pith. sign in

arxiv: 1907.06076 · v1 · pith:4SNE7T37new · submitted 2019-07-13 · 💻 cs.SI

Changing Views: Persuasion Modeling and Argument Extraction from Online Discussions

Subhabrata Dutta , Dipankar Das , Tanmoy Chakraborty This is my paper

Pith reviewed 2026-05-24 21:56 UTC · model grok-4.3

classification 💻 cs.SI
keywords persuasion modelingargument extractionLSTMReddit discussionsonline argumentationattention mechanismsemi-supervised learningdiscussion threads
0
0 comments X

The pith

A deep LSTM model classifies whether Reddit conversations lead to successful persuasion and implicitly identifies argument facets via attention.

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

The paper builds a deep LSTM model that determines if an online conversation on Reddit ends in successful persuasion. It then applies the same model to forecast whether a given sequence of arguments will persuade the other party. The model's attention mechanism surfaces important argument aspects during training without being told to do so. A separate semi-supervised technique extracts the main argumentative components from entire discussion threads. These tools together aim to reveal patterns in how people change each other's views through written exchanges.

Core claim

We propose a deep LSTM model to classify whether a conversation leads to a successful persuasion or not, and use this model to predict whether a certain chain of arguments can lead to persuasion. While learning persuasion dynamics, our model tends to identify argument facets implicitly, using an attention mechanism. We also propose a semi-supervised approach to extract argumentative components from discussion threads. Both these models provide useful insight into how people engage in argumentation on online discussion forums.

What carries the argument

Deep LSTM classifier with attention for predicting persuasion success, paired with semi-supervised argument component extraction from threads.

If this is right

  • The model can score entire argument chains for their likelihood of changing the recipient's view.
  • Attention weights inside the model highlight which parts of an argument drive the persuasion outcome.
  • The semi-supervised extractor can pull argumentative spans from unlabeled discussion threads at scale.
  • Insights from the model show which sequences of claims tend to succeed or fail in changing minds online.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same architecture could be retrained on other forums to test whether persuasion signatures are platform-specific.
  • If the attention mechanism reliably flags key facets, it could support automated tools that surface strong counter-arguments during live discussions.
  • Combining the persuasion classifier with the extractor might allow simulation of how adding or removing specific claims alters the predicted outcome of a thread.

Load-bearing premise

Reddit threads supply reliable labels of genuine opinion change that reflect real persuasion and that the learned patterns hold outside the training data.

What would settle it

Train the LSTM on one collection of labeled Reddit threads, then test prediction accuracy on an independently labeled set of threads from a different subreddit or platform; a large drop in accuracy would falsify the generalization claim.

Figures

Figures reproduced from arXiv: 1907.06076 by Dipankar Das, Subhabrata Dutta, Tanmoy Chakraborty.

Figure 1
Figure 1. Figure 1: (Color online) An example discussion thread from Reddit CMV. Singly colored [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (Color online) Persuasion modeling architecture using hierarchical LSTM with at [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (Color online) Similarity between argument components; first sentence contains both claim and premise, identified by the presence of discourse markers “I think that” and “because”. The two candidate segments shown below in the figure are undetected by the rules and need to be matched against the above components. to (n,m). The DTW distance between P and Q is then given by p M[n][m]. DTW gives a geometric s… view at source ↗
Figure 4
Figure 4. Figure 4: Three hypothetical 1-dimensional time series [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Architecture of hierarchical LSTM without attention (left) and LSTM with selected [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: (Color online) Persuasion prediction results for hierarchical LSTM with attention [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Precision, recall and F1 score for argumentative sentence selection while using different number of top unigrams and bi-grams as keywords (used separately). As the comparison in [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Plots showing presence of argument components and attention weighting varying [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Example of a chain with successful persuasion; shaded sentences represent sentences [PITH_FULL_IMAGE:figures/full_fig_p025_9.png] view at source ↗
read the original abstract

Persuasion and argumentation are possibly among the most complex examples of the interplay between multiple human subjects. With the advent of the Internet, online forums provide wide platforms for people to share their opinions and reasonings around various diverse topics. In this work, we attempt to model persuasive interaction between users on Reddit, a popular online discussion forum. We propose a deep LSTM model to classify whether a conversation leads to a successful persuasion or not, and use this model to predict whether a certain chain of arguments can lead to persuasion. While learning persuasion dynamics, our model tends to identify argument facets implicitly, using an attention mechanism. We also propose a semi-supervised approach to extract argumentative components from discussion threads. Both these models provide useful insight into how people engage in argumentation on online discussion forums.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a deep LSTM model to classify whether Reddit discussion threads result in successful persuasion (using delta awards as labels), employs the model to predict persuasion outcomes from argument chains, observes that the model implicitly identifies argument facets via attention, and introduces a semi-supervised method for extracting argumentative components from threads.

Significance. If the central empirical claims hold after addressing label validity, the work could contribute to computational social science by demonstrating how sequence models capture persuasion dynamics in online forums and by providing tools for argument extraction; the attention-based facet discovery offers a potential strength for interpretability if supported by results.

major comments (2)
  1. [Abstract and data description] The central claims rest on binary labels derived from subreddit deltas (likely CMV) as proxies for genuine opinion change. No validation against independent measures of belief change is described; if this proxy is noisy due to social signaling, both the LSTM persuasion classifier and the downstream argument-chain prediction inherit the misalignment, undermining the attention-based facet identification and semi-supervised extraction.
  2. [Abstract] The abstract states the LSTM architecture and semi-supervised extraction goals but provides no architecture details, dataset size, evaluation metrics, results, or error analysis. Without these in the methods and experiments sections, the claim that the model 'tends to identify argument facets implicitly' cannot be assessed for load-bearing support.
minor comments (1)
  1. [Data section] Clarify the exact definition of 'successful persuasion' and the subreddit(s) used for labeling in the data section to allow reproducibility.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments. We address each major point below, indicating where revisions will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract and data description] The central claims rest on binary labels derived from subreddit deltas (likely CMV) as proxies for genuine opinion change. No validation against independent measures of belief change is described; if this proxy is noisy due to social signaling, both the LSTM persuasion classifier and the downstream argument-chain prediction inherit the misalignment, undermining the attention-based facet identification and semi-supervised extraction.

    Authors: We agree that delta awards function as a scalable but imperfect proxy for persuasion and opinion change, and that social signaling or other factors could introduce noise. This labeling approach is standard in prior computational work on the ChangeMyView subreddit. In the revision we will expand the data description and limitations sections to explicitly discuss the proxy's validity, cite supporting literature on its use, and note potential misalignment as a limitation. We cannot add new independent validation experiments at this stage. revision: yes

  2. Referee: [Abstract] The abstract states the LSTM architecture and semi-supervised extraction goals but provides no architecture details, dataset size, evaluation metrics, results, or error analysis. Without these in the methods and experiments sections, the claim that the model 'tends to identify argument facets implicitly' cannot be assessed for load-bearing support.

    Authors: The abstract is intentionally brief. Full details appear in the manuscript body: the LSTM architecture and attention mechanism are specified in Section 3, the CMV dataset size and preprocessing in Section 4, and quantitative results, metrics (accuracy/F1), error analysis, and attention-based facet evidence in Section 5. We will add cross-references from the abstract and introduction to these sections for improved readability. revision: partial

standing simulated objections not resolved
  • Independent validation of delta labels against external measures of belief change

Circularity Check

0 steps flagged

No circularity: empirical LSTM training on external Reddit labels with no derivations or self-referential reductions

full rationale

The paper describes training a deep LSTM classifier on Reddit discussion threads labeled for successful persuasion (via deltas), then using the model for downstream prediction of argument chains and semi-supervised argument extraction via attention. No equations, fitted parameters renamed as predictions, self-citations as load-bearing premises, or ansatzes are present. All components are standard supervised/semi-supervised learning on external data; the central claims reduce to model performance on held-out threads rather than any definitional or citation-chain equivalence to inputs. This is self-contained empirical work.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract describes empirical machine learning models without any free parameters, mathematical axioms, or invented entities.

pith-pipeline@v0.9.0 · 5661 in / 1116 out tokens · 33171 ms · 2026-05-24T21:56:21.831358+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · 1 internal anchor

  1. [1]

    Fogg, Mass interpersonal persuasion: An early view of a new phe- nomenon, in: International Conference on Persuasive Technology, Springer, 2008, pp

    B. Fogg, Mass interpersonal persuasion: An early view of a new phe- nomenon, in: International Conference on Persuasive Technology, Springer, 2008, pp. 23–34

    work page 2008

  2. [2]

    Lippi, P

    M. Lippi, P. Torroni, Argumentation mining: State of the art and emerging trends, ACM Transactions on Internet Technology (TOIT) 16 (2) (2016) 10:1–10:25. 26

    work page 2016

  3. [3]

    S. E. Toulmin, The Uses of Argument, Cambridge University Press, 2003

    work page 2003

  4. [4]

    A. M. Farley, K. Freeman, Burden of proof in legal argumentation, in: Proceedings of the 5th international conference on Artificial intelligence and law, ACM, 1995, pp. 156–164

    work page 1995

  5. [5]

    P. M. Dung, On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games, Artificial intelligence 77 (2) (1995) 321–357

    work page 1995

  6. [6]

    Aharoni, A

    E. Aharoni, A. Polnarov, T. Lavee, D. Hershcovich, R. Levy, R. Rinott, D. Gutfreund, N. Slonim, A benchmark dataset for automatic detection of claims and evidence in the context of controversial topics, in: Proceedings of the First Workshop on Argumentation Mining, 2014, pp. 64–68

    work page 2014

  7. [7]

    Habernal, I

    I. Habernal, I. Gurevych, Argumentation mining in user-generated web discourse, Computational Linguistics 43 (1) (2017) 125–179

    work page 2017

  8. [8]

    C. Tan, V. Niculae, C. Danescu-Niculescu-Mizil, L. Lee, Winning argu- ments: Interaction dynamics and persuasion strategies in good-faith online discussions, in: Proceedings of the 25th international conference on world wide web, ACM, 2016, pp. 613–624

    work page 2016

  9. [9]

    Biran, S

    O. Biran, S. Rosenthal, J. Andreas, K. McKeown, O. Rambow, Detecting influencers in written online conversations, in: Proceedings of the Second Workshop on Language in Social Media, Association for Computational Linguistics, 2012, pp. 37–45

    work page 2012

  10. [10]

    Crismore, R

    A. Crismore, R. Markkanen, M. S. Steffensen, Metadiscourse in persua- sive writing: A study of texts written by american and finnish university students, Written communication 10 (1) (1993) 39–71

    work page 1993

  11. [11]

    Holtgraves, B

    T. Holtgraves, B. Lasky, Linguistic power and persuasion, Journal of Lan- guage and Social Psychology 18 (2) (1999) 196–205. 27

    work page 1999

  12. [12]

    Quercia, J

    D. Quercia, J. Ellis, L. Capra, J. Crowcroft, In the mood for being influen- tial on twitter, in: Privacy, Security, Risk and Trust (PASSAT) and 2011 IEEE Third Inernational Conference on Social Computing (SocialCom), 2011 IEEE Third International Conference on, IEEE, 2011, pp. 307–314

    work page 2011

  13. [13]

    G. M. Weiksner, B. Fogg, X. Liu, Six patterns for persuasion in online social networks, in: International Conference on Persuasive Technology, Springer, 2008, pp. 151–163

    work page 2008

  14. [14]

    R. E. Guadagno, R. B. Cialdini, Online persuasion: An examination of gender differences in computer-mediated interpersonal influence., Group Dynamics: Theory, Research, and Practice 6 (1) (2002) 38

    work page 2002

  15. [15]

    Z. Wei, Y. Liu, Y. Li, Is this post persuasive? ranking argumentative comments in online forum, in: Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), Vol. 2, 2016, pp. 195–200

    work page 2016

  16. [16]

    R. M. Palau, M.-F. Moens, Argumentation mining: the detection, classi- fication and structure of arguments in text, in: Proceedings of the 12th international conference on artificial intelligence and law, ACM, 2009, pp. 98–107

    work page 2009

  17. [17]

    Moens, E

    M.-F. Moens, E. Boiy, R. M. Palau, C. Reed, Automatic detection of argu- ments in legal texts, in: Proceedings of the 11th international conference on Artificial intelligence and law, ACM, 2007, pp. 225–230

    work page 2007

  18. [18]

    C. Stab, I. Gurevych, Annotating argument components and relations in persuasive essays, in: Proceedings of COLING 2014, the 25th Interna- tional Conference on Computational Linguistics: Technical Papers, 2014, pp. 1501–1510

    work page 2014

  19. [19]

    C. Stab, I. Gurevych, Identifying argumentative discourse structures in persuasive essays, in: Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2014, pp. 46–56. 28

    work page 2014

  20. [20]

    Schneider, T

    J. Schneider, T. Groza, A. Passant, A review of argumentation for the social semantic web, Semantic Web 4 (2) (2013) 159–218

    work page 2013

  21. [21]

    Goudas, C

    T. Goudas, C. Louizos, G. Petasis, V. Karkaletsis, Argument extraction from news, blogs, and social media, in: Hellenic Conference on Artificial Intelligence, Springer, 2014, pp. 287–299

    work page 2014

  22. [22]

    Biran, O

    O. Biran, O. Rambow, Identifying justifications in written dialogs by classi- fying text as argumentative, International Journal of Semantic Computing 5 (04) (2011) 363–381

    work page 2011

  23. [23]

    Habernal, I

    I. Habernal, I. Gurevych, What makes a convincing argument? empirical analysis and detecting attributes of convincingness in web argumentation, in: Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, 2016, pp. 1214–1223

    work page 2016

  24. [24]

    Ghosh, S

    D. Ghosh, S. Muresan, N. Wacholder, M. Aakhus, M. Mitsui, Analyzing argumentative discourse units in online interactions, in: Proceedings of the First Workshop on Argumentation Mining, 2014, pp. 39–48

    work page 2014

  25. [25]

    Hochreiter, J

    S. Hochreiter, J. Schmidhuber, Long short-term memory, Neural computa- tion 9 (8) (1997) 1735–1780

    work page 1997

  26. [26]

    Lawrence, C

    J. Lawrence, C. Reed, C. Allen, S. McAlister, D. Bourget, Mining argu- ments from 19th century philosophical texts using topic based modelling, in: Proceedings of the First Workshop on Argumentation Mining, 2014, pp. 79–87

    work page 2014

  27. [27]

    Bellman, R

    R. Bellman, R. Kalaba, On adaptive control processes, IRE Transactions on Automatic Control 4 (2) (1959) 1–9

    work page 1959

  28. [28]

    Myers, L

    C. Myers, L. Rabiner, A. Rosenberg, Performance tradeoffs in dynamic time warping algorithms for isolated word recognition, IEEE Transactions on Acoustics, Speech, and Signal Processing 28 (6) (1980) 623–635. 29

    work page 1980

  29. [29]

    Mikolov, I

    T. Mikolov, I. Sutskever, K. Chen, G. S. Corrado, J. Dean, Distributed rep- resentations of words and phrases and their compositionality, in: Advances in neural information processing systems, 2013, pp. 3111–3119

    work page 2013

  30. [30]

    M. P. Marcus, M. A. Marcinkiewicz, Building a large annotated corpus of english: The penn treebank, Computational Linguistics 19 (2) 313–330

    work page

  31. [31]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv

  32. [32]

    Agarwal, F

    A. Agarwal, F. Biadsy, K. R. Mckeown, Contextual phrase-level polarity analysis using lexical affect scoring and syntactic n-grams, in: Proceed- ings of the 12th Conference of the European Chapter of the Association for Computational Linguistics, Association for Computational Linguistics, 2009, pp. 24–32

    work page 2009

  33. [33]

    Sidorov, F

    G. Sidorov, F. Velasquez, E. Stamatatos, A. Gelbukh, L. Chanona- Hern´ andez, Syntactic n-grams as machine learning features for natural lan- guage processing, Expert Systems with Applications 41 (3) (2014) 853–860

    work page 2014

  34. [34]

    Sidorov, Should syntactic n-grams contain names of syntactic relations?., Int

    G. Sidorov, Should syntactic n-grams contain names of syntactic relations?., Int. J. Comput. Linguistics Appl. 5 (2) (2014) 25–47

    work page 2014

  35. [35]

    Baccianella, A

    S. Baccianella, A. Esuli, F. Sebastiani, Sentiwordnet 3.0: an enhanced lexical resource for sentiment analysis and opinion mining., in: LREC, Vol. 10, 2010, pp. 2200–2204

    work page 2010

  36. [36]

    Gunning, The fog index after twenty years, Journal of Business Com- munication 6 (2) (1969) 3–13

    R. Gunning, The fog index after twenty years, Journal of Business Com- munication 6 (2) (1969) 3–13

    work page 1969

  37. [37]

    Kusner, Y

    M. Kusner, Y. Sun, N. Kolkin, K. Weinberger, From word embeddings to document distances, in: International Conference on Machine Learning, 2015, pp. 957–966. 30

    work page 2015