Pith. sign in

REVIEW 3 major objections 5 minor 68 references

Algorithmic Collective Action with Two Collectives

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A collective that can steer a language model alone can lose up to 75% of its efficacy when a second collective acts simultaneously, even when they target different classes.

desk verdict The recommender half is solid and the framework is a useful extension, but the LM headline 75% drop is undermined by the authors' own admission that '100' and '101' are the same token to the model. read the letter →

arxiv 2505.00195 v1 pith:3V6ABWKT submitted 2025-04-30 cs.CY cs.GTcs.LG

Aditya Karan , Nicholas Vincent , Karrie Karahalios , Hari Sundaram This is my paper · ORCID
classification cs.CYcs.GTcs.LG
keywords AlgorithmicCollectiveActionSocialComputingDataCampaignsRecommenderSystemsLanguageModelsEfficacyConstructivenessScoreMulti-collectiveinteraction
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

The paper builds the first framework for studying what happens when two or more organized groups ('collectives') each try to steer the same data-driven algorithm toward their own goal at the same time. It argues that these groups are likely to differ in objectives, strategies, size, and internal similarity, and that such differences produce unintended interactions. Across experiments with a resume-classifying language model and a movie recommender, the paper finds that a group that can achieve near-perfect efficacy alone can lose as much as 75% of that efficacy when a second group acts simultaneously, even when the two groups target different classes or items. In recommender systems, group size matters more than internal homogeneity, and neither fully similar nor fully dissimilar groups are the most effective. The work matters because it suggests that collective-action organizers cannot predict success from their own group alone, and that platform developers need to account for competing data campaigns.

What carries the argument

The central mechanism is the two-collective experimental framework built on (1) a signal-planting strategy for language models—inserting a particular character every 20 words and relabeling those resumes to a target class—and (2) a cluster-based collective construction for recommender systems, where users are clustered by matrix-factorization vectors and members are sampled with propensity $p$ from a seed cluster to control homogeneity. The paper also introduces the constructiveness score $\mathrm{CT}(c_i, c_j)$, which measures how much collective $j$ helps or hurts collective $i$'s objective compared with $i$ acting alone; this score is the quantitative lens through which interference and synergy are detected.

What would settle it

Run the same two-collective language model experiment with a tokenizer verified to assign distinct token IDs to the two planted characters (e.g., '100' and '101'); if the ~75% efficacy drop in the '100' versus '101' condition does not occur, then that specific interference is an artifact of tokenizer conflation rather than a general property of multi-collective action.

Watch

Extended reading notes

Core claim

The central claim is that simultaneous collective action by two distinct groups produces substantial, often unintended, interactions: a 'targeted promoter' or 'targeted demoter' group that is highly effective in isolation can be sharply hampered by the presence of a second group with its own objective. In the language-model experiments, two collectives each plant a distinct character signal in resumes and relabel their training data; when both act, a collective that reached near-100% top-one accuracy alone can fall to about 25%, depending on whether the model's tokenizer conflates the two signals and on the relative sizes of the groups. In the recommender experiments, the paper defines a 'constructiveness score' that compares a group's hit-ratio gain when acting with a partner versus alone, and finds that two promoting groups help each other, two demoting groups help each other, but a promoter and a demoter actively interfere, even when their seed clusters are maximally far apart. It also reports that collective size has first-order influence on efficacy, while homogeneity—controlled by a sampling propensity from a seed cluster—plays a secondary role, with the most effective groups often not fully homogeneous.

Load-bearing premise

In the recommender experiments, the authors assume that matrix factorization user vectors faithfully capture user similarity and that sampling collective members from clusters with propensity $p$ models how real collectives form.

Editorial extensions

If this is right

  • Organizers of a data campaign cannot rely on their group's characteristics alone; the presence of an unrelated campaign can cut efficacy from near 100% to about 25%, so efficacy estimates should account for concurrent actors.
  • AI and platform developers should expect and monitor for competing data campaigns; the framework gives them a way to compute pairwise constructiveness scores to anticipate which groups interfere.
  • In recommender systems, demoting items from the top-10 is easier than promoting items into it, so demotivation campaigns can succeed with smaller and less homogeneous groups.
  • Homogeneity is a secondary factor relative to group size for both promoting and demoting collectives, so recruiting more members matters more than curating a similar membership.
  • The tokenizer conflation result implies that seemingly different campaign strategies can be treated as the same signal by a model, so 'distinctness' of strategies cannot be assumed from surface-level difference.

Reading between the lines

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

  • If real collectives form through social networks rather than rating-similarity clusters, the homogeneity findings may shift; a natural extension is to build collectives from observed follower or interaction graphs and compare constructiveness scores.
  • The paper's tokenizer-conflation explanation suggests a testable prediction: using a tokenizer that keeps the two planted characters as distinct tokens should reduce the interference seen in the '100' versus '101' experiment; this could be checked without new data collection.
  • For three or more collectives, pairwise constructiveness scores may not sum linearly; the paper's suggestion of VCG-style marginal-harm calculations hints that adding a collective could flip the balance in ways pairwise scores miss.
  • The claim that demoting is easier than promoting implies an asymmetry in content moderation: campaigns seeking to bury content may require less coordination than campaigns seeking to surface content, which platforms could exploit when designing countermeasures.
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

3 major / 5 minor

Summary. The paper introduces a framework for studying algorithmic collective action by two or more collectives that each manipulate a shared data-dependent machine learning system. The framework distinguishes collectives by their objectives, construction, available actions, model access, affected parties, and measurement. The authors instantiate the framework in two empirical settings: (1) fine-tuned DistilBERT classifiers on resume data, where two collectives plant different text signals and attempt to steer classification to their own target classes, and (2) matrix-factorization recommender systems on MovieLens 100k, where collectives promote or demote items and the authors vary collective size, homogeneity (via a sampling propensity over user clusters), and the pairing of promoter/demoter archetypes. The main reported finding is that interactions between simultaneously acting collectives can be large, with a claim in the Abstract and Conclusion that a collective's efficacy can drop from near 100% to roughly 25% when a second collective acts. A secondary finding is that in recommender systems, collective size has a first-order effect on efficacy while homogeneity has a second-order effect, and that neither fully homogeneous nor fully heterogeneous collectives are uniformly most effective.

Significance. The paper addresses a genuinely novel question: prior work on algorithmic collective action has largely considered a single collective, whereas real platforms are likely to host multiple, independently motivated campaigns. The proposed framework is a useful conceptual contribution, and the recommender experiments include robustness checks across four combinations of clustering algorithm (k-means vs. k-medoids) and distance metric (L2 vs. cosine), as well as across centroid selection strategies. The constructiveness score is a sensible operationalization of between-collective influence. The finding that interactions between collectives can be substantial, if it survives the concerns below, would be a valuable and falsifiable contribution for both researchers and platform designers. The paper does not ship code or machine-checked proofs, so its contribution rests on the clarity of the framework and the soundness of the empirical analysis.

major comments (3)
  1. [Abstract; Section 7; Section 5.1] The headline claim of a drop from near 100% efficacy to roughly 25% (a 75% drop) is based on the A100 vs B101 condition in Figures 3d and 4b, but Section 5.1 explicitly states that for the distilbert-based-uncased tokenizer, the characters mapped to '100' and '101' are treated as identical. This means the model receives the same planted signal from both collectives, so the observed interaction is effectively a same-strategy conflict rather than an unintentional interaction between two genuinely distinct strategies. The Abstract and Conclusion do not carry the qualification that appears in the results section, and the caption of Figure 4b even calls the strategies distinct. The authors should either rerun the condition with tokens that remain distinct after tokenization, or revise the Abstract and Conclusion to attribute the headline drop to a same-strategy conflict and temper the 'distinct strategies' framing.
  2. [Section 5.1; Figures 3 and 4; Appendix B] The RQ1 language-model results are reported as point estimates only. Figure 3 shows no error bars, and Figure 4's caption says the heatmaps are averaged across 5 trials, while Appendix B states that each experimental condition was run 10 times; no variance or confidence interval is reported anywhere. Because the paper makes precise quantitative claims (e.g., a drop to nearly 25% efficacy, or that 'B101 requires nearly 2x larger participation than A100'), the absence of uncertainty reporting makes it impossible to assess whether the observed differences are meaningful relative to run-to-run noise. Please report per-condition means with standard deviations or confidence intervals, and reconcile the trial-count discrepancy.
  3. [Section 4.2 (Collective Formation and Item Selection) and Section 5.2 (Figures 5 and 6)] The homogeneity manipulation is confounded with target-item selection. After sampling collective members with propensity p, the targeted items are chosen as the collectively highest-rated items of those members, so varying p changes not only member homogeneity but also the coherence, popularity, and overlap of the item sets being promoted or demoted. This confound undercuts the conclusion that homogeneity has only a secondary influence on efficacy: the observed differences could be driven by the properties of the items selected under different p values. Please hold the target item set fixed across homogeneity levels, or provide an analysis that separates member homogeneity from item-set properties.
minor comments (5)
  1. [Section 5.2 (Figure 6 discussion)] The sentence 'However, it is the opposite' in the discussion of the green lines is confusing; the preceding sentence says the promoter wants a positive score and the demoter wants a negative score, and the next sentence clarifies that both are hindered. Please rewrite to state directly that both collectives receive the opposite sign from what they desire.
  2. [Figure 4b caption] The caption states that 'the strategy and the target used by each collective is distinct,' which is inconsistent with the paper's own observation in Section 5.1 that the tokenizer treats the '100' and '101' characters as identical. Please correct the caption to reflect the tokenizer aliasing.
  3. [Appendix B vs. Figure 4 caption] Appendix B says each experimental condition was run 10 times, while the Figure 4 caption says the results are averaged across 5 trials. Please reconcile this inconsistency.
  4. [Figure 11 caption] The caption says 'solid lines represent demoting groups while solid lines represent promoting groups'; the second phrase should presumably be 'dashed lines represent promoting groups.'
  5. [References [12] and [13]] References [12] and [13] appear to cite the same paper (Etter and Albu 2021) with the same title, authors, and publication; please merge or differentiate them.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all reported quantities are directly measured from experimental outcomes, and no load-bearing claim reduces to its own inputs.

full rationale

This is an empirical study whose central quantities are defined and computed directly from experimental measurements rather than fitted to reproduce conclusions. The efficacy measure in Section 5.1 is the top-one accuracy on held-out test resumes containing a planted signal, and the recommender analyses use the relative hit ratio and constructiveness score, which are explicit arithmetic combinations of directly measured hit ratios (Section 4.2: CT(ci, cj) = gi(Xi|theta_i∧j)/gi(Xi|theta') - gi(Xi|theta_i)/gi(Xi|theta')). No parameter in these formulas was fitted to the interaction results; model training is a separate step from evaluation, and the constructiveness score is a definition, not a fitted prediction. The paper's use of prior work, including Hardt et al. [19], is for experimental setup and metric conventions, and that external anchor is not used to prove the paper's findings by self-citation. The tokenizer aliasing noted in Section 5.1, where the two distinct characters are treated as identical, is a genuine validity caveat about one configuration, but it is a confound in interpreting a headline effect, not circular reasoning: the paper does not claim to derive the 75% drop from the definition of those characters. The stated limitations about model types and collective formation assumptions further indicate that the authors treat their results as conditional empirical findings rather than analytic consequences. Therefore, no circular step meeting the required evidentiary standard is present.

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

The central findings rest on design choices in how collectives are formed (clusters, sampling propensity, size), how they act (character insertion, rating flips), and how success is measured (top-1 accuracy, HR@K). None of these quantities are fitted to force the conclusion, but several are hand-set and could affect the reported interaction magnitudes.

free parameters (5)
  • Sampling propensity p = 0.1, 0.25, 0.5, 0.75, 1.0
    Controls collective homogeneity in Section 4.2; varied by design as an independent variable, not fitted to data.
  • Collective size N = 10, 20, 50 users
    Independent variable for size effects in the recommender experiments; chosen by the authors to probe first-order vs second-order impact.
  • Number of user clusters Q = 10
    K in the clustering step (Section 4.2); hand-chosen and affects the granularity of collective homogeneity sampling.
  • Number of targeted items V = 10
    Each collective targets its top 10 rated items; this choice directly shapes the HR@K metric and the difficulty of promotion vs demotion.
  • Signal insertion frequency = one character every 20 words
    Resume modification strategy inherited from Hardt et al.; a design choice that determines the LM classification results.
assumptions (4)
  • domain assumption Matrix factorization user vectors capture meaningful user similarity for collective formation.
    Section 4.2: 'These user vectors can be used as features to cluster similar users together. We expect that users that have similar interests are more likely to interact with each other if embarking on a data campaign.' This is stated as an expectation, with no empirical validation of the formation mechanism.
  • domain assumption Collective members act unilaterally and uniformly.
    Section 6 'Sanctions and Trustworthiness' acknowledges real collectives involve disagreement and noise; the experiments assume all members change ratings to 5 or 1, or insert the same character and relabel consistently.
  • domain assumption Fine-tuning a model on modified data approximates how platforms incorporate user data campaigns.
    Section 4.1: modified resumes are used directly in training; real systems may update via online learning, filtering, or other mechanisms that could weaken or strengthen the measured effects.
  • standard math Standard machine learning and clustering algorithms behave as expected in this setting.
    The paper relies on distilbert fine-tuning, Surprise matrix factorization, k-means and k-medoids clustering; no formal verification is provided, but these are standard tools.
invented entities (1)
  • Targeted Promoter and Targeted Demoter collective archetypes
    purpose: Categorize distinct collective objectives in the proposed framework, such as promoting a movie or demoting a resume class.
    Introduced in Section 3.2 as conceptual labels. They have no falsifiable handle outside the paper, but they are analytic categories rather than explanatory entities pulled in to fit the results.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Algorithmic Collective Action with Two Collectives." pith.science (2026). https://pith.science/paper/3V6ABWKT

@misc{pith2026250500195,
  author       = {Pith},
  title        = {Pith review of: Algorithmic Collective Action with Two Collectives},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3V6ABWKT}},
  note         = {Machine review of arXiv:2505.00195}
}
abstract

Given that data-dependent algorithmic systems have become impactful in more domains of life, the need for individuals to promote their own interests and hold algorithms accountable has grown. To have meaningful influence, individuals must band together to engage in collective action. Groups that engage in such algorithmic collective action are likely to vary in size, membership characteristics, and crucially, objectives. In this work, we introduce a first of a kind framework for studying collective action with two or more collectives that strategically behave to manipulate data-driven systems. With more than one collective acting on a system, unexpected interactions may occur. We use this framework to conduct experiments with language model-based classifiers and recommender systems where two collectives each attempt to achieve their own individual objectives. We examine how differing objectives, strategies, sizes, and homogeneity can impact a collective's efficacy. We find that the unintentional interactions between collectives can be quite significant; a collective acting in isolation may be able to achieve their objective (e.g., improve classification outcomes for themselves or promote a particular item), but when a second collective acts simultaneously, the efficacy of the first group drops by as much as $75\%$. We find that, in the recommender system context, neither fully heterogeneous nor fully homogeneous collectives stand out as most efficacious and that heterogeneity's impact is secondary compared to collective size. Our results signal the need for more transparency in both the underlying algorithmic models and the different behaviors individuals or collectives may take on these systems. This approach also allows collectives to hold algorithmic system developers accountable and provides a framework for people to actively use their own data to promote their own interests.

Figures

Figures reproduced from arXiv: 2505.00195 by the authors.

Figure 1
Figure 1. Framework overview. Collectives [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Group construction process used for forming two [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Multiple Collective Action in the Resume Modification Task. Two collectives, each with their own strategy, insert [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Role of varying sizes. Here, there are two different strategy sets where, for each scenario, the strategy and the target [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 6
Figure 6. Figure 6: Constructive score vs. sampling propensity for a [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Role of a third collective on efficacy of the other two collectives. We see different types of potential impacts. The [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Collective Action in Linear Setting. Two groups tar [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: The role of varying level of participation on collective efficacy. Left Figure a group changing class A (Craft-repair) [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: Effect of homogeneity on efficacy. Here, we fix [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]
Figure 11
Figure 11. Figure 11: Impact on group size and similarity on changing the HIT ratio for different choices of grouping/metrics (bottom right [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Impact of multiple collective data action on recommender systems with different metric choices (bottom right having [PITH_FULL_IMAGE:figures/full_fig_p016_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 49 canonical work pages

  1. [1]

    Rozz Albon and Tony Jewels. 2014. Mutual Performance Monitoring: Elaborating the Development of a Team Learning Theory. Group Decision and Negotiation 23, 1 (Jan. 2014), 149–164. https://doi.org/10.1007/s10726-012-9311-9

    work page doi:10.1007/s10726-012-9311-9 2014

  2. [2]

    Michael Anderson and Jeremy Magruder. 2012. Learning from the crowd: Re- gression discontinuity estimates of the effects of an online review database. The Economic Journal 122, 563 (2012), 957–989

    work page 2012

  3. [3]

    Jack Bandy and Nicholas Diakopoulos. 2020. #TulsaFlop: A Case Study of Algorithmically-Influenced Collective Action on TikTok. https://doi.org/10. 48550/arXiv.2012.07716 arXiv:2012.07716 [cs]

    arXiv doi:10.48550/arxiv.2012.07716 2020

  4. [4]

    Joachim Baumann and Celestine Mendler-Dünner. 2024. Algorithmic Collective Action in Recommender Systems: Promoting Songs by Reordering Playlists. arXiv preprint arXiv:2404.04269 (2024)

    work page Pith review arXiv 2024

  5. [5]

    Burns, Rebecca Peters, and George H

    Matthew K. Burns, Rebecca Peters, and George H. Noell. 2008. Using performance feedback to enhance implementation fidelity of the problem-solving team process. Journal of School Psychology 46, 5 (Oct. 2008), 537–550. https://doi.org/10.1016/j. jsp.2008.04.001

    doi:10.1016/j 2008

  6. [6]

    Chrysanthos Dellarocas. 2000. Immunizing online reputation reporting systems against unfair ratings and discriminatory behavior. In Proceedings of the 2nd ACM conference on Electronic commerce (Minneapolis, Minnesota, USA) (EC ’00). Association for Computing Machinery, New York, NY, USA, 150–157

    work page 2000

  7. [7]

    Yepeng Deng, Chunkai Zhang, and Xuan Wang. 2019. A Multi-objective Exam- ples Generation Approach to Fool the Deep Neural Networks in the Black-Box Scenario. In 2019 IEEE Fourth International Conference on Data Science in Cy- berspace (DSC). 92–99. https://doi.org/10.1109/DSC.2019.00022

    arXiv 2019

  8. [8]

    Algorithms ruin everything

    Michael A DeVito, Darren Gergle, and Jeremy Birnholtz. 2017. “Algorithms ruin everything”: #RIPTwitter, Folk Theories, and Resistance to Algorithmic Change in Social Media. In Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (Denver, Colorado, USA) (CHI ’17). Association for Computing Machinery, New York, NY, USA, 3163–3174

    work page 2017

Show all 68 references
  1. [9]

    Michael A DeVito, Jeffrey T Hancock, Megan French, Jeremy Birnholtz, Judd Antin, Karrie Karahalios, Stephanie Tong, and Irina Shklovski. 2018. The algo- rithm and the user: How can hci use lay understandings of algorithmic systems?. Preprint — final version forthcoming Algorit...

    2018

  2. [10]

    predatory use of AI

    Cheyenne DeVon. 2024. Billie Eilish, Nicki Minaj, Jon Bon Jovi and over 200 artists call for protections against “predatory use of AI”. https://www.cnbc.com/2024/ 04/05/billie-eilish-nicki-minaj-200-artists-sign-letter-against-ai-music.html

    2024

  3. [11]

    Alicia DeVos, Aditi Dhabalia, Hong Shen, Kenneth Holstein, and Motahhare Eslami. 2022. Toward User-Driven Algorithm Auditing: Investigating users’ strategies for uncovering harmful algorithmic behavior. In Proceedings of the 2022 CHI Conference on Human Factors in Computing Sy...

    2022

  4. [12]

    Michael Etter and Oana Brindusa Albu. 2021. Activists in the dark: Social media algorithms and collective action in two social movement organizations. Organization 28, 1 (Jan. 2021), 68–91. https://doi.org/10.1177/1350508420961532 Publisher: SAGE Publications Ltd

    2021 doi

  5. [13]

    Michael Etter and Oana Brindusa Albu. 2021. Activists in the dark: Social media algorithms and collective action in two social movement organizations. Organization 28, 1 (2021), 68–91

    2021

  6. [14]

    Yunhe Feng and Chirag Shah. 2022. Has ceo gender bias really been fixed? adver- sarial attacking and improving gender fairness in image search. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 36. 11882–11890

    2022

  7. [15]

    Javier García, Rubén Majadas, and Fernando Fernández. 2020. Learning ad- versarial attack policies through multi-objective reinforcement learning. En- gineering Applications of Artificial Intelligence 96 (Nov. 2020), 104021. https: //doi.org/10.1016/j.engappai.2020.104021

    2020

  8. [16]

    The Authors Guilds. 2023. New Authors Guild AI Survey Reveals That Authors Overwhelmingly Want Consent and Compensation for Use of Their Works. https://authorsguild.org/news/ag-ai-survey-reveals-authors- overwhelmingly-want-consent-and-compensation-for-use-of-their-works/

    2023

  9. [17]

    Sihan Guo, Ting Bai, and Weihong Deng. 2023. Targeted Shilling Attacks on GNN-based Recommender Systems. In Proceedings of the 32nd ACM Inter- national Conference on Information and Knowledge Management (<conf-loc>, <city>Birmingham</city>, <country>United Kingdom</country>, <...

    2023

  10. [18]

    Isha Gupta, Hidde Lycklama, Emanuel Opel, Evan Rose, and Anwar Hithnawi

  11. [19]

    Moritz Hardt, Eric Mazumdar, Celestine Mendler-Dünner, and Tijana Zrnic. 2023. Algorithmic Collective Action in Machine Learning. https://doi.org/10.48550/ arXiv.2302.04262 arXiv:2302.04262 [cs, stat]

    2023 doi arXiv

  12. [20]

    Moritz Hardt, Nimrod Megiddo, Christos Papadimitriou, and Mary Wootters

  13. [21]

    Maxwell Harper and Joseph A

    F. Maxwell Harper and Joseph A. Konstan. 2015. The MovieLens Datasets: History and Context. ACM Trans. Interact. Intell. Syst. 5, 4, Article 19 (dec 2015), 19 pages. https://doi.org/10.1145/2827872

    2015 doi

  14. [22]

    Bethany Hastie. 2020. Platform Workers and Collective Labour Action in the Modern Economy. University of New Brunswick Law Journal 71 (2020),

    2020

  15. [23]

    Douglas D Heckathorn. 1996. The dynamics and dilemmas of collective action. American sociological review (1996), 250–277

    1996

  16. [24]

    Nicolas Hug. 2020. Surprise: A Python library for recommender systems. Journal of Open Source Software 5, 52 (2020), 2174. https://doi.org/10.21105/joss.02174

    2020 doi

  17. [25]

    Matthew Jagielski, Giorgio Severi, Niklas Pousette Harger, and Alina Oprea. 2021. Subpopulation data poisoning attacks. In Proceedings of the 2021 ACM SIGSAC Conference on Computer and Communications Security . 3104–3122

    2021

  18. [26]

    Meijerink Jeroen, Jansen Giedo, and Daskalova Victoria. 2021. Platform Economy Puzzles: A Multidisciplinary Perspective on Gig Work . Edward Elgar Publishing. Google-Books-ID: cXg8EAAAQBAJ

    2021

  19. [27]

    Jiang, Lauren Brown, Jessica Cheng, Mehtab Khan, Abhishek Gupta, Deja Workman, Alex Hanna, Johnathan Flowers, and Timnit Gebru

    Harry H. Jiang, Lauren Brown, Jessica Cheng, Mehtab Khan, Abhishek Gupta, Deja Workman, Alex Hanna, Johnathan Flowers, and Timnit Gebru. 2023. AI Art and its Impact on Artists. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society (AIES ’23) . Association f...

    2023

  20. [28]

    Kameni Florentin Flambeau Jiechieu and Norbert Tsopze. 2021. Skills prediction based on multi-label resume classification using CNN with model predictions explanation. Neural Computing and Applications 33, 10 (2021), 5069–5087

    2021

  21. [29]

    Nadia Karizat, Dan Delmonaco, Motahhare Eslami, and Nazanin Andalibi. 2021. Algorithmic Folk Theories and Identity: How TikTok Users Co-Produce Knowl- edge of Identity and Engage in Algorithmic Resistance. Proc. ACM Hum.-Comput. Interact. 5, CSCW2 (Oct. 2021), 1–44

    2021

  22. [30]

    Peter Kollock. 1998. Social dilemmas: The anatomy of cooperation. Annual review of sociology 24, 1 (1998), 183–214

    1998

  23. [31]

    Shyong K Lam and John Riedl. 2004. Shilling recommender systems for fun and profit. In Proceedings of the 13th international conference on World Wide Web (New York, NY, USA) (WWW ’04). Association for Computing Machinery, New York, NY, USA, 393–402

    2004

  24. [32]

    Ya-Wen Lei. 2021. Delivering Solidarity: Platform Architecture and Collective Contention in China’s Platform Economy. American Sociological Review 86, 2 (April 2021), 279–309. https://doi.org/10.1177/0003122420979980 Publisher: SAGE Publications Inc

    2021 doi

  25. [33]

    Jing Lin, Long Dang, Mohamed Rahouti, and Kaiqi Xiong. 2021. ML Attack Models: Adversarial Attacks and Data Poisoning Attacks. https://doi.org/10. 48550/arXiv.2112.02797 arXiv:2112.02797 [cs]

    2021 doi arXiv

  26. [34]

    Michael Luca. 2016. Reviews, reputation, and revenue: The case of Yelp. com. Com (March 15, 2016). Harvard Business School NOM Unit Working Paper 12-016 (2016)

    2016

  27. [35]

    Caleb Malchik and Joan Feigenbaum. 2022. From Data Leverage to Data Co- Ops: An Institutional Model for User Control over Information Access. https: //doi.org/10.48550/arXiv.2201.10677 arXiv:2201.10677 [cs]

    2022 doi arXiv

  28. [36]

    Dragan, and Moritz Hardt

    Smitha Milli, John Miller, Anca D. Dragan, and Moritz Hardt. 2019. The Social Cost of Strategic Classification. In Proceedings of the Conference on Fairness, Ac- countability, and Transparency (FAT* ’19). Association for Computing Machinery, New York, NY, USA, 230–239. https:/...

    2019

  29. [37]

    Gemma Newlands, Christoph Lutz, and Christian Fieseler. 2018. Collective action and provider classification in the sharing economy. New Technology, Work and Employment 33, 3 (2018), 250–267. https://doi.org/10.1111/ntwe.12119 _eprint: https://onlinelibrary.wiley.com/doi/pdf/10...

    2018 doi

  30. [38]

    Annabel Nugent. 2021. Taylor Swift fans share tips on how to make old ‘Fear- less’ album disappear to mark release of ‘Taylor’s Version’. Independent (Apr 2021). https://www.independent.co.uk/arts-entertainment/music/news/taylor- swift-fearless-fans-b1829051.html

    2021

  31. [39]

    Elinor Ostrom. 1990. Governing the commons: The evolution of institutions for collective action. Cambridge university press

    1990

  32. [40]

    https://heinonline.org/HOL/Page?handle=hein.journals/unblj71&id=46& div=&collection=

  33. [41]

    Will B Payne. 2024. Review bombing the platformed city: Contested po- litical speech in online local reviews. Big Data & Society 11, 3 (2024), 20539517241275879

    2024

  34. [42]

    Derek Robertson. 2024. Exclusive poll: Americans favor AI data regula- tion. https://www.politico.com/newsletters/digital-future-daily/2024/05/06/ exclusive-poll-americans-favor-ai-data-regulation-00156350

    2024

  35. [43]

    Louise Overington and Gabriela Ionita. 2012. Progress monitoring measures: A brief guide. Canadian Psychology / Psychologie canadienne 53, 2 (2012), 82–92. https://doi.org/10.1037/a0028017 Place: US Publisher: Educational Publishing Foundation

    2012 doi

  36. [44]

    Tim Roughgarden. 2010. Algorithmic game theory. Commun. ACM 53, 7 (2010), 78–86

    2010

  37. [45]

    Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2019. Dis- tilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter. ArXiv abs/1910.01108 (2019)

    2019 arXiv

  38. [46]

    Rosen and Aaron S

    Michael A. Rosen and Aaron S. Dietz. 2017. Team Performance Mea- surement. In The Wiley Blackwell Handbook of the Psychology of Team Working and Collaborative Processes . John Wiley & Sons, Ltd, 479–

    2017

  39. [47]

    Shawn Shan, Wenxin Ding, Josephine Passananti, Stanley Wu, Haitao Zheng, and Ben Y Zhao. 2024. Nightshade: Prompt-Specific Poisoning Attacks on Text- to-Image Generative Models. In 2024 IEEE Symposium on Security and Privacy (SP). IEEE Computer Society, 212–212

    2024

  40. [48]

    Aaron Shaw, Haoqi Zhang, Andrés Monroy-Hernández, Sean Munson, Ben- jamin Mako Hill, Elizabeth Gerber, Peter Kinnaird, and Patrick Minder. 2014. Computer supported collective action. Interactions 21, 2 (March 2014), 74–77. https://doi.org/10.1145/2576875

    2014 doi

  41. [49]

    Hong Shen, Alicia DeVos, Motahhare Eslami, and Kenneth Holstein. 2021. Ev- eryday Algorithm Auditing: Understanding the Power of Everyday Users in Surfacing Harmful Algorithmic Behaviors. Proc. ACM Hum.-Comput. Interact. 5, CSCW2 (Oct. 2021), 1–29

    2021

  42. [50]

    Mehdi Shadmehr and Dan Bernhardt. 2011. Collective Action with Uncertain Payoffs: Coordination, Public Signals, and Punishment Dilemmas. American Political Science Review 105, 4 (Nov. 2011), 829–851. https://doi.org/10.1017/ S0003055411000359 Publisher: Cambridge University Press

    2011

  43. [51]

    Zhiyi Tian, Lei Cui, Jie Liang, and Shui Yu. 2022. A comprehensive survey on poisoning attacks and countermeasures in machine learning. Comput. Surveys 55, 8 (2022), 1–35

    2022

  44. [52]

    Conscious Data Contribution

    Nicholas Vincent and Brent Hecht. 2021. Can "Conscious Data Contribution" Help Users to Exert "Data Leverage" Against Technology Companies?Proceedings of the ACM on Human-Computer Interaction 5, CSCW1 (April 2021), 103:1–103:23. https://doi.org/10.1145/3449177

    2021 doi

  45. [53]

    Nicholas Vincent, Hanlin Li, Nicole Tilly, Stevie Chancellor, and Brent Hecht

  46. [54]

    Mingdan Si and Qingshan Li. 2020. Shilling attacks against collaborative recom- mender systems: a review. Artificial Intelligence Review 53 (2020), 291–319

    2020

  47. [55]

    Neil Widmeyer and Kimberly Ducharme

    W. Neil Widmeyer and Kimberly Ducharme. 1997. Team building through team goal setting. Journal of Applied Sport Psychology 9, 1 (March 1997), 97–

    1997

  48. [56]

    Phoenix Neale Williams and Ke Li. 2023. Black-Box Sparse Ad- versarial Attack via Multi-Objective Optimisation. 12291–12301. https://openaccess.thecvf.com/content/CVPR2023/html/Williams_Black- Box_Sparse_Adversarial_Attack_via_Multi-Objective_Optimisation_CVPR_ 2023_paper.html

    2023

  49. [57]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement De- langue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Ma...

    2020 arXiv

  50. [58]

    Jonathan Woodside, Tara Vinodrai, and Markus Moos. 2021. Bottom-up strategies, platform worker power and local action: Learning from ridehailing drivers.Local Economy 36, 4 (June 2021), 325–343. https://doi.org/10.1177/02690942211040170 Publisher: SAGE Publications Ltd

    2021 doi

  51. [59]

    Just-in-Place

    Katie J Wells, Kafui Attoh, and Declan Cullen. 2021. “Just-in-Place” labor: Driver organizing in the Uber workplace. Environment and Planning A: Economy and Space 53, 2 (March 2021), 315–331. https://doi.org/10.1177/0308518X20949266 Publisher: SAGE Publications Ltd

    2021 doi

  52. [60]

    A Reasonable Thing to Ask For

    Yuxi Wu, W Keith Edwards, and Sauvik Das. 2022. “A Reasonable Thing to Ask For”: Towards a Unified Voice in Privacy Collective Action. In Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems . 1–17. Preprint — final version forthcoming Algorithmic Colle...

    2022

  53. [65]

    A Reasonable Thing to Ask For

    Yuxi Wu, W. Keith Edwards, and Sauvik Das. 2022. “A Reasonable Thing to Ask For”: Towards a Unified Voice in Privacy Collective Action. InProceedings of the 2022 CHI Conference on Human Factors in Computing Systems (CHI ’22) . Association for Computing Machinery, New York, NY,...

    2022

  54. [67]

    (b)𝐿2 + K-means - Max distance 0.2 0.4 0.6 0.8 1.0 Homogeneity (p) □0.4 □0.3 □0.2 □0.1 0.0 CT Score CT (c↑ 1, c↑ 2) CT (c↑ 2, c↑ 1) CT (c↓ 1, c↓ 2) CT (c↓ 2, c↓ 1) CT (c↑ 1, c↓ 2) CT (c↑ 2, c↓ 1) (c) Cosine + K-Medoids - Uniform 0.2 0.4 0.6 0.8 1.0 Homogeneity (p) □0.3 □0.2 □0...

  55. [68]

    promoting

    (d) Cosine/K-Medoids - Max Distance Figure 12: Impact of multiple collective data action on recommender systems with different metric choices (bottom right having appeared in the main body). Homogeneity is on the x-axis while constructivess score is shows on the 𝑦-axis. Blue l...

  56. [113]

    https://doi.org/10.1080/10413209708415386 Publisher: Routledge _eprint: https://doi.org/10.1080/10413209708415386

    doi

  57. [502]

    https://doi.org/10.1002/9781118909997.ch21 Section: 21 _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/9781118909997.ch21

    doi

  58. [2016]

    In Proceedings of the 2016 ACM Conference on Inno- vations in Theoretical Computer Science (ITCS ’16)

    Strategic Classification. In Proceedings of the 2016 ACM Conference on Inno- vations in Theoretical Computer Science (ITCS ’16). Association for Computing Ma- chinery, New York, NY, USA, 111–122. https://doi.org/10.1145/2840728.2840730

    2016

  59. [2021]

    In Proceedings of the 2021 ACM Conference on Fair- ness, Accountability, and Transparency (FAccT ’21)

    Data Leverage: A Framework for Empowering the Public in its Relationship with Technology Companies. In Proceedings of the 2021 ACM Conference on Fair- ness, Accountability, and Transparency (FAccT ’21). Association for Computing Ma- chinery, New York, NY, USA, 215–227. https:/...

    2021

  60. [2024]

    arXiv:2410.08872 [cs.LG] https://arxiv.org/abs/2410.08872

    Fragile Giants: Understanding the Susceptibility of Models to Subpopula- tion Attacks. arXiv:2410.08872 [cs.LG] https://arxiv.org/abs/2410.08872

    arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.