023 arXiv

  6. [14]

    Daphne Ippolito, Daniel Duckworth, Chris Callison-Burch, and Douglas Eck. 2020. https://doi.org/10.18653/v1/2020.acl-main.164 Automatic detection of generated text is easiest when humans are fooled . In Proceedings of the 58th Annual Meeting of the Association for Computationa...

    2020 doi

  7. [15]

    Rahul Radhakrishnan Iyer and Katia Sycara. 2019. https://arxiv.org/abs/1912.06745 An unsupervised domain-independent framework for automated detection of persuasion tactics in text . Preprint, arXiv:1912.06745

    2019 arXiv

  8. [16]

    Tianbo Ji, Yvette Graham, Gareth Jones, Chenyang Lyu, and Qun Liu. 2022. https://doi.org/10.18653/v1/2022.acl-long.445 Achieving reliable human assessment of open-domain dialogue systems . In Proceedings of the 60th Annual Meeting of the Association for Computational Linguisti...

    2022 doi

  9. [17]

    Chuhao Jin, Kening Ren, Lingzhen Kong, Xiting Wang, Ruihua Song, and Huan Chen. 2024. https://doi.org/10.18653/v1/2024.acl-long.92 Persuading across diverse domains: a dataset and persuasion large language model . In Proceedings of the 62nd Annual Meeting of the Association fo...

    2024 doi

  10. [18]

    Daniel Gordon Jones. 2024. Detecting propaganda in news articles using large language models. Eng OA, 2(1):01--12

    2024

  11. [19]

    Pei Ke, Jian Guan, Minlie Huang, and Xiaoyan Zhu. 2018. https://doi.org/10.18653/v1/P18-1139 Generating informative responses with controlled sentence function . In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers),...

    2018 doi

  12. [20]

    Deepak Kumar, Yousef Anees AbuHashem, and Zakir Durumeric. 2024. Watch your language: Investigating content moderation with large language models. In Proceedings of the International AAAI Conference on Web and Social Media, volume 18, pages 865--878

    2024

  13. [21]

    Yaman Kumar, Rajat Jha, Arunim Gupta, Milan Aggarwal, Aditya Garg, Tushar Malyan, Ayush Bhardwaj, Rajiv Ratn Shah, Balaji Krishnamurthy, and Changyou Chen. 2023. Persuasion strategies in advertisements. In Proceedings of the AAAI Conference on Artificial Intelligence

    2023

  14. [22]

    Bolin Lai, Hongxin Zhang, Miao Liu, Aryan Pariani, Fiona Ryan, Wenqi Jia, Shirley Anugrah Hayati, James M Rehg, and Diyi Yang. 2022. Werewolf among us: A multimodal dataset for modeling persuasion behaviors in social deduction games. arXiv preprint arXiv:2212.08279

    2022 arXiv

  15. [23]

    Jingyuan Li and Xiao Sun. 2018. https://doi.org/10.18653/v1/D18-1071 A syntactically constrained bidirectional-asynchronous approach for emotional conversation generation . In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 678--68...

    2018 doi

  16. [24]

    Hongru Liang and Huaqing Li. 2021. https://arxiv.org/abs/2105.11197 Towards standard criteria for human evaluation of chatbots: A survey . Preprint, arXiv:2105.11197

    2021 arXiv

  17. [25]

    Sue Lim and Ralf Schm \"a lzle. 2023. Artificial intelligence for health message generation: an empirical study using a large language model (llm) and prompt engineering. Frontiers in Communication, 8:1129082

    2023

  18. [26]

    Zhaojiang Lin, Andrea Madotto, Jamin Shin, Peng Xu, and Pascale Fung. 2019. https://doi.org/10.18653/v1/D19-1012 M o EL : Mixture of empathetic listeners . In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint...

    2019 doi

  19. [27]

    Stephanie M Lukin, Pranav Anand, Marilyn Walker, and Steve Whittaker. 2017. Argument strength is in the eye of the beholder: Audience effects in persuasion. arXiv preprint arXiv:1708.09085

    2017 arXiv

  20. [28]

    Yongqiang Ma, Jiawei Liu, Fan Yi, Qikai Cheng, Yong Huang, Wei Lu, and Xiaozhong Liu. 2023. https://api.semanticscholar.org/CorpusID:256826708 Ai vs. human -- differentiation analysis of scientific content generation

    2023

  21. [29]

    SC Matz, JD Teeny, Sumer S Vaid, H Peters, GM Harari, and M Cerf. 2024. The potential of generative ai for personalized persuasion at scale. Scientific Reports, 14(1):4692

    2024

  22. [30]

    Elyas Meguellati, Lei Han, Abraham Bernstein, Shazia Sadiq, and Gianluca Demartini. 2024. How good are llms in generating personalized advertisements? In Companion Proceedings of the ACM on Web Conference 2024, pages 826--829

    2024

  23. [31]

    Raphael Meier. 2024. Llm-aided social media influence operations. Large, page 105

    2024

  24. [32]

    Nikita Moghe, Siddhartha Arora, Suman Banerjee, and Mitesh M. Khapra. 2018. https://doi.org/10.18653/v1/D18-1255 Towards exploiting background knowledge for building conversation systems . In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processin...

    2018 doi

  25. [33]

    OpenAI. 2024. https://openai.com/index/learning-to-reason-with-llms Learning to reason with llms

    2024

  26. [34]

    Joon Sung Park, Joseph O'Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, pages 1--22

    2023

  27. [35]

    Amalie Pauli, Leon Derczynski, and Ira Assent. 2022. Modelling persuasion through misuse of rhetorical appeals. In Proceedings of the Second Workshop on NLP for Positive Impact (NLP4PI), pages 89--100

    2022

  28. [36]

    Jakub Piskorski, Nicolas Stefanovitch, Nikolaos Nikolaidis, Giovanni Da San Martino, and Preslav Nakov. 2023. Multilingual multifaceted understanding of online news in terms of genre, framing, and persuasion techniques. In Proceedings of the 61st Annual Meeting of the Associat...

    2023

  29. [37]

    Robin Schaefer, Ren \'e Knaebel, and Manfred Stede. 2023. Towards fine-grained argumentation strategy analysis in persuasive essays. In Proceedings of the 10th Workshop on Argument Mining, pages 76--88

    2023

  30. [38]

    LJ Shrum, Min Liu, Mark Nespoli, and Tina M Lowrey. 2012. Persuasion in the Marketplace. Sage

    2012

  31. [39]

    Assaf Toledo, Shai Gretz, Edo Cohen-Karlik, Roni Friedman, Elad Venezian, Dan Lahav, Michal Jacovi, Ranit Aharonov, and Noam Slonim. 2019. Automatic argument quality assessment--new datasets and methods. arXiv preprint arXiv:1909.01007

    2019 arXiv

  32. [40]

    Xuewei Wang, Weiyan Shi, Richard Kim, Yoojung Oh, Sijia Yang, Jingwen Zhang, and Zhou Yu. 2019. Persuasion for good: Towards a personalized persuasive dialogue system for social good. arXiv preprint arXiv:1906.06725

    2019 arXiv

  33. [41]

    Yuxin Wang, Ivory Yang, Saeed Hassanpour, and Soroush Vosoughi. 2024. M ental M anip: A dataset for fine-grained analysis of mental manipulation in conversations. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)...

    2024

  34. [42]

    Wenquan Wu, Zhen Guo, Xiangyang Zhou, Hua Wu, Xiyuan Zhang, Rongzhong Lian, and Haifeng Wang. 2019. https://doi.org/10.18653/v1/P19-1369 Proactive human-machine conversation with explicit conversation goal . In Proceedings of the 57th Annual Meeting of the Association for Comp...

    2019 doi

  35. [43]

    Jin Xu, Xiaojiang Liu, Jianhao Yan, Deng Cai, Huayang Li, and Jian Li. 2022. Learning to break the loop: Analyzing and mitigating repetitions for neural text generation. Advances in Neural Information Processing Systems, 35:3082--3095

    2022

  36. [44]

    Ivory Yang, Xiaobo Guo, Sean Xie, and Soroush Vosoughi. 2024. Enhanced detection of conversational mental manipulation through advanced prompting techniques. arXiv preprint arXiv:2408.07676

    2024 arXiv

  37. [45]

    Tom Young, Erik Cambria, Iti Chaturvedi, Hao Zhou, Subham Biswas, and Minlie Huang. 2018. https://doi.org/10.1609/aaai.v32i1.11923 Augmenting end-to-end dialogue systems with commonsense knowledge . In Proceedings of the AAAI Conference on Artificial Intelligence, volume 32

    2018 doi

  38. [46]

    Qingfu Zhu, Lei Cui, Wei-Nan Zhang, Furu Wei, and Ting Liu. 2019. https://doi.org/10.18653/v1/P19-1366 Retrieval-enhanced adversarial training for neural response generation . In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 376...

    2019 doi

  39. [47]

    Caleb Ziems, Jane Dwivedi-Yu, Yi-Chia Wang, Alon Halevy, and Diyi Yang. 2023. Normbank: A knowledge bank of situational social norms. arXiv preprint arXiv:2305.17008

    2023 arXiv

  40. [48]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  41. [49]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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