Pith. sign in

REVIEW 4 major objections 6 minor 66 references

Toxic behavior silences online political conversations

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read The paper finds that YouTube political conversations tend to end in a terminal state where toxic and insulting comments are more likely, and interprets this as evidence that hostile behavior silences participants.

desk verdict Solid descriptive finding on end-of-conversation toxicity, but the causal 'silencing' claim outruns the design. read the letter →

arxiv 2412.05741 v1 pith:OEWE6KK4 submitted 2024-12-07 cs.SI cs.CY

classification cs.SIcs.CY
keywords socialmediatoxiccontenthiddenMarkovmodelspiralofsilencepoliticaldeliberationself-censorshipYouTubecomments
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

Online political threads under six major US news channels tend to end not with a reasoned conclusion but with a burst of toxic and insulting comments, and the paper argues this pattern is evidence of self-censorship. Using around 32.5 million YouTube comments from the 2020 US presidential election period, the authors fit a two-state hidden Markov model to each conversation, treating comments as toxic, non-toxic, or an appended end marker. They find a terminal latent state that emits both the end of the thread and a higher probability of toxic and insulting posts relative to the earlier state, for almost every channel and across topic clusters. If right, this means public comment sections understate the diversity of political opinion because toxicity pushes participants out while amplifying hostile voices.

What carries the argument

A two-state hidden Markov model fitted to comment sequences, with three observed symbols: $X_1 = 0$ (an end-of-conversation marker appended to every thread), $X_2 = 1$ (a non-toxic and non-insulting comment), and $X_3 = 2$ (a toxic or insulting comment, scored by the Perspective API toxicity classifier). The model estimates transition probabilities between two latent conversational states and emission probabilities from each state to each symbol. The terminal state $Z_1$ is identified by $P(X_1|Z_1) > 0$ and $P(X_1|Z_2) = 0$; its character is quantified by the relative risks $RR_{X_2} = P(X_2|Z_1)/P(X_2|Z_2)$ and $RR_{X_3} = P(X_3|Z_1)/P(X_3|Z_2)$.

What would settle it

Compare matched threads where the final comment is toxic versus non-toxic: if the terminal state reflects self-censorship, users in toxic-ending threads should show a measurable drop in subsequent commenting on the same channel or video relative to users in non-toxic-ending threads. In addition, place the appended zero at a random position within a subset of sequences; if the terminal-state toxicity disappears, it is an artifact of the appended marker rather than a property of real conversations.

Watch

Extended reading notes

Core claim

The paper's central discovery is a latent silence state at the end of online political conversations. After converting each YouTube thread into a sequence of toxic, non-toxic, and end-of-conversation symbols, a two-state hidden Markov model learns that conversations almost never end in the ordinary-posting state Z2; instead they terminate in state Z1, where non-toxic and non-insulting posts are relatively rare and toxic and insulting posts are relatively common, with the relative risk exceeding 1 for nearly all channels and CNN as the exception. Because the end marker was appended by the authors, this finding asserts that the way a thread closes is systematically toxic, and the authors interpret that as toxicity-driven self-censorship: users exposed to hostile replies stop contributing, leaving the final word to toxic content. The pattern is robust across channel-level and topic-level groupings, with the strongest effects for Fox News and for topics such as police brutality, the Black Lives Matter movement, and COVID-19 vaccination.

Load-bearing premise

The load-bearing premise is that appending an end-of-conversation symbol to every comment sequence and fitting a two-state hidden Markov model produces a hidden state that genuinely represents self-censorship, rather than simply the structural fact that every thread has an endpoint.

Editorial extensions

If this is right

  • Platforms that want to sustain civil discussion should treat a rise in toxic replies as a warning that a thread is approaching a terminal state dominated by hostile content.
  • If the terminal state reflects self-censorship, then visible comment sections understate the diversity of political opinion, because moderate voices drop out while toxic ones have the last word.
  • The U-shaped pattern across channel bias ratings suggests polarization is self-reinforcing: channels further from the political center show higher terminal-state toxicity, which may push moderate users away.
  • Because the effect appears across topic clusters rather than only channel groupings, the silencing dynamic is likely tied to the content being discussed, not just to the outlet hosting it.

Reading between the lines

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

  • An extension would be to test whether individual users who receive toxic replies subsequently reduce their commenting activity in the same channel; that would move the terminal-state pattern from an aggregate signature to a per-user behavioral claim.
  • The appended zero conflates 'the conversation ended' with 'the participants were silenced'; a control that appends the end marker at a random position in the sequence would show whether the terminal state's toxicity is a real conversational pattern or a modeling artifact.
  • The same two-state hidden Markov model could be applied to Reddit or Twitter threads to see whether terminal-state toxicity is a general property of online deliberation or specific to YouTube's reply structure.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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. This paper studies toxic and insulting behavior in YouTube conversations under six US news channels during the 2020 US election period. The authors classify comments with Google's Perspective API, define conversations as top-level comments plus replies received within 10 days, append an artificial end-of-conversation symbol X1=0 to every sequence, and fit a two-state hidden Markov model with observations X1 (end), X2 (non-toxic/non-insulting), and X3 (toxic/insulting). They report that one state, Z1, emits the end marker with positive probability while the other does not, and that the relative risk of toxic/insulting comments in Z1 versus Z2 exceeds 1 for most channels. The authors interpret Z1 as a latent terminal state consistent with toxicity-driven silence and conclude that toxic behavior silences users. They also repeat the analysis with topic-based video clusters and find similar relative-risk patterns.

Significance. The paper brings a very large dataset (32.5M comments) to an important question about self-censorship in online political deliberation. The descriptive finding that toxic and insulting comments are relatively more probable near the end of conversations is potentially interesting, and the topic-cluster robustness check is a plus. However, the interpretive leap from 'the observed comment stream ended' to 'users were silenced by toxicity' is not supported by the design. The end marker is an artifact of the data-collection procedure and platform structure, not an observation of user disengagement. Without an independent measure of silencing (e.g., a user who stops replying after receiving a toxic comment, or a conversation that ends earlier than expected given its activity), the central claim remains a restatement of the correlation between conversation position and toxicity. As a descriptive study of toxicity at the end of threads, the paper might be salvageable, but the current framing overstates the evidence.

major comments (4)
  1. [SI S4; Section 2.4; Abstract] The operationalization of 'silence' as the appended end-of-conversation symbol X1=0 has no construct validity for self-censorship. The authors state in SI S4 that they 'appended a zero (X1=0) to the end of every sequence/conversation to indicate its conclusion.' Since every conversation is artificially terminated by the data-collection window (10 days) and by YouTube's first-level-reply structure, the terminal state Z1, which is identified by its emission of X1, cannot be interpreted as a state of user silence without external validation. The abstract's claim that the authors 'observe patterns of self-censorship' is therefore not supported by the measurements.
  2. [Section 2.4, Eqs. (1)-(2)] The relative risks RRX2 and RRX3 are computed from the same fitted emission probabilities that define the latent states, so the finding RRX3>1 is a property of the fitted model rather than an independent test of a silencing mechanism. The paper does not report uncertainty estimates or confidence intervals for these ratios across the repeated HMM fits described in SI Table 4; adding such intervals would be necessary to assess whether the >1 pattern is statistically stable.
  3. [Section 3 (Discussion)] The discussion goes beyond the observational evidence by describing 'a circular causality between toxicity and disengagement.' The data cannot distinguish between toxicity causing users to stop commenting, users ceasing to comment for unrelated reasons and the conversation then turning toxic, or platform and moderator effects truncating threads. A concrete alternative explanation is that YouTube only permits first-level replies (as the authors note in Section 2.1) and the 10-day cutoff, so the 'end' marker is independent of user silence.
  4. [Abstract; Section 2.4] The claim that state Z1 is 'characterized by reduced user activity' (abstract) is not measured. The model has no observation of user activity beyond the presence or absence of comments; the appended X1 symbol is not a measure of activity level. This characterization should be removed or replaced with a user-level analysis (e.g., reply rates or return probabilities) to support the silence interpretation.
minor comments (6)
  1. [Section 2.4] There is a typographical error: 'P (X1|Z2 = 0)' should be 'P (X1|Z2) = 0' in the sentence about the four-state model; the same notation appears again in the paragraph describing Figure 4.
  2. [Section 2.1] The sentence 'It is possible to use these mentions to reconstruct sub-threads, but doing so is beyond the scope of this research. Using these mentions, it is therefore possible to reconstruct sub-threads; however, this is beyond the scope of this work.' is repeated with slight variation; the duplication should be removed.
  3. [Section 2.4 / Figure 4] The relative risk values in Figure 4 are described only in prose and without error bars; please provide uncertainty or bootstrap intervals, or state that they are omitted.
  4. [Table 3] The bias scores use a decimal comma (e.g., '-2,40' for ABC News); the manuscript should use decimal points consistently.
  5. [SI S4] Reference [66] (a protein-protein interactions overview) is used as the source for the HMM description; a standard HMM reference (e.g., Rabiner 1989) would be more appropriate.
  6. [Section 2.5] The paper says hSBM 'automatically identifies the number of topics and hierarchical levels,' but then it says 'we work with the clusters at the fourth hierarchical level.' Please clarify whether the fourth level is chosen by the model or by the authors.

Circularity Check

2 steps flagged · score 6.0 of 10

The 'silence' state is the appended end-of-conversation marker relabeled as self-censorship, and the reported relative risks are ratios of the same fitted emission probabilities, making the central claim substantially circular.

  1. self definitional [SI S4 'Hidden Markov Model fit details'; Section 2.4; abstract]
    "Additionally, we appended a zero (X1 = 0) to the end of every sequence/conversation to indicate its conclusion, ensuring that all conversations end with a 0. ... Notably, P (X1|Z2) = 0 for all channels ... Recall that X1 denotes the end of a conversation; thus, P (X1|Z2) = 0 indicates that conversations do not conclude in state Z2. Combined with the fact that P (X1|Z1) > 0 across all channels, we can interpret state Z1 as the likely terminal state for conversations."

    Every observed conversation is forced, by construction, to end with the artificial symbol 0. The model's state Z1 is identified as 'terminal' precisely because it emits this symbol (P(X1|Z1) > 0). The paper then labels this state 'silence' and the abstract describes it as 'reduced user activity' consistent with self-censorship. That is a restatement of the input encoding: 'lack of activity' was defined as the appended end marker. No independent observation of user disengagement, deleted comments, or suppressed participation is used, so the conclusion that conversations end in a 'silence' state is true by definition rather than discovered from the data.

  2. fitted input called prediction [Section 2.4, Eqs. (1)-(2) and Figure 4]
    "RRX3 = P (X3|Z1) / P (X3|Z2) ... We found that RRX3 > 1 for nearly all channels ... This finding indicates that when a conversation is in state Z1, toxic and insulting content is more likely to appear—suggesting that conversations approaching their end are more likely to contain toxic or insulting posts."

    The relative risks RRX2 and RRX3 are defined directly as ratios of the emission probabilities P(Xj|Zi) produced by the same HMM that supplies the Z1/Z2 labels. Reporting RRX3 > 1 as evidence that 'toxicity silences users' is an in-sample description of the fitted parameters, not a prediction tested on held-out data or validated against an independent measure of silencing. The causal claim in the abstract goes beyond what the ratio can show: the ratio only states that, in the fitted model, the state labeled 'terminal' has a higher fitted emission probability for toxic/insulting posts. The finding is therefore the fitted input reported as the output.

full rationale

The descriptive statistics, time-series correlations, conditional toxicity distributions, and the HMM estimation pipeline are self-contained empirical work; I found no load-bearing self-citation or imported uniqueness theorem. However, the paper's central contribution—the 'toxicity-driven silence' state—is built on the appended X1=0 marker in SI S4. Because every conversation is forced to end in this symbol, identifying a state that emits it and calling that state 'silence' / 'self-censorship' makes the abstract's central claim substantially circular: the 'silence' is defined as the end-of-conversation marker, and the 'reduced user activity' attributed to the silence state is just the imposed absence of further comments. The relative risks in Eqs. (1)-(2) are ratios of the same fitted emission probabilities used to name the states, so the reported pattern that toxicity concentrates at the end is an in-sample property of the fitted model rather than an independent prediction. The topic-clustering reanalysis (Section 2.5) uses the same appended-zero encoding, so it does not break the circularity. Score 6 reflects a central claim that partly reduces to its own construction, while leaving room for the paper's other descriptive results to stand independently.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

The central claim rests on several author-chosen thresholds and a modeling assumption that appending an end-of-conversation marker captures self-censorship. The latent 'silence' state is an interpretive label applied to a fitted HMM state, with no external validation.

free parameters (4)
  • Toxicity/insult classification threshold = 0.6
    Used to classify comments as toxic/insulting; chosen by following prior studies, with robustness checks mentioned but not shown.
  • Number of hidden states = 2
    Fixed at 2 although the 4-state model had better log-likelihood (SI S4); the choice is justified by overfitting concerns but was made post hoc.
  • Conversation reply window = 10 days
    Replies within 10 days of the top-level comment are included; no sensitivity analysis is reported for this window.
  • hSBM hierarchical level for topic clusters = 4
    The fourth hierarchical level is used for topic clusters; no sensitivity analysis is reported for this choice.
assumptions (5)
  • domain assumption Perspective API scores validly measure the constructs of toxicity and insult in this dataset.
    The paper relies on Perspective API despite acknowledging known biases (references 38-39); no validation on this specific dataset is provided.
  • ad hoc to paper Appending an end-of-conversation marker is a valid representation of conversation termination and silence.
    In SI S4, X1=0 is appended to every sequence to indicate conclusion; interpreting this as self-censorship is an unvalidated modeling assumption.
  • standard math HMM Markov property and stationarity hold for comment sequences.
    HMM assumes the future state depends only on the current state; no evidence is given that comment sequences satisfy this.
  • domain assumption All replies target the top-level comment even when mentions suggest otherwise.
    Section 2.1 states this simplification; it ignores sub-threads reconstructed via mentions.
  • domain assumption Conversations are fully captured within 10 days.
    Replies after 10 days are excluded, which may truncate long-running discussions.
invented entities (1)
  • Latent state Z1 interpreted as 'toxicity-driven silence'
    purpose: To explain why toxic comments are more likely near the end of conversations.
    The latent state is inferred from the data via HMM; there is no independent behavioral measure of silence or self-censorship, and no falsifiable out-of-sample prediction is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Toxic behavior silences online political conversations." pith.science (2026). https://pith.science/paper/OEWE6KK4

@misc{pith2026241205741,
  author       = {Pith},
  title        = {Pith review of: Toxic behavior silences online political conversations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OEWE6KK4}},
  note         = {Machine review of arXiv:2412.05741}
}
read the original abstract

Quantifying how individuals react to social influence is crucial for tackling collective political behavior online. While many studies of opinion in public forums focus on social feedback, they often overlook the potential for human interactions to result in self-censorship. Here, we investigate political deliberation in online spaces by exploring the hypothesis that individuals may refrain from expressing minority opinions publicly due to being exposed to toxic behavior. Analyzing conversations under YouTube videos from six prominent US news outlets around the 2020 US presidential elections, we observe patterns of self-censorship signaling the influence of peer toxicity on users' behavior. Using hidden Markov models, we identify a latent state consistent with toxicity-driven silence. Such state is characterized by reduced user activity and a higher likelihood of posting toxic content, indicating an environment where extreme and antisocial behaviors thrive. Our findings offer insights into the intricacies of online political deliberation and emphasize the importance of considering self-censorship dynamics to properly characterize ideological polarization in digital spheres.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 44 canonical work pages

  1. [1]

    Science 381(6656), 392–398 (2023) https://doi.org/10.1126/science.ade7138

    Gonz´ alez-Bail´ on, S., Lazer, D., Barber´ a, P., Zhang, M., Allcott, H., Brown, T., Crespo-Tenorio, A., Freelon, D., Gentzkow, M., Guess, A.,et al.: Asymmetric ideological segregation in exposure to political news on facebook. Science 381(6656), 392–398 (2023) https://doi.org/10.1126/science.ade7138

  2. [2]

    Guess, A., Malhotra, N., Pan, J., Barber´ a, P., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M.,et al.: How do social media feed algorithms affect attitudes and behavior in an election campaign? Science 381(6656), 398–404 (2023) https://doi.org/10.1126/science.abp9364

  3. [3]

    Philosophy & Social Criticism 27(4), 21–39 (2001) https://doi.org/10.1177/019145370102700402

    Gimmler, A.: Deliberative democracy, the public sphere and the internet. Philosophy & Social Criticism 27(4), 21–39 (2001) https://doi.org/10.1177/019145370102700402

  4. [4]

    How does neural activity encode spontaneous motor behavior in zebrafish larvae ?

    Bressers, B., Hume, J.: Message boards, public discourse, and historical meaning: An online community reacts to September 11. American Journalism 29(4), 9–33 (2012) https://doi.org/10.1080/08821127. 2012.10677846

  5. [5]

    Nature human behaviour 7(1), 74–101 (2023) https://doi.org/10.1038/s41562-022-01460-1

    Lorenz-Spreen, P., Oswald, L., Lewandowsky, S., Hertwig, R.: A systematic review of worldwide causal and correlational evidence on digital media and democracy. Nature human behaviour 7(1), 74–101 (2023) https://doi.org/10.1038/s41562-022-01460-1

  6. [6]

    CNN (26 July 2020)

    Asmelash, L.: How Black Lives Matter went from a hashtag to a global rallying cry. CNN (26 July 2020). https://edition.cnn.com/2020/07/26/us/black-lives-matter-explainer-trnd/index.html

  7. [7]

    New York Times (1 January 2021)

    Frenkel, S.: The storming of Capitol Hill was organized on social media. New York Times (1 January 2021). https://www.nytimes.com/2021/01/06/us/politics/protesters-storm-capitol-hill-building.html

  8. [8]

    Yasseri, T., Menczer, F.: Can crowdsourcing rescue the social marketplace of ideas? Communications of the ACM 66(9), 42–45 (2023) https://doi.org/10.1145/3578645

Show all 66 references
  1. [9]

    Algorithm Watch (26 May 2020)

    Stark, B., Stegmann, D., Magin, M., J¨ urgens, P.: Are algorithms a threat to democracy? The rise of intermediaries: A challenge for public discourse. Algorithm Watch (26 May 2020). https://algorithmwatch.org/de/wp-content/uploads/2020/05/ Governing-Platforms-communications-st...

  2. [10]

    Proceedings of the National Academy of Sciences 118(27), 2025764118 (2021) https://doi.org/10.1073/pnas.2025764118 13

    Bak-Coleman, J., Alfano, M., Barfuss, W., Bergstrom, C., Centeno, M., Couzin, I., Donges, J., Galesic, M., Gersick, A., Jacquet, J.,et al.: Stewardship of global collective behavior. Proceedings of the National Academy of Sciences 118(27), 2025764118 (2021) https://doi.org/10....

  3. [11]

    International Public Management Journal 20(3), 409–441 (2017) https://doi.org/10.1080/10967494.2016.1269859

    Porumbescu, G.: Not all bad news after all? Exploring the relationship between citizens’ use of online mass media for government information and trust in government. International Public Management Journal 20(3), 409–441 (2017) https://doi.org/10.1080/10967494.2016.1269859

  4. [12]

    Proceed- ings of the National Academy of Sciences 118(14), 2013464118 (2021) https://doi.org/10.1073/pnas

    Guess, A., Barber´ a, P., Munzert, S., Yang, J.: The consequences of online partisan media. Proceed- ings of the National Academy of Sciences 118(14), 2013464118 (2021) https://doi.org/10.1073/pnas. 2013464118

  5. [13]

    Information, Communication & Society 25(13), 1934–1951 (2022) https://doi.org/10.1080/1369118X.2021.1907435

    Schumann, S., Thomas, F., Ehrke, F., Bertlich, T., Dupont, J.: Maintenance or change? Examining the reinforcing spiral between social media news use and populist attitudes. Information, Communication & Society 25(13), 1934–1951 (2022) https://doi.org/10.1080/1369118X.2021.1907435

  6. [14]

    Systematic review

    Casta˜ o-Pulgar ´ ın, S., Su´ arez-Betancur, N., Tilano-Vega, L., Herrera-L´ pez, H.: Internet, social media and online hate speech. Systematic review. Aggression and Violent Behavior 58, 101608 (2021) https: //doi.org/10.1016/j.avb.2021.101608

  7. [15]

    Jour- nal of Information Technology & Politics 17(1), 66–78 (2020) https://doi.org/10.1080/19331681.2019

    Vidgen, B., Yasseri, T.: Detecting weak and strong islamophobic hate speech on social media. Jour- nal of Information Technology & Politics 17(1), 66–78 (2020) https://doi.org/10.1080/19331681.2019. 1702607

  8. [16]

    Proceedings of the National Academy of Sciences 115(37), 9216–9221 (2018) https://doi.org/10.1073/ pnas.1804840115

    Bail, C., Argyle, L., Brown, T., Bumpus, J., Chen, H., Hunzaker, F., Lee, J., Mann, M., Merhout, F., Volfovsky, A.: Exposure to opposing views on social media can increase political polarization. Proceedings of the National Academy of Sciences 115(37), 9216–9221 (2018) https:/...

  9. [17]

    Political Communication 38(1-2), 98–139 (2021) https://doi.org/10.1080/10584609.2020.1785067

    Yarchi, M., Baden, C., Kligler-Vilenchik, N.: Political polarization on the digital sphere: A cross- platform, over-time analysis of interactional, positional, and affective polarization on social media. Political Communication 38(1-2), 98–139 (2021) https://doi.org/10.1080/10...

  10. [18]

    Nature 489(7415), 295–298 (2012) https: //doi.org/10.1038/nature11421

    Bond, R., Fariss, C., Jones, J., Kramer, A., Marlow, C., Settle, J., Fowler, J.: A 61-million-person experiment in social influence and political mobilization. Nature 489(7415), 295–298 (2012) https: //doi.org/10.1038/nature11421

  11. [19]

    Communication Research 47(7), 947–966 (2020) https://doi.org/10.1177/0093650218808186

    Boulianne, S.: Twenty years of digital media effects on civic and political participation. Communication Research 47(7), 947–966 (2020) https://doi.org/10.1177/0093650218808186

  12. [20]

    Princeton University Press, Princeton (2016)

    Margetts, H., John, P., Hale, S., Yasseri, T.: Political Turbulence: How Social Media Shape Collective Action. Princeton University Press, Princeton (2016)

  13. [21]

    New Media & Society 10(3), 455–474 (2008) https://doi.org/10.1177/1461444807085327

    Beaudoin, C.: The internet’s impact on international knowledge. New Media & Society 10(3), 455–474 (2008) https://doi.org/10.1177/1461444807085327

  14. [22]

    Journalism & Mass Communication Quarterly 95(4), 1103–1127 (2018) https://doi.org/10.1177/1077699017753149

    Park, C., Kaye, B.: News engagement on social media and democratic citizenship: Direct and moderat- ing roles of curatorial news use in political involvement. Journalism & Mass Communication Quarterly 95(4), 1103–1127 (2018) https://doi.org/10.1177/1077699017753149

  15. [23]

    Chinese Journal of Communication 9(4), 331–347 (2016) https://doi.org/10.1080/17544750.2016

    Skoric, M., Zhu, Q., Pang, N.: Social media, political expression, and participation in confucian asia. Chinese Journal of Communication 9(4), 331–347 (2016) https://doi.org/10.1080/17544750.2016. 1143378

  16. [24]

    In: 2023 5th International Conference on Artificial Intelligence and Computer Applications (ICAICA), pp

    Dai, W., Tao, J., Yan, X., Feng, Z., Chen, J.: Addressing unintended bias in toxicity detection: An lstm and attention-based approach. In: 2023 5th International Conference on Artificial Intelligence and Computer Applications (ICAICA), pp. 375–379 (2023). https://doi.org/10.11...

  17. [25]

    Public Discourse in Online Environments (2023) https://doi.org/10.31235/osf.io/evdxy

    Oswald, L.: Effects of preemptive empathy interventions on reply toxicity among highly active social media users. Public Discourse in Online Environments (2023) https://doi.org/10.31235/osf.io/evdxy

  18. [26]

    Pew Research Center: Teens and cyberbullying 2022 (December 2022)

  19. [27]

    Pew Research Center: The state of online harassment (January 2021) 14

  20. [28]

    In: 2021 IEEE Symposium on Security and Privacy (SP), pp

    Thomas, K., Akhawe, D., Bailey, M., Boneh, D., Bursztein, E., Consolvo, S., Dell, N., Durumeric, Z., Kelley, P., Kumar, D., et al.: Sok: Hate, harassment, and the changing landscape of online abuse. In: 2021 IEEE Symposium on Security and Privacy (SP), pp. 247–267 (2021). IEEE

  21. [29]

    Princeton University Press, Princeton (2022)

    Bail, C.A.: Breaking the Social Media Prism: How to Make Our Platforms Less Polarizing. Princeton University Press, Princeton (2022)

  22. [30]

    nasty effect:

    Anderson, A.A., Brossard, D., Scheufele, D.A., Xenos, M.A., Ladwig, P.: The “nasty effect:” online incivility and risk perceptions of emerging technologies. Journal of computer-mediated communication 19(3), 373–387 (2014) https://doi.org/10.1111/jcc4.12009

  23. [31]

    Journal of Communication 71(6), 922–946 (2021) https://doi.org/10.1093/joc/jqab034

    Kim, J., Guess, A., Nyhan, B., Reifler, J.: The distorting prism of social media: How self-selection and exposure to incivility fuel online comment toxicity. Journal of Communication 71(6), 922–946 (2021) https://doi.org/10.1093/joc/jqab034

  24. [32]

    American political science review 16(1), 1–18 (2022) https://doi.org/ 10.1017/S0003055421000885

    Bor, A., Petersen, M.B.: The psychology of online political hostility: A comprehensive, cross-national test of the mismatch hypothesis. American political science review 16(1), 1–18 (2022) https://doi.org/ 10.1017/S0003055421000885

  25. [33]

    Nature 628(8008), 582–589 (2024) https://doi.org/10.1038/s41586-024-07229-y

    Avalle, M., Di Marco, N., Etta, G., Sangiorgio, E., Alipour, S., Bonetti, A., Alvisi, L., Scala, A., Baronchelli, A., Cinelli, M., et al.: Persistent interaction patterns across social media platforms and over time. Nature 628(8008), 582–589 (2024) https://doi.org/10.1038/s415...

  26. [34]

    In: Proceedings of the 26th International Conference on World Wide Web, pp

    Wulczyn, E., Thain, N., Dixon, L.: Ex machina: Personal attacks seen at scale. In: Proceedings of the 26th International Conference on World Wide Web, pp. 1391–1399 (2017). https://doi.org/10.1145/ 3038912.3052591

  27. [35]

    In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp

    Lees, A., Tran, V., Tay, Y., Sorensen, J., Gupta, J., Metzler, D., Vasserman, L.: A new generation of perspective api: Efficient multilingual character-level transformers. In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pp. 3197–3207 (...

  28. [36]

    https://www.perspectiveapi.com/how-it-works/

    Perspective API - How it works. https://www.perspectiveapi.com/how-it-works/. Accessed: 21 October 2024

  29. [37]

    Plos one 15(12), 0243300 (2020) https://doi.org/10.1371/journal.pone.0243300

    Vidgen, B., Derczynski, L.: Directions in abusive language training data, a systematic review: Garbage in, garbage out. Plos one 15(12), 0243300 (2020) https://doi.org/10.1371/journal.pone.0243300

  30. [38]

    Medium (9 March 2018)

    Jigsaw: Unintended Bias and Identity Terms. Medium (9 March 2018). https://medium.com/jigsaw/ unintended-bias-and-names-of-frequently-targeted-groups-8e0b81f80a23

  31. [39]

    Preprint at https://arxiv.org/abs/2312.12651 (2023)

    Nogara, G., Pierri, F., Cresci, S., Luceri, L., T¨ ornberg, P., Giordano, S.: Toxic Bias: Perspective API misreads German as more toxic. Preprint at https://arxiv.org/abs/2312.12651 (2023)

  32. [40]

    Standford University

    Ramage, D.: CS229: Hidden Markov Models Fundamentals. Standford University. https://cs229. stanford.edu/section/cs229-hmm.pdf (2007)

  33. [41]

    In: Proceedings of the ACL-02 Workshop on Effective Tools and Methodologies for Teaching Natural Language Processing and Computational Linguistics, pp

    Eisner, J.: An interactive spreadsheet for teaching the forward-backward algorithm. In: Proceedings of the ACL-02 Workshop on Effective Tools and Methodologies for Teaching Natural Language Processing and Computational Linguistics, pp. 10–18 (2002). https://aclanthology.org/W0...

  34. [42]

    Department of Computer Science, Brown University

    Istrail, S.: HMM: The Learning Problem. Department of Computer Science, Brown University. https: //cs.brown.edu/courses/csci1820/resources/HMMs The Learning Problem slides.pdf (2020)

  35. [43]

    Proceedings of the National Academy of Sciences 118(32), 2101967118 (2021) https://doi.org/10.1073/pnas.2101967118

    Hosseinmardi, H., Ghasemian, A., Clauset, A., Mobius, M., Rothschild, D., Watts, D.: Examining the consumption of radical content on youtube. Proceedings of the National Academy of Sciences 118(32), 2101967118 (2021) https://doi.org/10.1073/pnas.2101967118

  36. [44]

    Pro- ceedings of the National Academy of Sciences 120(50), 2213020120 (2023) https://doi.org/10.1073/ pnas.221302012 15

    Haroon, M., Wojcieszak, M., Chhabra, A., Liu, X., Mohapatra, P., Shafiq, Z.: Auditing youtube’s recommendation system for ideologically congenial, extreme, and problematic recommendations. Pro- ceedings of the National Academy of Sciences 120(50), 2213020120 (2023) https://doi...

  37. [45]

    AllSides (1 February 2019)

    Mastrine, J.: Introducing the AllSides Media Bias Chart. AllSides (1 February 2019). https://www. allsides.com/blog/introducing-allsides-media-bias-chart

  38. [46]

    https://developers.perspectiveapi.com/s/ about-the-api-attributes-and-languages?language=en US

    Perspective API - Attributes & Languages. https://developers.perspectiveapi.com/s/ about-the-api-attributes-and-languages?language=en US. Accessed: 21 October 2024

  39. [47]

    https://developers.perspectiveapi.com/s/about-the-api-score?language=en US

    Perspective API - Score. https://developers.perspectiveapi.com/s/about-the-api-score?language=en US. Accessed: 21 October 2024

  40. [48]

    In: Leskovec, J., et al (ed.) Proceedings of the Web Conference 2021, pp

    Saveski, M., Roy, B., Roy, D.: The structure of toxic conversations on twitter. In: Leskovec, J., et al (ed.) Proceedings of the Web Conference 2021, pp. 1086–1097. Association for Computing Machinery, New YorkNYUnited States (2021). https://doi.org/10.1145/3442381.344986

  41. [49]

    Neurocomputing 490, 312–318 (2022) https://doi.org/10.1016/j.neucom.2021.11

    Sheth, A., Shalin, V.L., Kursuncu, U.: Defining and detecting toxicity on social media: context and knowledge are key. Neurocomputing 490, 312–318 (2022) https://doi.org/10.1016/j.neucom.2021.11. 095

  42. [50]

    In: Proceedings of the International AAAI Conference on Web and Social Media, pp

    Rajadesingan, A., Resnick, P., Budak, C.: Quick, community-specific learning: How distinctive toxicity norms are maintained in political subreddits. In: Proceedings of the International AAAI Conference on Web and Social Media, pp. 557–568 (2020). https://doi.org/10.1609/icwsm....

  43. [51]

    Social Science Computer Review 38(5), 584–599 (2020) https://doi.org/10.1177/ 089443931881424

    Hopp, T., Vargo, C.J., Dixon, L., Thain, N.: Correlating self-report and trace data measures of incivility: A proof of concept. Social Science Computer Review 38(5), 584–599 (2020) https://doi.org/10.1177/ 089443931881424

  44. [52]

    AllSides (3 January 2023)

    AllSides Staff: Introducing the AllSides Media Bias Meter ™. AllSides (3 January 2023). https://www. allsides.com/blog/introducing-allsides-media-bias-meter

  45. [53]

    Proceedings of the National Academy of Sciences114(28), 7313–7318 (2017) https://doi.org/10.1073/pnas.1618923114

    Brady, W., Wills, J., Jost, J., Tucker, J., Van Bavel, J.: Emotion shapes the diffusion of moralized content in social networks. Proceedings of the National Academy of Sciences114(28), 7313–7318 (2017) https://doi.org/10.1073/pnas.1618923114

  46. [54]

    CNN (28 September 2020)

    McLaughlin, E.: Anger erupts in American cities after charging decision in Breonna Taylor case. CNN (28 September 2020). https://edition.cnn.com/2020/09/28/us/weekend-protests-breonna-taylor/ index.html

  47. [55]

    BBC (22 April 2021)

    BBC Staff: George Floyd: Timeline of black deaths and protests. BBC (22 April 2021). https://www. bbc.com/news/world-us-canada-52905408

  48. [56]

    COVID-19 Death Toll Reaches 500,000

    Huang, P.: ’A Loss To The Whole Society’: U.S. COVID-19 Death Toll Reaches 500,000. NPR (22 February 2021). https://www.npr.org/sections/health-shots/2021/02/22/969494791/ a-loss-to-the-whole-society-u-s-covid-19-death-toll-reaches-500-000

  49. [57]

    Journal of Agricultural, Bio- logical and Environmental Statistics 22, 270–293 (2017) https://doi.org/10.1007/s13253-017-0283-8

    Pohle, J., Langrock, R., Van Beest, F.M., Schmidt, N.M.: Selecting the number of states in hidden markov models: pragmatic solutions illustrated using animal movement. Journal of Agricultural, Bio- logical and Environmental Statistics 22, 270–293 (2017) https://doi.org/10.1007...

  50. [58]

    Information Systems 94, 101582 (2020) https://doi.org/10.1016/j.is.2020.101582

    Vayansky, I., Kumar, S.: A review of topic modeling methods. Information Systems 94, 101582 (2020) https://doi.org/10.1016/j.is.2020.101582

  51. [59]

    Science Advances 4(7), 1360 (2018) https://doi.org/10.1126/sciadv.aaq1360

    Gerlach, M., Peixoto, T.P., Altmann, E.G.: A network approach to topic models. Science Advances 4(7), 1360 (2018) https://doi.org/10.1126/sciadv.aaq1360

  52. [60]

    The Journal of Mathematical Sociology 46(1), 80–97 (2022) https://doi.org/10.1080/0022250X.2020

    Blex, C., Yasseri, T.: Positive algorithmic bias cannot stop fragmentation in homophilic networks. The Journal of Mathematical Sociology 46(1), 80–97 (2022) https://doi.org/10.1080/0022250X.2020. 1818078

  53. [61]

    Nature Communications 15(1), 9560 (2024) https://doi.org/10.1038/s41467-024-53868-0

    Falkenberg, M., Zollo, F., Quattrociocchi, W., Pfeffer, J., Baronchelli, A.: Patterns of partisan toxicity and engagement reveal the common structure of online political communication across countries. Nature Communications 15(1), 9560 (2024) https://doi.org/10.1038/s41467-024-53868-0

  54. [62]

    American 16 Journal of Political Science 61(2), 490–503 (2017) https://doi.org/10.1111/ajps.12274

    Beauchamp, N.: Predicting and interpolating state-level polls using twitter textual data. American 16 Journal of Political Science 61(2), 490–503 (2017) https://doi.org/10.1111/ajps.12274

  55. [63]

    JMIR public health and surveillance 6(3), 17103 (2020) https://doi.org/10.2196/ 17103

    Nguyen, T.T., Adams, N., Huang, D., Glymour, M.M., Allen, A.M., Nguyen, Q.C.: The association between state-level racial attitudes assessed from twitter data and adverse birth outcomes: Obser- vational study. JMIR public health and surveillance 6(3), 17103 (2020) https://doi.o...

  56. [64]

    https://simonmunzert.github.io/meof/material/schulz-et-al-ideology-twitter-apsa.pdf

    Schulz, W.S., Guess, A.M., Barber´ a, P., Munzert, S., Gottlieb, A., Hughes, A., Remy, E., Shah, S., Smith, A.: (Mis)representing ideology on twitter: How social influence shapes online political expression (2020). https://simonmunzert.github.io/meof/material/schulz-et-al-ideo...

  57. [65]

    New Journal of Physics 14(1), 013055 (2012) https://doi.org/10.1088/1367-2630/14/1/ 013055

    Jo, H., Karsai, M., Kert´ esz, J., Kaski, K.: Circadian pattern and burstiness in mobile phone com- munication. New Journal of Physics 14(1), 013055 (2012) https://doi.org/10.1088/1367-2630/14/1/ 013055

  58. [66]

    In: Ranganathan, S., Gribskov, M., Nakai, K., Sch¨ onbach, C

    Folador, E., Tiwari, S., Da Paz Barbosa, C., Jamal, S., Da Costa Schulze, M., Barh, D., Azevedo, V.: Protein-protein interactions: an overview. In: Ranganathan, S., Gribskov, M., Nakai, K., Sch¨ onbach, C. (eds.) Encyclopedia of Bioinformatics and Computational Biology, pp. 82...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.