Detecting Winning Arguments with Large Language Models and Persuasion Strategies

ACL ARR 2025 July Submission1209 Authors

Abstract:

Detecting persuasion in text is a challenging task, with important implications for understanding human communication. In this work, we address the problem using the Winning Arguments dataset built from the Change My View subreddit, where users award a "delta" to comments that successfully change their opinion.
Given a pair of similar messages where only one of which received a delta, our goal is to identify the successful one.
We approach the task by leveraging large language models (LLMs) through a chain-of-thought framework that guides them to reason about six persuasion strategies that have been widely studied in the literature. Our method directs LLMs to reflect on the use of each strategy within a message and to assess its overall persuasiveness.
To better understand the influence of content, we also organize the dataset into broad discussion topics and examine performance across them. Finally, we release this topic-annotated version of the dataset to support future research on persuasion detection.
Our results show that LLMs, when guided through explicit reasoning steps, can effectively capture persuasive signals. This highlights the value of strategy-based prompting for improving interpretability and robustness in argument quality assessment.

Paper Type: Long

Research Area: Computational Social Science and Cultural Analytics

Research Area Keywords: quantitative analyses of news and/or social media

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Data resources

Languages Studied: English

Reassignment Request Area Chair: This is not a resubmission

Reassignment Request Reviewers: This is not a resubmission

A1 Limitations Section: This paper has a limitations section.

A2 Potential Risks: N/A

B Use Or Create Scientific Artifacts: Yes

B1 Cite Creators Of Artifacts: Yes

B1 Elaboration: 3

B2 Discuss The License For Artifacts: Yes

B2 Elaboration: Ethical Considerations

B3 Artifact Use Consistent With Intended Use: Yes

B3 Elaboration: 3

B4 Data Contains Personally Identifying Info Or Offensive Content: Yes

B4 Elaboration: Ethical Considerations

B5 Documentation Of Artifacts: Yes

B5 Elaboration: 3

B6 Statistics For Data: Yes

B6 Elaboration: 3, 6

C Computational Experiments: Yes

C1 Model Size And Budget: Yes

C1 Elaboration: Ethical Considerations

C2 Experimental Setup And Hyperparameters: Yes

C2 Elaboration: 6, Appendix I

C3 Descriptive Statistics: Yes

C3 Elaboration: 6, Appendixes A, J

C4 Parameters For Packages: Yes

C4 Elaboration: 6, Appendix H

D Human Subjects Including Annotators: No

D1 Instructions Given To Participants: N/A

D2 Recruitment And Payment: N/A

D3 Data Consent: N/A

D4 Ethics Review Board Approval: N/A

D5 Characteristics Of Annotators: N/A

E Ai Assistants In Research Or Writing: No

E1 Information About Use Of Ai Assistants: N/A

Author Submission Checklist: yes

Submission Number: 1209