pith. sign in

arxiv: 2410.13036 · v4 · submitted 2024-10-16 · 💻 cs.HC · cs.SI

Uncovering the Internet's Hidden Values: An Empirical Study of Desirable Behavior Using Highly-Upvoted Content on Reddit

Agam Goyal , Charlotte Lambert , Yoshee Jain , Eshwar Chandrasekharan This is my paper

Pith reviewed 2026-05-23 18:49 UTC · model grok-4.3

classification 💻 cs.HC cs.SI
keywords Redditupvotescommunity normsdesirable behaviorprosocialitylarge language modelsonline communitiesvalue extraction
0
0 comments X

The pith

Upvotes on Reddit reveal a much wider range of community values than existing prosociality models capture.

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

The paper treats highly-upvoted comments as a direct signal of what online communities consider desirable behavior. It processes 16,000 such comments from 80 subreddits in 2016 and 2022, using a large language model to surface dozens of macro-, meso-, and micro-level values. The extracted values include most items from prior qualitative taxonomies yet also many others that actually appear in practice. Standard computational measures of prosociality are shown to miss 82 percent of these values on average. The work therefore argues that platforms and moderators need new, more nuanced ways to detect and promote the full spectrum of behaviors communities reward.

Core claim

By treating upvotes as a proxy for desirability and applying large-language-model extraction to 16,000 highly-upvoted comments across 80 subreddits, the authors compile 64 values in 2016 and 72 values in 2022; existing prosociality models cover only 18 percent of these values on average while the new extraction recovers nearly all previously identified values plus additional ones that communities demonstrably encourage.

What carries the argument

Upvotes as proxy for desirability, combined with large-language-model extraction of macro-, meso-, and micro-level values from comments.

If this is right

  • Moderator tools can surface a broader set of examples than prosociality scores alone provide.
  • Automated systems for highlighting desirable content must incorporate community-specific value lists rather than generic prosocial metrics.
  • Platform design that relies on existing prosocial detectors will systematically under-promote many behaviors communities actually reward.
  • Longitudinal comparisons between 2016 and 2022 values can track how community priorities shift over time.

Where Pith is reading between the lines

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

  • The method could be applied to other platforms that record explicit approval signals to test whether the same gap between prosocial measures and observed values appears elsewhere.
  • If the extracted value lists prove stable within subreddits, they could serve as lightweight community charters for new moderators.
  • Future models might be trained directly on upvote signals rather than on curated prosocial datasets to close the observed coverage gap.

Load-bearing premise

Upvotes serve as a reliable proxy for the full spectrum of desirable behavior that communities wish to encourage.

What would settle it

A direct comparison between the values extracted from upvoted comments and the behaviors that subreddit moderators and active users explicitly list as desirable in surveys or rules would show large mismatch.

Figures

Figures reproduced from arXiv: 2410.13036 by Agam Goyal, Charlotte Lambert, Eshwar Chandrasekharan, Yoshee Jain.

Figure 1
Figure 1. Figure 1: Plot representing the macro and meso values extracted from 80 subreddits in [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Plot depicting odds ratios by logistic regression [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Plot depicting quantiles of the score and two thresholds marked in red at 0.9 and 0.95. The first signifi￾cant rise in the score occurs at 0.95 which we therefore use as the threshold for “high” upvote comments. Since there is little to no rise until 0.7, we use that as our threshold for “low” upvote comments. politics, entertaining, education, identity, explicit, contro￾versial, thoughtfulness, quality, c… view at source ↗
read the original abstract

A major task for moderators of online spaces is norm-setting, essentially creating shared norms for user behavior in their communities. Platform design principles emphasize the importance of highlighting norm-adhering examples and explicitly stating community norms. However, norms and values vary between communities and go beyond content-level attributes, making it challenging for platforms and researchers to provide automated ways to identify desirable behavior to be highlighted. Current automated approaches to detect desirability are limited to measures of prosocial behavior, but we do not know whether these measures fully capture the spectrum of what communities value. In this paper, we use upvotes, which express community approval, as a proxy for desirability and examine 16,000 highly-upvoted comments across 80 popular sub-communities on Reddit. Using a large language model, we extract values from these comments across two years (2016 and 2022) and compile 64 and 72 $\textit{macro}$, $\textit{meso}$, and $\textit{micro}$ values for 2016 and 2022 respectively, based on their frequency across communities. Furthermore, we find that existing computational models for measuring prosociality were inadequate to capture on average $82\%$ of the values we extracted. Finally, we show that our approach can not only extract most of the qualitatively-identified values from prior taxonomies, but also uncover new values that are actually encouraged in practice. Our findings highlight the need for nuanced models of desirability that go beyond preexisting prosocial measures. This work has implications for improving moderator understanding of their community values and provides a framework that can supplement qualitative approaches with larger-scale content analyses.

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

3 major / 2 minor

Summary. The paper claims that upvotes on Reddit can serve as a proxy for desirable community behavior; by applying an LLM to extract values from 16,000 highly-upvoted comments across 80 subreddits in 2016 and 2022, the authors compile 64 and 72 macro/meso/micro values respectively (retained by frequency), show that these include most prior qualitative taxonomies plus novel ones, and report that existing prosociality models capture only 18% of the extracted values on average, calling for more nuanced desirability models.

Significance. If the proxy and extraction steps are valid, the work supplies a large-scale, reproducible empirical method for surfacing community-endorsed values that existing prosocial measures miss, directly informing moderator tools and platform design; the scale (80 communities, two years) and the demonstration that new values are actually practiced give it practical utility beyond purely qualitative approaches.

major comments (3)
  1. [Abstract] Abstract (paragraph 2) and Methods: the central 82% coverage gap rests on the untested assumption that highly-upvoted comments faithfully exemplify the values communities wish to encourage; upvotes are known to be driven by orthogonal factors (humor, visibility, recency, agreement) that need not align with the extracted macro/meso/micro values, yet no validation, control condition, or robustness check against these confounds is reported.
  2. [Methods] Methods (value extraction and coverage analysis): the LLM pipeline for identifying and categorizing values lacks any reported accuracy metrics, prompt-sensitivity tests, or inter-rater agreement with human coders; without these, both the 64/72 value inventories and the subsequent claim that prosocial models miss 82% of them cannot be evaluated for reliability.
  3. [Results] Results (frequency-threshold step and 82% calculation): the retention of macro/meso/micro values is governed by an explicit frequency threshold (listed as a free parameter), yet no sensitivity analysis across thresholds or details on the exact matching procedure used to compute coverage against published prosocial models are provided; this directly affects whether the reported gap is robust.
minor comments (2)
  1. [Abstract] Abstract: the phrase 'on average 82%' should be accompanied by the exact set of prosocial models compared and the per-model coverage numbers for transparency.
  2. [Abstract] Notation: the use of italicized 'macro', 'meso', and 'micro' is introduced without an explicit definition or example in the abstract; a short parenthetical gloss would aid readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their detailed and constructive feedback. We have carefully considered each major comment and provide point-by-point responses below. Where appropriate, we have revised the manuscript to address the concerns raised.

read point-by-point responses

  1. Referee: [Abstract] Abstract (paragraph 2) and Methods: the central 82% coverage gap rests on the untested assumption that highly-upvoted comments faithfully exemplify the values communities wish to encourage; upvotes are known to be driven by orthogonal factors (humor, visibility, recency, agreement) that need not align with the extracted macro/meso/micro values, yet no validation, control condition, or robustness check against these confounds is reported.

    Authors: We acknowledge that upvotes on Reddit can be influenced by factors beyond the values expressed in the comment, such as humor or timing. However, our approach follows established practices in computational social science where aggregate upvotes serve as a signal of community endorsement and desirability. The proxy is not claimed to be perfect but provides a scalable way to surface practiced values. To address the concern, we will add a dedicated limitations subsection discussing these potential confounds and their implications for interpretation. revision: partial

  2. Referee: [Methods] Methods (value extraction and coverage analysis): the LLM pipeline for identifying and categorizing values lacks any reported accuracy metrics, prompt-sensitivity tests, or inter-rater agreement with human coders; without these, both the 64/72 value inventories and the subsequent claim that prosocial models miss 82% of them cannot be evaluated for reliability.

    Authors: This is a valid concern. The original manuscript did not include validation metrics for the LLM-based value extraction. We will revise the Methods section to include: (1) accuracy metrics from a human-coded subsample (e.g., precision/recall against expert annotations), (2) prompt sensitivity analysis by varying prompts and reporting consistency, and (3) inter-rater agreement statistics (Cohen's kappa) between the LLM and multiple human coders on a validation set. These additions will allow readers to assess the reliability of the 64/72 values and the 82% gap. revision: yes

  3. Referee: [Results] Results (frequency-threshold step and 82% calculation): the retention of macro/meso/micro values is governed by an explicit frequency threshold (listed as a free parameter), yet no sensitivity analysis across thresholds or details on the exact matching procedure used to compute coverage against published prosocial models are provided; this directly affects whether the reported gap is robust.

    Authors: We agree that sensitivity to the frequency threshold is important. In the revised manuscript, we will: (1) report results for a range of thresholds (e.g., values appearing in at least 5%, 10%, 20% of communities) to show robustness of the 64/72 counts, (2) provide the exact procedure for matching extracted values to those in prior prosocial models, including how we handled semantic similarity, and (3) recompute the coverage gap under different thresholds to confirm the 82% figure is stable. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical coverage count against external models

full rationale

The paper performs an empirical extraction of values from highly-upvoted Reddit comments using an LLM, compiles frequency-based lists of macro/meso/micro values, and reports that existing prosociality models cover only 18% on average. This 82% figure is a direct empirical count, not a fitted parameter or self-referential prediction. The assumption that upvotes proxy desirability is stated explicitly but functions as an input premise rather than a derived result that loops back to itself. No equations, self-citations, or ansatzes reduce the central claim to its own inputs by construction. The derivation chain is self-contained against external published models.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim depends on three untested modeling choices: (1) upvotes equal desirability, (2) LLM extraction faithfully surfaces community values, and (3) frequency across communities is the right aggregation rule. No free parameters are explicitly fitted in the abstract, but the frequency threshold for retaining a value functions as an implicit free parameter.

free parameters (1)
  • frequency threshold for macro/meso/micro value retention
    Values are retained only if they appear across multiple communities; the exact cutoff is not stated but directly determines the final lists of 64 and 72 values.
axioms (2)
  • domain assumption Upvotes reliably indicate community approval of the underlying value expressed in a comment
    Invoked in abstract paragraph 2 when upvotes are chosen as the proxy for desirability.
  • domain assumption LLM can accurately extract and categorize values from short text without systematic bias
    Implicit in the extraction step that produces the 64/72 value sets.

pith-pipeline@v0.9.0 · 5842 in / 1424 out tokens · 22640 ms · 2026-05-23T18:49:31.231820+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

66 extracted references · 66 canonical work pages · 6 internal anchors

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

    work page

  2. [2]

    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 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

    work page

  3. [3]

    J.; Trager, J.; Park, P

    Abdurahman, S.; Atari, M.; Karimi-Malekabadi, F.; Xue, M. J.; Trager, J.; Park, P. S.; Golazizian, P.; Omrani, A.; and Dehghani, M. 2024. Perils and opportunities in using large language models in psychological research. PNAS nexus, 3(7): pgae245

    work page 2024

  4. [4]

    GPT-4 Technical Report

    Achiam, J.; Adler, S.; Agarwal, S.; Ahmad, L.; Akkaya, I.; Aleman, F. L.; Almeida, D.; Altenschmidt, J.; Altman, S.; Anadkat, S.; et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  5. [5]

    Bao, J.; Wu, J.; Zhang, Y.; Chandrasekharan, E.; and Jurgens, D. 2021. Conversations Gone Alright : Quantifying and Predicting Prosocial Outcomes in Online Conversations . In Proceedings of the Web Conference 2021 , 1134--1145. Ljubljana Slovenia: ACM. ISBN 978-1-4503-8312-7

    work page 2021

  6. [6]

    Bicchieri, C.; and Muldoon, R. 2011. Social Norms

    work page 2011

  7. [7]

    M.; Ng, A

    Blei, D. M.; Ng, A. Y.; and Jordan, M. I. 2003. Latent dirichlet allocation. Journal of machine Learning research, 3(Jan): 993--1022

    work page 2003

  8. [8]

    Bonikowski, B.; and Nelson, L. K. 2022. From ends to means: The promise of computational text analysis for theoretically driven sociological research. Sociological Methods & Research, 51(4): 1469--1483

    work page 2022

  9. [9]

    D.; and Smart, S

    Brown, J. D.; and Smart, S. 1991. The self and social conduct: Linking self-representations to prosocial behavior. Journal of Personality and Social psychology, 60(3): 368

    work page 1991

  10. [10]

    Brown, T. B. 2020. Language models are few-shot learners. arXiv preprint arXiv:2005.14165

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  11. [11]

    W.; and Gilbert, E

    Chandrasekharan, E.; Gandhi, C.; Mustelier, M. W.; and Gilbert, E. 2019. Crossmod: A Cross - Community Learning -based System to Assist Reddit Moderators . In Proceedings of the ACM on Human - Computer Interaction , volume 3, 1--30

    work page 2019

  12. [12]

    Chandrasekharan, E.; Jhaver, S.; Bruckman, A.; and Gilbert, E. 2022. Quarantined! Examining the Effects of a Community - Wide Moderation Intervention on Reddit . In ACM Transactions on Computer - Human Interaction , volume 29, 1--26

    work page 2022

  13. [13]

    Chandrasekharan, E.; Pavalanathan, U.; Srinivasan, A.; Glynn, A.; Eisenstein, J.; and Gilbert, E. 2017. You Can 't Stay Here : The Efficacy of Reddit 's 2015 Ban Examined Through Hate Speech . In Proceedings of the ACM on Human - Computer Interaction , volume 1, 1--22

    work page 2017

  14. [14]

    Chandrasekharan, E.; Samory, M.; Jhaver, S.; Charvat, H.; Bruckman, A.; Lampe, C.; Eisenstein, J.; and Gilbert, E. 2018. The Internet 's Hidden Rules : An Empirical Study of Reddit Norm Violations at Micro , Meso , and Macro Scales . In Proceedings of the ACM on Human - Computer Interaction , volume 2, 1--25

    work page 2018

  15. [15]

    P.; and Danescu-Niculescu-Mizil, C

    Chang, J. P.; and Danescu-Niculescu-Mizil, C. 2019. Trajectories of Blocked Community Members : Redemption , Recidivism and Departure . In Proceedings of WWW

    work page 2019

  16. [16]

    Chew, R.; Bollenbacher, J.; Wenger, M.; Speer, J.; and Kim, A. 2023. LLM-assisted content analysis: Using large language models to support deductive coding. arXiv preprint arXiv:2306.14924

    work page arXiv 2023

  17. [17]

    Choi, F.; Bajpai, T.; Pratipati, S.; and Chandrasekharan, E. 2023. ConvEx : A Visual Conversation Exploration System for Discord Moderators . In Proceedings of the ACM on Human - Computer Interaction , volume 7, 262:1--262:30

    work page 2023

  18. [18]

    Choi, F.; Lambert, C.; Koshy, V.; Pratipati, S.; Do, T.; and Chandrasekharan, E. 2024. Creator Hearts : Investigating the Impact Positive Signals from YouTube Creators in Shaping Comment Section Behavior . ArXiv:2404.03612 [cs]

    work page arXiv 2024

  19. [19]

    L.; Cai, J.; and Wohn, D

    Cook, C. L.; Cai, J.; and Wohn, D. Y. 2022. Awe Versus Aww : The Effectiveness of Two Kinds of Positive Emotional Stimulation on Stress Reduction for Online Content Moderators . ArXiv:2202.05964 [cs]

    work page arXiv 2022

  20. [20]

    Cunha, T.; Weber, I.; and Pappa, G. 2017. A Warm Welcome Matters !: The Link Between Social Feedback and Weight Loss in /r/loseit. In Proceedings of the 26th International Conference on World Wide Web Companion - WWW '17 Companion , 1063--1072. Perth, Australia: ACM Press. ISBN 978-1-4503-4914-7

    work page 2017

  21. [21]

    Danescu-Niculescu-Mizil, C.; Sudhof, M.; Jurafsky, D.; Leskovec, J.; and Potts, C. 2013. A computational approach to politeness with application to social factors. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics ( Volume 1: Long Papers ) , 250--259. Sofia, Bulgaria: Association for Computational Linguistics

    work page 2013

  22. [22]

    Diakopoulos, N. A. 2015. The Editor 's Eye : Curation and Comment Relevance on the New York Times . In Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing , 1153--1157. ACM

    work page 2015

  23. [23]

    Dignan, L. 2024. Reddit's data licensing play: Do you want your LLM trained on Reddit data?

    work page 2024

  24. [24]

    Dunivin, Z. O. 2024. Scalable Qualitative Coding with LLMs: Chain-of-Thought Reasoning Matches Human Performance in Some Hermeneutic Tasks. arXiv preprint arXiv:2401.15170

    work page arXiv 2024

  25. [25]

    Gilardi, F.; Alizadeh, M.; and Kubli, M. 2023. ChatGPT outperforms crowd workers for text-annotation tasks. Proceedings of the National Academy of Sciences, 120(30): e2305016120

    work page 2023

  26. [26]

    Grimmelmann, J. 2015. The virtues of moderation. Yale JL & Tech., 17: 42. Publisher: HeinOnline

    work page 2015

  27. [27]

    Halfaker, A.; Kittur, A.; and Riedl, J. 2011. Don't bite the newbies: how reverts affect the quantity and quality of Wikipedia work. In Proceedings of the 7th International Symposium on Wikis and Open Collaboration , 163--172. Mountain View California: ACM. ISBN 978-1-4503-0909-7

    work page 2011

  28. [28]

    F.; and Krippendorff, K

    Hayes, A. F.; and Krippendorff, K. 2007. Answering the call for a standard reliability measure for coding data. Communication methods and measures, 1(1): 77--89

    work page 2007

  29. [29]

    Jhaver, S.; Boylston, C.; Yang, D.; and Bruckman, A. 2021. Evaluating the Effectiveness of Deplatforming as a Moderation Strategy on Twitter . In Proceedings of the ACM on Human - Computer Interaction , volume 5, 1--30

    work page 2021

  30. [30]

    Jurgens, D.; Chandrasekharan, E.; and Hemphill, L. 2019. A Just and Comprehensive Strategy for Using NLP to Address Online Abuse . arXiv. ArXiv:1906.01738 [cs]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  31. [31]

    Kiesler, S.; Kraut, R.; Resnick, P.; and Kittur, A. 2012. Regulating behavior in online communities. Building successful online communities: Evidence-based social design. Publisher: MIT Press, Cambridge, MA, USA

    work page 2012

  32. [32]

    S.; Reid, M.; Matsuo, Y.; and Iwasawa, Y

    Kojima, T.; Gu, S. S.; Reid, M.; Matsuo, Y.; and Iwasawa, Y. 2022. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35: 22199--22213

    work page 2022

  33. [33]

    Kolhatkar, V.; and Taboada, M. 2017. Constructive language in news comments. In Proceedings of the first workshop on abusive language online, 11--17

    work page 2017

  34. [34]

    E.; Resnick, P.; Kiesler, S.; Burke, M.; Chen, Y.; Kittur, N.; Konstan, J.; Ren, Y.; and Riedl, J

    Kraut, R. E.; Resnick, P.; Kiesler, S.; Burke, M.; Chen, Y.; Kittur, N.; Konstan, J.; Ren, Y.; and Riedl, J. 2011. Building Successful Online Communities : Evidence - Based Social Design . The MIT Press. ISBN 978-0-262-01657-5

    work page 2011

  35. [35]

    Kumar, S.; Cheng, J.; and Leskovec, J. 2017. Antisocial behavior on the web: Characterization and detection. In Proceedings of the 26th International Conference on World Wide Web Companion, 947--950

    work page 2017

  36. [36]

    Lambert, C.; Choi, F.; and Chandrasekharan, E. 2024. ``Positive reinforcement helps breed positive behavior'': Moderator Perspectives on Encouraging Desirable Behavior. Proceedings of the ACM on Human-Computer Interaction, (CSCW)

    work page 2024

  37. [37]

    Lambert, C.; Rajagopal, A.; and Chandrasekharan, E. 2022. Conversational Resilience : Quantifying and Predicting Conversational Outcomes Following Adverse Events . In Proceedings of the International AAAI Conference on Web and Social Media , volume 16, 548--559

    work page 2022

  38. [38]

    Lambert, C.; Saha, K.; and Chandrasekharan, E. 2025. Does Positive Reinforcement Work?: A Quasi-Experimental Study of the Effects of Positive Feedback on Reddit. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, CHI '25. New York, NY, USA: Association for Computing Machinery. ISBN 9798400713941

    work page 2025

  39. [39]

    Liu, P.; Guberman, J.; Hemphill, L.; and Culotta, A. 2018. Forecasting the presence and intensity of hostility on Instagram using linguistic and social features. In Proceedings of the International AAAI Conference on Web and Social Media, volume 12

    work page 2018

  40. [40]

    McClintock, C. G. 1978. Social values: Their definition, measurement and development. Journal of Research & Development in Education

    work page 1978

  41. [41]

    Niculae, V.; and Danescu-Niculescu-Mizil, C. 2016. Conversational Markers of Constructive Discussions . In Proceedings of NAACL - HLT , 568--578

    work page 2016

  42. [42]

    Ouyang, L.; Wu, J.; Jiang, X.; Almeida, D.; Wainwright, C.; Mishkin, P.; Zhang, C.; Agarwal, S.; Slama, K.; Ray, A.; et al. 2022. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35: 27730--27744

    work page 2022

  43. [43]

    Y.; Li, S

    Park, C. Y.; Li, S. S.; Jung, H.; Volkova, S.; Mitra, T.; Jurgens, D.; and Tsvetkov, Y. 2024. ValueScope: Unveiling Implicit Norms and Values via Return Potential Model of Social Interactions. arXiv preprint arXiv:2407.02472

    work page arXiv 2024

  44. [44]

    S.; Seering, J.; and Bernstein, M

    Park, J. S.; Seering, J.; and Bernstein, M. S. 2022. Measuring the Prevalence of Anti - Social Behavior in Online Communities . arXiv. ArXiv:2208.13094 [cs]

    work page arXiv 2022

  45. [45]

    Pavlopoulos, J.; Malakasiotis, P.; and Androutsopoulos, I. 2017. Deeper attention to abusive user content moderation. In Proceedings of the 2017 conference on empirical methods in natural language processing, 1125--1135

    work page 2017

  46. [46]

    Peters, H.; and Matz, S. C. 2024. Large language models can infer psychological dispositions of social media users. PNAS nexus, 3(6): pgae231

    work page 2024

  47. [47]

    Reimers, N.; and Gurevych, I. 2019. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics

    work page 2019

  48. [48]

    XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models

    R \"o ttger, P.; Kirk, H. R.; Vidgen, B.; Attanasio, G.; Bianchi, F.; and Hovy, D. 2023. Xstest: A test suite for identifying exaggerated safety behaviours in large language models. arXiv preprint arXiv:2308.01263

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  49. [49]

    Saha, K.; Chandrasekharan, E.; and De Choudhury, M. 2019. Prevalence and Psychological Effects of Hateful Speech in Online College Communities . In Proceedings of the 10th ACM Conference on Web Science , 255--264. Boston Massachusetts USA: ACM. ISBN 978-1-4503-6202-3

    work page 2019

  50. [50]

    Saha, K.; and Sharma, A. 2020. Causal Factors of Effective Psychosocial Outcomes in Online Mental Health Communities . Proceedings of the International AAAI Conference on Web and Social Media, 14: 590--601

    work page 2020

  51. [51]

    L.; Pennebaker, J

    Saha, K.; Yousuf, A.; Boyd, R. L.; Pennebaker, J. W.; and De Choudhury, M. 2022. Social Media Discussions Predict Mental Health Consultations on College Campuses . Scientific Reports, 12(1): 123

    work page 2022

  52. [52]

    Team, G.; Kamath, A.; Ferret, J.; Pathak, S.; Vieillard, N.; Merhej, R.; Perrin, S.; Matejovicova, T.; Ram \'e , A.; Rivi \`e re, M.; et al. 2025. Gemma 3 technical report. arXiv preprint arXiv:2503.19786

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  53. [53]

    Verma, G.; Bhardwaj, A.; Aledavood, T.; De Choudhury, M.; and Kumar, S. 2022. Examining the impact of sharing COVID -19 misinformation online on mental health. Scientific Reports, 12(1): 8045

    work page 2022

  54. [54]

    P.; Prabhakaran, V.; Hamilton, W

    Voigt, R.; Camp, N. P.; Prabhakaran, V.; Hamilton, W. L.; Hetey, R. C.; Griffiths, C. M.; Jurgens, D.; Jurafsky, D.; and Eberhardt, J. L. 2017. Language from police body camera footage shows racial disparities in officer respect. Proceedings of the National Academy of Sciences, 114(25): 6521--6526

    work page 2017

  55. [55]

    Wang, Z.; and Jurgens, D. 2018 a . It’s going to be okay: Measuring access to support in online communities. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, 33--45

    work page 2018

  56. [56]

    Wang, Z.; and Jurgens, D. 2018 b . It’s going to be okay: Measuring access to support in online communities. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing , 33--45

    work page 2018

  57. [57]

    Warner, W.; and Hirschberg, J. 2012. Detecting hate speech on the world wide web. In Proceedings of the second workshop on language in social media, 19--26

    work page 2012

  58. [58]

    V.; Zhou, D.; et al

    Wei, J.; Wang, X.; Schuurmans, D.; Bosma, M.; Xia, F.; Chi, E.; Le, Q. V.; Zhou, D.; et al. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35: 24824--24837

    work page 2022

  59. [59]

    X.; and Althoff, T

    Weld, G.; Zhang, A. X.; and Althoff, T. 2021. Making Online Communities ' Better ': A Taxonomy of Community Values on Reddit . Publisher: arXiv Version Number: 2

    work page 2021

  60. [60]

    X.; and Althoff, T

    Weld, G.; Zhang, A. X.; and Althoff, T. 2022. What Makes Online Communities ‘ Better ’? Measuring Values , Consensus , and Conflict across Thousands of Subreddits . Proceedings of the International AAAI Conference on Web and Social Media, 16: 1121--1132

    work page 2022

  61. [61]

    X.; and Althoff, T

    Weld, G.; Zhang, A. X.; and Althoff, T. 2024. Making Online Communities ‘ Better ’: A Taxonomy of Community Values on Reddit . Proceedings of the International AAAI Conference on Web and Social Media, 18: 1611--1633

    work page 2024

  62. [62]

    Williamson, D. 2020. US social media usage: How the coronavirus is changing consumer behavior. eMarketer, 2: 1649--023

    work page 2020

  63. [63]

    Wold, S.; Esbensen, K.; and Geladi, P. 1987. Principal component analysis. Chemometrics and intelligent laboratory systems, 2(1-3): 37--52

    work page 1987

  64. [64]

    Conversations Gone Awry: Detecting Early Signs of Conversational Failure

    Zhang, J.; Chang, J. P.; Danescu-Niculescu-Mizil, C.; Dixon, L.; Hua, Y.; Thain, N.; and Taraborelli, D. 2018. Conversations Gone Awry : Detecting Early Signs of Conversational Failure . arXiv. ArXiv:1805.05345 [physics]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  65. [65]

    Zhou, N.; and Jurgens, D. 2020. Condolence and empathy in online communities. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), 609--626

    work page 2020

  66. [66]

    Ziems, C.; Held, W.; Shaikh, O.; Chen, J.; Zhang, Z.; and Yang, D. 2024. Can large language models transform computational social science? Computational Linguistics, 50(1): 237--291

    work page 2024