Pith. sign in

REVIEW 3 major objections 4 minor 68 references

Signals in the Noise: Decoding Unexpected Engagement Patterns on Twitter

T0 review · 3 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read Content, not popularity, decides a tweet's surprise engagement

desk verdict A clean residual-based metric for engagement composition, but the §3.1 filter requiring all three engagement types may bias the topic comparisons that anchor the paper's central claims. read the letter →

arxiv 2509.08128 v1 pith:DK2VT2V3 submitted 2025-09-09 cs.SI cs.CYcs.HC

Yulin Yu , Houming Chen , Daniel Romero , Paramveer S. Dhillon This is my paper · ORCID
classification cs.SIcs.CYcs.HC
keywords unexpectedengagementcompositionTwitter/Xtypessignalingtheoryattentioneconomyquantileregressionsocialmediacontentfeatures
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 tries to establish that on Twitter, the kind of engagement a post over-achieves on—likes, retweets, or comments—is systematically driven by the tweet's topic and wording, not just by how popular it is. The authors construct an 'unexpectedness quotient' for each interaction type by dividing observed counts by the count predicted from the other two types, which strips out overall popularity and leaves a measure of compositional surprise. Applying this to 642,108 tweets from 2018, they find stable associations: news, politics, and business text over-achieves on retweets and comments; games, sports, art, celebrity, and movie text over-achieves on likes (and often retweets); subjective posts attract extra likes, objective posts attract extra retweets; and URL-rich, longer, harder-to-read posts attract extra retweets. If these patterns hold, they give content creators and platform designers a way to aim for a specific interaction type rather than raw volume, and they undermine the common practice of treating engagement as a single thing that varies only in intensity.

What carries the argument

The unexpectedness quotient, defined as the ratio of a tweet's observed engagement count for one type to its predicted count for that type. The prediction comes from a 90th-percentile quantile regression of each engagement type on the other two (e.g., likes ~ retweets + comments), chosen for robustness to power-law-skewed, outlier-heavy social media data. The quotient normalizes away overall popularity and isolates compositional deviation; a second-stage OLS regression of the log quotient on content and author features then identifies which attributes drive each type of unexpected engagement.

What would settle it

Compute the same quotient on the full set of 2018 hashtag tweets without the 'at least one of each engagement type' filter (adding a smoothing constant before taking ratios) and re-estimate the OLS models; if the topic coefficients for unexpected retweets and comments change sign or shrink to zero, the paper's central pattern is an artifact of the filter rather than a property of engagement.

Watch

Extended reading notes

Core claim

This paper claims that on Twitter, which engagement type a tweet over-achieves on is systematically predictable from the tweet's topic and linguistic style, independent of overall popularity. Using 642,108 English hashtag-containing tweets from 2018, the authors compute an unexpectedness quotient for likes, retweets, and comments: each observed count is divided by the count predicted from the other two counts using 90th-percentile quantile regression, so a high quotient means that engagement type is disproportionately strong for that tweet. Regressing the log quotient on topic, sentiment, subjectivity, readability, length, concreteness, and URL presence, they find that news, politics, and bu

Load-bearing premise

The central result depends on restricting the dataset to tweets that already have at least one like, one retweet, and one comment; if that filter systematically drops the tweets with the most extreme engagement compositions, the measured topic-content relationships may not describe engagement in general.

Editorial extensions

If this is right

  • Content creators can target a specific engagement type by adjusting what they post: objective, URL-rich, moderately complex text for retweets; subjective, personal, concrete language for likes; longer, discussion-prompting text for comments.
  • Platforms can flag 'surprising' content algorithmically without waiting for it to trend overall, because the quotient identifies tweets whose engagement composition is anomalous relative to their volume.
  • Engagement baselines should be topic-specific: judging a news tweet's comment count against the whole feed rather than against other news tweets will misclassify normal behavior as unexpected.
  • Composite engagement scores that sum likes, retweets, and comments obscure the structure documented here; a quotient-based decomposition is a workable alternative for any platform with differentiated interaction actions.
  • The higher average unexpectedness of comments implies that prompting discussion is a more attainable goal than generating unexpected sharing, a distinction with direct design consequences for comment-heavy community features.

Reading between the lines

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

  • Editorial inference: The 'at least one like, one retweet, and one comment' filter likely removes the most compositionally extreme tweets (e.g., a widely retweeted post with zero comments), so the measured effects may understate how strongly content shapes engagement composition; reanalysing with a zero-handling method is a direct test.
  • Editorial inference: The quotient is portable to platforms with distinct actions such as Facebook reactions versus shares or LinkedIn reactions versus reposts, so if the cost-signal interpretation is right, analogous topic-to-composition patterns should appear there; a cross-platform replication would separate platform affordances from general signaling behavior.
  • Editorial inference: Because the data end in 2018, before the algorithmic home timeline settled and before the platform rebrand, the paper's temporal-stability argument is indirect; recomputing the same coefficients on post-2020 data would either support or overturn the claim that the relationships persist.
  • Editorial inference: The finding that subjective content attracts unexpected likes while objective content attracts unexpected retweets suggests a testable content strategy—swapping a post's framing from personal to factual should shift its over-achieving engagement type in a predictable direction.
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

3 major / 4 minor

Summary. The paper proposes an 'unexpectedness quotient' for Twitter engagement: for each tweet, quantile regressions of likes, retweets, and comments on the other two counts (Eqs. 1–3) produce a predicted value, and the quotient is observed/predicted (Eqs. 4–6). The quotient is then log-transformed and regressed on textual complexity, valence, topic, and author controls (Eqs. 7–9). Using 642,108 hashtag-containing tweets from 2018 with at least one like, retweet, and comment, the authors report that news/politics/business topics show higher-than-expected retweets/comments, games/sports show higher-than-expected likes/comments, subjective content attracts likes, objective content attracts retweets, and longer/URL-containing tweets attract retweets.

Significance. If the findings are valid, the paper provides a novel metric and a large descriptive map of how content features relate to engagement composition. Strengths include the large sample, the two-stage design, robustness checks at alternative quantiles (75%, 50%), a threshold robustness check, and 10-fold cross-validation, as well as an explicit acknowledgment that the analysis is correlational. The quotient is a portable descriptive tool that could be applied to other platforms. However, the validity of the central topic claims depends heavily on the sample-selection rule in §3.1, which has not been adequately stress-tested.

major comments (3)
  1. [§3.1 and §4.2/Fig. 6(b)] Requiring at least one like, retweet, and comment truncates the joint engagement distribution on every margin. The first-stage quantile regressions (Eqs. 1–3) and the second-stage topic coefficients (Eqs. 7–9) are therefore conditional on survival of this filter. If the probability of receiving at least one of each type varies by topic (e.g., comment propensities are lower for art/games than for politics), the tweets retained within each topic are a selected sample: an art tweet that receives any comment is likely to have an unusually high comment count, mechanically inflating its comment quotient. The robustness check in Fig. 6(b) only moves the truncation from 1 to 10 in one interaction and does not test whether the filter itself distorts topic comparisons. Please report the fraction and topic-specific rates of tweets excluded by the all-three filter, and provide an analysis that addre
  2. [§4.1 and §4.3] The quotient in Eqs. (4)–(6) is computed from quantile regressions fit on the same 642,108 tweets used in the second-stage OLS (Eqs. 7–9). The reported standard errors and p-values in Fig. 6 do not account for the first-stage estimation error, so they are likely understated. Section 4.3's cross-validation description is too brief to address this: it states that the quotient is recomputed per fold, but it does not say whether the second stage is refit within folds or how coefficient variation is measured, and it reports no corrected standard errors. Please add a bootstrap or delta-method procedure that propagates first-stage uncertainty, or explicitly justify that with two regressors and N>600k the effect on inference is negligible.
  3. [§4.1 and abstract] The denominator in Eqs. (4)–(6) is the conditional 90th percentile of the target engagement given the other two, not the conditional expectation. Calling this 'expected' (throughout §4 and the abstract) is therefore misleading: a quotient above 1 means 'in the top decile of the conditional distribution,' not 'higher than average/expected.' The choice of the 90th percentile is also a free parameter, although the 50th/75th percentile robustness checks mitigate this concern. Please rename the denominator (e.g., 'predicted high-level engagement') and interpret the quotient accordingly, or switch the benchmark to the conditional median with robustness to the quantile.
minor comments (4)
  1. [§5.1] The sentence 'Subjective tweets promoted more likes than expected, while objective tweets gathered unexpectedly high comments' contradicts §4.2.1 and the abstract, which state that objective tweets receive more unexpected retweets. Please correct this inconsistency.
  2. [§6.1 vs §4.3] Section 6.1 claims 'random half-sample replications' and 'five-fold cross-validation,' while §4.3 reports 10-fold cross-validation and no half-sample analysis. These robustness checks are not described elsewhere in the paper. Either add the promised analyses or revise the claims to match what is actually reported.
  3. [Figure 6 caption] The note says '95% confidential interval' instead of 'confidence interval.' Also, Table 3 has a typo: 'dateset' should be 'dataset.'
  4. [§3.1] The paper says 'we only included tweets that received at least one like, one retweet, and one comment.' This should be explicitly framed as a definition of the study population in the title/abstract; otherwise the abstract's claim of analyzing 'over 600,000 tweets' may suggest a broader sample than is actually studied.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the two-stage residual design is self-contained; the only self-citation (topic taxonomy) is a published external instrument, not a derivation of the findings.

full rationale

The paper's derivation chain is a two-step residual analysis. Equations (1)-(3) fit 90th-percentile quantile regressions of each engagement count on the other two engagement counts; Equations (4)-(6) define the unexpectedness quotient as the ratio of the observed to the fitted value. Because the first-stage regressors are only the other engagement counts, the quotient is a residual that does not encode the content features used in Equations (7)-(9). Regressing log E on complexity, valence, topic, and author features is therefore a standard second-stage association analysis, not a prediction of a quantity that was fit from those same features. No equation-level circularity exists. The topic taxonomy is credited to Romero et al. [45], a prior published paper with a co-author overlap, and the authors explicitly extend it with three new manually inspected categories; this is a measurement instrument, not a uniqueness theorem or ansatz, and it does not presuppose the paper's findings. The sample filter requiring at least one like, retweet, and comment is a data-selection concern that could bias topic comparisons, but it is a correctness/validity issue, not a circularity of derivation. Overall, the central claim is an empirical regularity derived from data, not an input recycled as an output.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The central claim relies on standard text-analysis tools, a hashtag-based topic taxonomy, a hand-chosen quantile level, and a linear-quantile functional form. No new physical or conceptual entities are postulated; the unexpectedness quotient is a constructed metric, not an entity with independent evidence outside the paper.

free parameters (1)
  • quantile level for expected engagement = 0.90 (tested at 0.75, 0.50)
    Hand-chosen threshold in Equations 1-3 defines the benchmark 'expected' engagement level; the quotient's scale and interpretation depend on this choice, though robustness checks were run at lower quantiles.
assumptions (5)
  • domain assumption Likes, retweets, and comments form a stable cost/effort hierarchy (low, medium, high) that users internalize.
    The interpretation of unexpectedness in Sections 4.1 and 5 relies on Signaling Theory treating likes as cheap, retweets as moderate, and comments as costly.
  • ad hoc to paper The relationship between engagement counts is adequately captured by a linear quantile regression at the 90th percentile with the other two engagement counts as predictors.
    Equations 1-3 assume linearity in counts and a single quantile level; if the true conditional relationship is nonlinear or varies across topics or author types, the quotient would be misspecified.
  • domain assumption Hashtags are reliable topic markers and the manual 11-topic annotation is accurate.
    Section 3.2.3 builds topic features on this assumption; mislabeled topics could bias coefficient estimates.
  • domain assumption VADER and TextBlob subjectivity correctly measure sentiment and subjectivity in short text.
    Sentiment and subjectivity features in Section 3.2.2 are standard tools, but measurement error in these proxies could attenuate or distort effects.
  • domain assumption The 2018 Decahose sample and the hashtag filter provide a representative enough corpus for the studied relationships.
    Section 3.1 uses a 10% random sample and requires hashtags; the paper compares hashtag and non-hashtag tweets, but does not fully establish representativeness of the full Twitter population.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Signals in the Noise: Decoding Unexpected Engagement Patterns on Twitter." pith.science (2026). https://pith.science/paper/DK2VT2V3

@misc{pith2026250908128,
  author       = {Pith},
  title        = {Pith review of: Signals in the Noise: Decoding Unexpected Engagement Patterns on Twitter},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DK2VT2V3}},
  note         = {Machine review of arXiv:2509.08128}
}
read the original abstract

Social media platforms offer users multiple ways to engage with content--likes, retweets, and comments--creating a complex signaling system within the attention economy. While previous research has examined factors driving overall engagement, less is known about why certain tweets receive unexpectedly high levels of one type of engagement relative to others. Drawing on Signaling Theory and Attention Economy Theory, we investigate these unexpected engagement patterns on Twitter (now known as "X"), developing an "unexpectedness quotient" to quantify deviations from predicted engagement levels. Our analysis of over 600,000 tweets reveals distinct patterns in how content characteristics influence unexpected engagement. News, politics, and business tweets receive more retweets and comments than expected, suggesting users prioritize sharing and discussing informational content. In contrast, games and sports-related topics garner unexpected likes and comments, indicating higher emotional investment in these domains. The relationship between content attributes and engagement types follows clear patterns: subjective tweets attract more likes while objective tweets receive more retweets, and longer, complex tweets with URLs unexpectedly receive more retweets. These findings demonstrate how users employ different engagement types as signals of varying strength based on content characteristics, and how certain content types more effectively compete for attention in the social media ecosystem. Our results offer valuable insights for content creators optimizing engagement strategies, platform designers facilitating meaningful interactions, and researchers studying online social behavior.

Figures

Figures reproduced from arXiv: 2509.08128 by the authors.

Figure 1
Figure 1. Empirical distribution of the readability, length, and concreteness features. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Empirical distribution of the sentiment and subjectivity features. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Empirical distribution of the author features: number of followers and number of tweets. [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Empirical distribution of the number of likes, retweets, and comments. [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Unexpectedness quotient for likes, retweets, and comments. [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: OLS Regression estimation results.The left panel (a) shows point estimates for equation ( [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: OLS Regression estimation results for non-linear effect. The panel shows point estimates for equation ( [PITH_FULL_IMAGE:figures/full_fig_p030_7.png]
Figure 8
Figure 8. Figure 8: OLS regression results examining the interaction effect on the outcome variable: unexpected likes. [PITH_FULL_IMAGE:figures/full_fig_p031_8.png]
Figure 9
Figure 9. Figure 9: OLS regression results examining the interaction effect on the outcome variable: unexpected retweet. [PITH_FULL_IMAGE:figures/full_fig_p032_9.png]
Figure 10
Figure 10. Figure 10: OLS regression results examining the interaction effect on the outcome variable: unexpected comment. [PITH_FULL_IMAGE:figures/full_fig_p033_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 58 canonical work pages

  1. [1]

    Kholoud Khalil Aldous, Jisun An, and Bernard J Jansen. 2019. View, like, comment, post: Analyzing user engagement by topic at 4 levels across 5 social media platforms for 53 news organizations. InProceedings of the International AAAI Conference on Web and Social Media, Vol. 13. 47–57

    work page 2019

  2. [2]

    2008.Mostly harmless econometrics: An empiricist’s companion

    Joshua D Angrist and Jörn-Steffen Pischke. 2008.Mostly harmless econometrics: An empiricist’s companion. Princeton university press

    work page 2008

  3. [3]

    Eytan Bakshy, Jake M Hofman, Winter A Mason, and Duncan J Watts. 2011. Everyone’s an influencer: quantifying influence on twitter. InProceedings of the fourth ACM international conference on Web search and data mining. 65–74

    work page 2011

  4. [4]

    Jonah Berger and Katherine L Milkman. 2012. What makes online content viral?Journal of marketing research49, 2 (2012), 192–205

    work page 2012

  5. [5]

    Michael S Bernstein, Eytan Bakshy, Moira Burke, and Brian Karrer. 2013. Quantifying the invisible audience in social networks. InProceedings of the SIGCHI conference on human factors in computing systems. 21–30

    2013

  6. [6]

    danah boyd, Scott Golder, and Gilad Lotan. 2010. Tweet, tweet, retweet: Conversational aspects of retweeting on twitter. In2010 43rd Hawaii international conference on system sciences. IEEE, 1–10

    work page 2010

  7. [7]

    William J Brady, Julian A Wills, John T Jost, Joshua A Tucker, and Jay J Van Bavel. 2017. Emotion shapes the diffusion of moralized content in social networks.Proceedings of the National Academy of Sciences114, 28 (2017), 7313–7318

    2017

  8. [8]

    Axel Bruns. 2021. After the ‘APIcalypse’: Social media platforms and their fight against critical scholarly research. Disinformation and data lockdown on social platforms(2021), 14–36

    work page 2021

Show all 68 references
  1. [9]

    Marc Brysbaert, Amy Beth Warriner, and Victor Kuperman. 2014. Concreteness ratings for 40 thousand generally known English word lemmas.Behavior research methods46, 3 (2014), 904–911

    2014

  2. [10]

    Moira Burke and Robert E Kraut. 2016. The relationship between Facebook use and well-being depends on communi- cation type and tie strength.Journal of computer-mediated communication21, 4 (2016), 265–281

    2016

  3. [11]

    Moira Burke, Cameron Marlow, and Thomas Lento. 2010. Social network activity and social well-being. InProceedings of the SIGCHI conference on human factors in computing systems. 1909–1912

    2010

  4. [12]

    Meeyoung Cha, Hamed Haddadi, Fabricio Benevenuto, and Krishna Gummadi. 2010. Measuring user influence in twitter: The million follower fallacy. InProceedings of the international AAAI conference on web and social media, Vol. 4. 10–17

    2010

  5. [13]

    Brian L Connelly, S Trevis Certo, R Duane Ireland, and Christopher R Reutzel. 2011. Signaling theory: A review and assessment.Journal of management37, 1 (2011), 39–67

    2011

  6. [14]

    Michele Coscia. 2017. Popularity spikes hurt future chances for viral propagation of protomemes.Commun. ACM61, 1 (2017), 70–77

    2017

  7. [15]

    Thomas H Davenport and John C Beck. 2001. The attention economy.Ubiquity2001, May (2001), 1–es

    2001

  8. [16]

    Judith Donath. 2007. Signals in social supernets.Journal of computer-mediated communication13, 1 (2007), 231–251

    2007

  9. [17]

    Georg Franck. 2019. The economy of attention.Journal of sociology55, 1 (2019), 8–19

    2019

  10. [18]

    Deen Freelon. 2018. Computational research in the post-API age.Political Communication35, 4 (2018), 665–668

    2018

  11. [19]

    Eric Gilbert. 2013. Widespread underprovision on reddit. InProceedings of the 2013 conference on Computer supported cooperative work. 803–808

    2013

  12. [20]

    Scott A Golder and Michael W Macy. 2014. Digital footprints: Opportunities and challenges for online social research. Annual review of sociology40, 1 (2014), 129–152

    2014

  13. [21]

    Michael H Goldhaber. 1997. The attention economy and the net.First Monday(1997)

    1997

  14. [22]

    Sandra Gonzalez-Bailon, Andreas Kaltenbrunner, and Rafael E Banchs. 2010. The structure of political discussion networks: a model for the analysis of online deliberation.Journal of Information Technology25, 2 (2010), 230–243

    2010

  15. [23]

    Catherine Grevet, Loren G Terveen, and Eric Gilbert. 2014. Managing political differences in social media. InProceedings of the 17th ACM conference on Computer supported cooperative work & social computing. 1400–1408

    2014

  16. [24]

    Lars Kai Hansen, Adam Arvidsson, Finn Aarup Nielsen, Elanor Colleoni, and Michael Etter. 2011. Good friends, bad news-affect and virality in twitter. InFuture Information Technology: 6th International Conference, FutureTech 2011, Loutraki, Greece, June 28-30, 2011, Proceedings...

    2011

  17. [25]

    Xiangnan He, Ming Gao, Min-Yen Kan, Yiqun Liu, and Kazunari Sugiyama. 2014. Predicting the popularity of web 2.0 items based on user comments. InProceedings of the 37th international ACM SIGIR conference on Research & development in information retrieval. 233–242

    2014

  18. [26]

    Bernardo A Huberman, Daniel M Romero, and Fang Wu. 2008. Social networks that matter: Twitter under the microscope.arXiv preprint arXiv:0812.1045(2008)

    2008 arXiv

  19. [27]

    Clayton J Hutto and Eric Gilbert. 2014. Vader: A parsimonious rule-based model for sentiment analysis of social media text. InEighth international AAAI conference on weblogs and social media

    2014

  20. [28]

    Andreas Jungherr. 2015. Analyzing political communication with digital trace data.Cham, Switzerland: Springer (2015)

    2015

  21. [29]

    1975.Derivation of new readability formulas (automated readability index, fog count and flesch reading ease formula) for navy enlisted personnel

    J Peter Kincaid, Robert P Fishburne Jr, Richard L Rogers, and Brad S Chissom. 1975.Derivation of new readability formulas (automated readability index, fog count and flesch reading ease formula) for navy enlisted personnel. Technical Report. Naval Technical Training Command Mi...

    1975

  22. [30]

    Haewoon Kwak, Changhyun Lee, Hosung Park, and Sue Moon. 2010. What is Twitter, a social network or a news media?. InProceedings of the 19th international conference on World wide web. 591–600

    2010

  23. [31]

    Cliff Lampe, Nicole Ellison, and Charles Steinfield. 2006. A Face (book) in the crowd: Social searching vs. social browsing. InProceedings of the 2006 20th anniversary conference on Computer supported cooperative work. 167–170

    2006

  24. [32]

    Janette Lehmann, Bruno Gonçalves, José J Ramasco, and Ciro Cattuto. 2012. Dynamical classes of collective attention in twitter. InProceedings of the 21st international conference on World Wide Web. 251–260

    2012

  25. [33]

    Ana Levordashka, Sonja Utz, and Renee Ambros. 2016. What’s in a like? Motivations for pressing the like button. In Proceedings of the International AAAI Conference on Web and Social Media, Vol. 10

    2016

  26. [34]

    Mufan Luo and Jeffrey T Hancock. 2020. Self-disclosure and social media: motivations, mechanisms and psychological well-being.Current opinion in psychology31 (2020), 110–115

    2020

  27. [35]

    Sofus Macskassy and Matthew Michelson. 2011. Why do people retweet? anti-homophily wins the day!. InProceedings of the International AAAI Conference on Web and Social Media, Vol. 5

    2011

  28. [36]

    Jalal Mahmud, Jilin Chen, and Jeffrey Nichols. 2013. When will you answer this? estimating response time in twitter. InProceedings of the International AAAI Conference on Web and Social Media, Vol. 7. 697–700

    2013

  29. [37]

    Florian Meier, David Elsweiler, and Max Wilson. 2014. More than liking and bookmarking? towards understanding twitter favouriting behaviour. InProceedings of the international AAAI conference on web and social media, Vol. 8

    2014

  30. [38]

    Gillian Moran, Laurent Muzellec, and Devon Johnson. 2020. Message content features and social media engagement: evidence from the media industry.Journal of Product & Brand Management29, 5 (2020), 533–545

    2020

  31. [39]

    Nasir Naveed, Thomas Gottron, Jérôme Kunegis, and Arifah Che Alhadi. 2011. Bad news travel fast: A content-based analysis of interestingness on twitter. InProceedings of the 3rd international web science conference. 1–7

    2011

  32. [40]

    Chong Oh and Savan Kumar. 2017. How trump won: the role of social media sentiment in political elections. (2017)

    2017

  33. [41]

    Ethan Pancer and Maxwell Poole. 2016. The popularity and virality of political social media: hashtags, mentions, and links predict likes and retweets of 2016 US presidential nominees’ tweets.Social Influence11, 4 (2016), 259–270

    2016

  34. [42]

    Bo Pang and Lillian Lee. 2004. A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts.arXiv preprint cs/0409058(2004)

    2004 arXiv

  35. [43]

    Indraneil Paul, Abhinav Khattar, Ponnurangam Kumaraguru, Manish Gupta, and Shaan Chopra. 2019. Elites tweet? Characterizing the Twitter verified user network. In2019 IEEE 35th International Conference on Data Engineering Workshops (ICDEW). IEEE, 278–285

    2019

  36. [44]

    Lisa Posch, Claudia Wagner, Philipp Singer, and Markus Strohmaier. 2013. Meaning as collective use: predicting semantic hashtag categories on twitter. InProceedings of the 22nd International Conference on World Wide Web. 621–628

    2013

  37. [45]

    Daniel M Romero, Brendan Meeder, and Jon Kleinberg. 2011. Differences in the mechanics of information diffusion across topics: idioms, political hashtags, and complex contagion on twitter. InProceedings of the 20th international conference on World wide web. 695–704

    2011

  38. [46]

    Lauren Scissors, Moira Burke, and Steven Wengrovitz. 2016. What’s in a Like? Attitudes and behaviors around receiving Likes on Facebook. InProceedings of the 19th acm conference on computer-supported cooperative work & social computing. 1501–1510

    2016

  39. [47]

    Kenshin Sekimoto, Yoshifumi Seki, Mitsuo Yoshida, and Kyoji Umemura. 2020. The metrics of keywords to understand the difference between retweet and like in each category. In2020 IEEE/WIC/ACM International Joint Conference on Web Intelligence and Intelligent Agent Technology (W...

    2020

  40. [48]

    Bryan Semaan, Heather Faucett, Scott Robertson, Misa Maruyama, and Sara Douglas. 2015. Navigating imagined audiences: Motivations for participating in the online public sphere. InProceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing. ...

    2015

  41. [49]

    Bryan C Semaan, Scott P Robertson, Sara Douglas, and Misa Maruyama. 2014. Social media supporting political deliberation across multiple public spheres: towards depolarization. InProceedings of the 17th ACM conference on Computer supported cooperative work & social computing. ...

    2014

  42. [50]

    Hamidreza Shahbaznezhad, Rebecca Dolan, and Mona Rashidirad. 2021. The role of social media content format and platform in users’ engagement behavior.Journal of Interactive Marketing53, 1 (2021), 47–65

    2021

  43. [51]

    Naveen Kumar Sharma, Saptarshi Ghosh, Fabricio Benevenuto, Niloy Ganguly, and Krishna Gummadi. 2012. Inferring who-is-who in the Twitter social network.ACM SIGCOMM Computer Communication Review42, 4 (2012), 533–538

    2012

  44. [52]

    Herbert A Simon. 1996. Designing organizations for an information-rich world.International Library of Critical Writings in Economics70 (1996), 187–202

    1996

  45. [53]

    Stefan Stieglitz and Linh Dang-Xuan. 2013. Emotions and information diffusion in social media—sentiment of microblogs and sharing behavior.Journal of management information systems29, 4 (2013), 217–248

    2013

  46. [54]

    Bongwon Suh, Lichan Hong, Peter Pirolli, and Ed H Chi. 2010. Want to be retweeted? large scale analytics on factors impacting retweet in twitter network. In2010 IEEE Second International Conference on Social Computing. IEEE, 177–184

    2010

  47. [55]

    Chenhao Tan, Lillian Lee, and Bo Pang. 2014. The effect of wording on message propagation: Topic-and author- controlled natural experiments on Twitter.arXiv preprint arXiv:1405.1438(2014)

    2014 arXiv

  48. [56]

    Ori Tenenboim and Akiba A Cohen. 2015. What prompts users to click and comment: A longitudinal study of online news.Journalism16, 2 (2015), 198–217

    2015

  49. [57]

    Nava Tintarev and Judith Masthoff. 2007. Effective explanations of recommendations: user-centered design. In Proceedings of the 2007 ACM conference on Recommender systems. 153–156

    2007

  50. [58]

    Gang Wang, Bolun Wang, Tianyi Wang, Ana Nika, Haitao Zheng, and Ben Y Zhao. 2014. Whispers in the dark: analysis of an anonymous social network. InProceedings of the 2014 conference on internet measurement conference. 137–150

    2014

  51. [59]

    Yuan Wang, Jie Liu, Yalou Huang, and Xia Feng. 2016. Using hashtag graph-based topic model to connect semantically- related words without co-occurrence in microblogs.IEEE Transactions on Knowledge and Data Engineering28, 7 (2016), 1919–1933

    2016

  52. [60]

    2014.The marketplace of attention: How audiences take shape in a digital age

    James G Webster. 2014.The marketplace of attention: How audiences take shape in a digital age. Mit Press

    2014

  53. [61]

    Lilian Weng, Alessandro Flammini, Alessandro Vespignani, and Fillipo Menczer. 2012. Competition among memes in a world with limited attention.Scientific reports2, 1 (2012), 335

    2012

  54. [62]

    Lei Yang, Tao Sun, Ming Zhang, and Qiaozhu Mei. 2012. We know what@ you# tag: does the dual role affect hashtag adoption?. InProceedings of the 21st international conference on World Wide Web. 261–270

    2012

  55. [63]

    Sarita Yardi and Danah Boyd. 2010. Dynamic debates: An analysis of group polarization over time on twitter.Bulletin of science, technology & society30, 5 (2010), 316–327

    2010

  56. [64]

    Yulin Yu, Julie Jiang, and Paramveer S. Dhillon. 2024. Characterizing the Structure of Online Conversations Across Reddit.Proc. ACM Hum.-Comput. Interact.8, CSCW2, Article 374 (Nov. 2024), 23 pages. doi:10.1145/3686913

    2024 doi

  57. [65]

    Amy X Zhang and Scott Counts. 2015. Modeling ideology and predicting policy change with social media: Case of same-sex marriage. Inproceedings of the 33rd Annual ACM conference on human factors in computing systems. 2603–2612

    2015

  58. [66]

    Justine Zhang, Cristian Danescu-Niculescu-Mizil, Christina Sauper, and Sean J Taylor. 2018. Characterizing online public discussions through patterns of participant interactions.Proceedings of the ACM on Human-Computer Interaction 2, CSCW (2018), 1–27

    2018

  59. [67]

    Space Force

    Jing Zhang and En Mao. 2016. From online motivations to ad clicks and to behavioral intentions: An empirical study of consumer response to social media advertising.Psychology & Marketing33, 3 (2016), 155–164. Proc. ACM Hum.-Comput. Interact., Vol. 9, No. 7, Article CSCW476. Pu...

    2016

  60. [68]

    102 308 5 (2) Indigenous People of Biafra: Biafra Education Awareness Day (B.E.A.D) lectures will be held at Goldsmiths, University of London

    Spain. 102 308 5 (2) Indigenous People of Biafra: Biafra Education Awareness Day (B.E.A.D) lectures will be held at Goldsmiths, University of London . . . 10 53 1 Table 7. Tweets (paraphrased) illustrating key findings. Received October 2024; revised April 2025; accepted Augus...

    2024

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.