Game Theory in Social Media: A Stackelberg Model of Collaboration, Conflict, and Algorithmic Incentives

Arjan Khadka

REVIEW 3 major objections 6 minor 24 references

Game Theory in Social Media: A Stackelberg Model of Collaboration, Conflict, and Algorithmic Incentives

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper claims that social media content strategy can be read as a Stackelberg game where the algorithm's weights on clicks, watch time, and shares, plus sponsor penalties, determine whether creators collaborate or engage in public…

desk verdict A clean, honest, but elementary exercise that never solves the Stackelberg game it claims to analyze; the central equilibrium claim is unsupported. read the letter →

arxiv 2506.05373 v2 pith:AVYBG3IB submitted 2025-05-29 cs.GT

Arjan Khadka This is my paper

classification cs.GT MSC 91A6591A80

keywords Stackelberggamesocialmediaalgorithmscreatorincentivescollaborationvsbeefingengagementmetricssponsorpenaltyalgorithmicgovernanceequilibriumanalysis

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 tries to show that the most visible strategic choice on social media—whether creators collaborate or pick public fights ('beefing')—can be modeled as the follower move in a leader-follower game. The algorithm is the leader and chooses weights $\alpha$ on clicks, $\beta$ on watch time, and $\gamma$ on shares; the creator then picks whichever strategy maximizes engagement reward minus a sponsor penalty $\delta$ for drama risk. The paper's central claim is that small shifts in those weights can flip the equilibrium between cooperative and conflictual content. If true, this gives platforms a concrete lever: tune the reward weights or the effective sponsor penalty to steer the content ecosystem toward or away from drama.

What carries the argument

The engine of the model is the creator utility function $U_{\text{creator}}(s) = \alpha\cdot \mathrm{Clicks}_s + \beta\cdot \mathrm{Watch}_s + \gamma\cdot \mathrm{Shares}_s - \delta\cdot \mathrm{DramaRisk}_s$, together with the algorithm's utility $U_{\text{algorithm}}(\alpha,\beta,\gamma) = \sum_s P_s(\alpha\,\mathrm{Clicks}_s + \beta\,\mathrm{Watch}_s + \gamma\,\mathrm{Shares}_s)$. The named object is the Stackelberg equilibrium: a leader-follower fixed point in which the algorithm moves first by setting the weights, creators choose the strategy that maximizes their payoff, and the algorithm optimizes its engagement objective while anticipating that response. The machinery works by reducing the whole content ecosystem to a comparison of two numbers—the utilities of collaboration versus beefing—so equilibrium selection is decided by which linear combination of engagement metrics is larger after the sponsor penalty.

What would settle it

The decisive calculation is the leader's problem in Section 2.8 written out literally. With $U_{\text{algorithm}} = \alpha\,\mathrm{Clicks}_{s^\ast} + \beta\,\mathrm{Watch}_{s^\ast} + \gamma\,\mathrm{Shares}_{s^\ast}$ and no constraint on $\alpha,\beta,\gamma$, scaling any weights by a factor $t>1$ scales the leader's utility by $t$, so the argmax over all triples does not exist. A reader can verify this directly from Table 1; if no finite equilibrium exists under the paper's stated assumptions, the 'tuning' narrative requires an added constraint or cost.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a platform-and-creator system reaches a Stackelberg equilibrium in which the algorithm's reward weights select the content style. With the illustrative engagement numbers used throughout—collaboration gives 2 clicks, 5 watch time, 3 shares, and zero drama risk; beefing gives 5 clicks, 2 watch time, 4 shares, and drama risk 3—a creator's best response is collaboration when watch time and the sponsor penalty matter, and beefing when clicks and shares dominate. The paper's worked examples put numbers on the switch: with $\alpha=1.0, \beta=2.0, \gamma=1.5, \delta=1.0$, collaboration wins $16.5$ to $12.0$; with $\alpha=2.5, \beta=0.5, \gamma=2.0, \delta=1.0$, beefing wins $18.5$ to $13.5$. The conclusion drawn is that algorithmic design and sponsor brand-safety pressure jointly determine what kinds of content become prevalent.

Load-bearing premise

The load-bearing premise is that the algorithm's optimization over weights has a finite solution: in Section 2.8 the leader maximizes a linear objective over unconstrained $\alpha,\beta,\gamma$, and a linear function over the whole space is unbounded. If that premise fails, the claimed equilibrium cannot be defined, even though the creators' side of the arithmetic is correct.

Editorial extensions

If this is right

If the algorithm heavily rewards clicks and shares, creators' best response moves toward beefing despite the sponsor penalty; the paper's Example 3 shows beefing at 18.5 versus collaboration at 13.5.
If the algorithm rewards watch time and sponsors penalize drama, collaboration is the equilibrium outcome; Example 1 shows collaboration at 16.5 versus beefing at 12.0.
Raising the sponsor sensitivity $\delta$ suppresses beefing even when the algorithm keeps click and share weights high.
Shifts in algorithmic weights are the channel through which viewer preferences indirectly change creator behavior.
Treating toxic content as a bad equilibrium rather than an inevitability gives platforms a formal reason to adjust incentive design.

Reading between the lines

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

Beyond the paper: if the weights are constrained to a simplex (for instance $\alpha+\beta+\gamma=1$), the unbounded leader problem becomes well-posed and the model's threshold structure predicts exact parameter regions where creators flip from collaboration to beefing.
Beyond the paper: the same payoff comparison can be tested empirically by measuring a creator's content style before and after a documented change in a platform's engagement weights; the direction of change should match the model's best-response formula.
Beyond the paper: relaxing the binary strategy set to a drama level $d\in[0,1]$ would replace the all-or-nothing switch with interior equilibria, a direction only sketched in the paper's future-work section.
Beyond the paper: because viewers act only through the algorithm's weights, the model implies that audience pressure changes creator behavior only when platforms reweight metrics; cross-platform comparisons with different weight policies would isolate that channel.

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.

Desk Editor's Note

read the letter

Colleague, the short version: this is a clearly written but elementary exercise that never actually solves the Stackelberg game it claims to analyze. The arithmetic is right, the exposition is clean, and the limitations section is genuinely honest, but the central equilibrium claim is not supported.

What is new: very little. The idea that platforms set reward weights and creators respond is already in the cited Tang et al. [7] and Hron et al. [8] papers. The binary collab-versus-beef choice with a sponsor penalty is a special case, not a generalization. The paper's main insight—higher alpha/gamma favors beefing, higher beta favors collaboration—is read directly off Table 1, which the paper itself labels 'illustrative.' With Clicks 5 vs 2, Shares 4 vs 3, Watch 2 vs 5, and DramaRisk 3 vs 0, the comparative statics follow mechanically. The numerical 'simulations' are just plugging numbers into a linear formula; there is no simulation.

The soft spot the stress-test note identifies is real and load-bearing. Section 2.8 defines the leader's problem as an unconstrained argmax over (alpha, beta, gamma) of a linear objective. Because the objective is linear in the weights, no finite maximum exists—if all weights are scaled up, both the leader's payoff and the creators' payoffs scale up without bound. The paper never solves the leader's problem, even under a normalized constraint. So there is no Stackelberg equilibrium as defined. The follower best response is characterized correctly, but the leader half of the game is missing.

To be fair, the paper does not hide its scope: the Limitations section lists binary strategies, static environment, homogeneous agents, linear utility, and passive viewers/sponsors, and calls the model 'simplified.' That honesty earns some credit. But a simplified model still needs to be internally well-posed. The unbounded leader problem is not a minor gap; it invalidates the claimed equilibrium shifts.

The behavioral evidence section is mostly a literature survey with plausible citations, but it does not test the model or calibrate the table. The case studies are stylized illustrations, not evidence.

Who gets value: an undergraduate or a curious layperson might find this a tidy demonstration of how to set up a two-strategy game with a linear payoff and derive threshold conditions. As a research contribution, it does not advance the state of the art.

Recommendation: I would not send this to peer review in its current form. It needs a properly constrained leader problem, an actual solution or at least existence proof, and ideally calibrated data. If the author fixes those, it could become a decent teaching note or workshop paper. For now, treat it as a well-meaning exercise.

Referee Report

3 major / 6 minor

Summary. The paper proposes a Stackelberg game in which a social media platform's algorithm (leader) chooses weights α, β, γ on clicks, watch time, and shares, and content creators (followers) choose between 'collaboration' and 'beefing' to maximize a linear utility that includes a sponsor penalty δ. Using an illustrative engagement table, the paper computes worked examples, asserts that small changes in algorithmic weights lead to dramatic shifts in equilibrium behavior, and draws platform-policy implications. The central claimed contribution is that platforms can steer creators between collaboration and conflict by tuning engagement weights and sponsor sensitivity.

Significance. If the model were correctly specified and calibrated, it would provide a compact formal illustration of how algorithm design shapes creator incentives, a topic of clear practical importance. The paper deserves credit for writing down a simple, transparent arithmetic framework and for explicitly listing many limitations. However, the significance in its current form is limited: the leader's optimization problem is not solved and is in fact unbounded as written, so the advertised Stackelberg equilibrium does not exist; the main behavioral conclusions are direct restatements of the hand-set Table 1 values; and the numerical section contains only hand computations, not simulations. These issues are load-bearing because they concern the paper's core claim about equilibrium shifts under changing algorithm weights.

major comments (3)

[§2.8, Step 2] The leader's optimization problem is ill-posed. Step 2 defines (α*, β*, γ*) = argmax U_algorithm(s*(α, β, γ)) with U_algorithm = α·Clicks_s* + β·Watch_s* + γ·Shares_s* and no constraints on α, β, γ. Since this objective is linear and homogeneous in the weights, it is unbounded above. For example, set α=β=γ=t and δ=1. Then for t>3, beefing is the follower's best response (beef utility 11t−3 versus collaboration utility 10t), and the leader's payoff is 11t, which grows without bound as t→∞. Hence no finite Stackelberg equilibrium exists as defined, contradicting the claims in §2.8 and §8 of a stable equilibrium and of equilibrium shifts. A normalization or compact constraint on the weights would be needed, but Section 2.8 does not solve even such a restricted problem. The paper's headline conclusion therefore is not supported by the formal model as written.
[§2.3, Table 1, and §4.3–4.5, §5.1–5.2] The behavioral conclusions are direct restatements of the assumed engagement values. Table 1 fixes beefing to yield more clicks (5 vs. 2) and shares (4 vs. 3), collaboration to yield more watch time (5 vs. 2), and assigns drama risk only to beefing. Consequently, the worked examples in §4.3–4.5 and the implications in §5.1–5.2 that high α/γ favor beefing and high β favors collaboration merely restate those input assumptions. The values are explicitly described as illustrative, and §7.2 lists empirical calibration as future work, but the paper nevertheless advances 'small shifts in weights lead to dramatic shifts in equilibrium behavior' as a substantive finding. To support that claim, the model would need either calibrated engagement values or an explicit statement that the exercise is purely illustrative and carries no empirical implications.
[§2.6 vs. §2.8 Step 2] The algorithm's objective is defined inconsistently between two sections. Equation (2) in §2.6 defines U_algorithm as a sum over strategies weighted by P_s, the proportion of creators choosing each strategy. In §2.8 Step 2, however, U_algorithm is written as α·Clicks_s* + β·Watch_s* + γ·Shares_s*, with no P_s term. The latter implicitly assumes all creators choose the same pure strategy or that P_s is degenerate for the chosen strategy. If mixed populations are intended, the leader's objective is a convex combination over strategies, and the maximizer can differ from the single-representative-creator case. The paper never states which interpretation is intended, and this matters for any claimed equilibrium characterization.

minor comments (6)

[§4 title] The section is titled 'Numerical Examples and Simulations,' but no simulation is performed; the examples are a few hand computations with fixed parameter values.
[§2.4] There are typographical issues in the variable definitions, including 'Sharess' in the explanation of Shares and the mixed use of 'click-through rate' where a count 'Clicks' is used in the formula.
[§2.5] The nonlinear utility extension contains a formatting artifact ('β · p Watchs' instead of a square-root expression) and is never used in the subsequent analysis, so it is unclear what role this extension plays in the paper.
[§3] Section 3 discusses bounded rationality, satisficing, and level-k thinking, but the formal model in Sections 2 and 4 uses fully rational best responses; the connection between the behavioral discussion and the model is not made.
[§5.3] The claim that 'strategic transparency reduces manipulation' is not derived from the model, since transparency (knowledge of α, β, γ) is not a parameter in the game and the model does not analyze hidden vs. public weights.
[§6] The case studies in Section 6 are anecdotal post-hoc illustrations; the statement that they 'validate the theoretical model's assumptions and structure' overstates what can be concluded from qualitative examples.

Circularity Check

1 steps flagged · score 7.0 of 10

The central behavioral predictions are restatements of the hand-set engagement numbers in Table 1, so the Stackelberg 'insight' is an input rather than a derived result.

self definitional [Section 2.3 Table 1; Section 2.8 Implications of the Model]
"Strategy Clicks Watch Time Shares Drama Risk; Collaboration 2 5 3 0; Beefing 5 2 4 3. ... If the algorithm puts a high weight on clicks (α) and shares (γ), creators will lean towards beefing because this strategy yields more clicks and shares despite the sponsor penalty."

The paper's qualitative prediction—high α and γ select beefing, high β selects collaboration—is exactly the ordering of the illustrative numbers in Table 1 inserted into the linear utility (1). The inequality U_beef > U_collab reduces to 3α + γ > 3β + 3δ, which is nothing but the assumed fact that beefing has more clicks (5>2) and shares (4>3) while collaboration has more watch time (5>2). The numerical examples in Sections 4.3 and 4.5 recompute this same inequality. Thus the central claimed insight is put in by hand via Table 1, not derived from independent data, calibration, or a substantive equilibrium argument.

full rationale

The model's central claim that algorithmic weight shifts change creator behavior between beefing and collaboration is an arithmetic consequence of the author-chosen payoff table. Because Table 1 is explicitly 'illustrative' and not calibrated to any data, the comparative statics do not constitute an independent prediction; they formalize the assumption that beefing is clicky and collaborative content is long-watch-time content. That is the clear self-definitional circularity counted here. Separately, the leader's optimization in Section 2.8 is unbounded as written because the algorithm's objective is linear in unconstrained weights, so no finite Stackelberg equilibrium is actually established; that is a formal-correctness problem, not itself a circularity, but it further weakens the paper's headline claim. I do not count the absence of self-citations or the heavy reliance on external references as circular, since those are not load-bearing in a self-referential way. The paper is transparent about its illustrative parameters, but the claimed 'demonstration' of equilibrium dynamics is a restatement of those inputs rather than a result that could fail against data.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The central claim rests on a hand-assigned payoff table and linear utility forms, with the leader's optimization problem unconstrained and unsolved. No data, code, or external benchmarks are provided.

free parameters (3)

Engagement outcome table (Table 1 values for Clicks, Watch, Shares, DramaRisk) = Collab: (2,5,3,0); Beef: (5,2,4,3)
Hand-selected 'illustrative' values (Table 1 note); they encode the rankings that drive all conclusions and are not calibrated to any platform data.
Algorithm weights (alpha, beta, gamma) = Example sets (1.0,2.0,1.5) and (2.5,0.5,2.0)
Chosen by hand in Section 4 to represent long-form and short-form platforms; no calibration or sensitivity range is provided.
Sponsor sensitivity delta = 1.0 and 2.5 in the examples
Chosen by hand in Sections 4.3-4.4 to illustrate stricter brand safety; no empirical basis is given.

assumptions (4)

domain assumption Creators are rational expected-utility maximizers who know alpha, beta, and gamma and choose the single best strategy (Eq. 1, Section 2.4).
Needed for the best response s*; Section 3 later introduces bounded rationality and heuristics, which is never reconciled with this assumption.
ad hoc to paper The engagement outcome of each strategy is fixed and identical for all creators and all platforms (Table 1).
No empirical source is given; the paper labels the values illustrative, yet every numerical conclusion depends on them.
ad hoc to paper The algorithm can choose arbitrary real weights (alpha, beta, gamma) with no normalization or cost and maximizes its linear utility (Section 2.8).
As written, this makes the leader's objective unbounded, so a finite Stackelberg equilibrium is not guaranteed.
domain assumption Viewer preferences are fully summarized by the algorithm's weights (Section 2.6), so viewers need not be modeled as players.
A deliberate simplification acknowledged in Section 7.1; it removes viewer backlash or boycott dynamics.

invented entities (1)

DramaRisk scalar
purpose: Quantifies controversy or risk of a strategy; assigned 0 for collaboration and 3 for beefing in Table 1.
A constructed index with no external measurement or benchmark; it directly sets the size of the sponsor penalty.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Game Theory in Social Media: A Stackelberg Model of Collaboration, Conflict, and Algorithmic Incentives." pith.science (2026). https://pith.science/paper/AVYBG3IB

@misc{pith2026250605373,
  author       = {Pith},
  title        = {Pith review of: Game Theory in Social Media: A Stackelberg Model of Collaboration, Conflict, and Algorithmic Incentives},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AVYBG3IB}},
  note         = {Machine review of arXiv:2506.05373}
}

read the original abstract

This research models the social media content creation and the choices that creators make as a Stackelberg game. The platform's algorithms, such as TikTok's and YouTube's, function as leaders, and they set rules to maximize users' engagement with their platforms. Then, content creators, who function as followers in this Stackelberg Game, respond to this by selecting strategies; in this instance, we are specifically focusing on collaboration or conflict, referred to in this paper as 'beefing.' They do this in order to maximize views and personal payoffs. The viewer's preferences are already placed within the algorithmic utility function, while the external sponsors will impose penalties on high-risk strategies, namely, beefing multiple times. This paper ultimately demonstrates, through the use of math, how shifts in algorithmic weights determine equilibrium creator behavior.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages

[7]

Tang, J., Jiang, M., Zhang, Y., & Yan, X. (2019). A Stackelberg Game Model for Con- tent Promotion in Social Media Platforms. IEEE Transactions on Computational Social Systems, 6(2), 278–287. Models content promotion using Stackelberg games to analyze platform and creator in- teractions

work page 2019
[8]

exposure game

Hron, J., Krauth, K., Jordan, M. I., Kilbertus, N., & Dean, S. (2023). Modeling Content Creator Incentives on Algorithm-Curated Platforms. In Proc. of ICLR 2023 . Formalizes an “exposure game” to study how recommender-system choices shape creator incentives. 17

work page 2023
[1]

Stackelberg, H. von. (1934). Marktform und Gleichgewicht . Vienna: Springer-Verlag. Foundational work introducing Stackelberg competition, forming the basis for leader- follower strategic models

work page 1934
[2]

Tufekci, Z. (2015). Algorithmic harms beyond Facebook and Google: Emergent chal- lenges of computational agency. Colorado Technology Law Journal, 13(203), 203–218. Discusses algorithmic influence on user behavior, which inspired our treatment of algo- rithms as strategic actors

work page 2015
[3]

Goldfarb, A., & Tucker, C. (2011). Online Display Advertising: Targeting and Obtru- siveness. Marketing Science, 30(3), 389–404. Explores the relationship between engagement metrics and platform incentives

work page 2011
[4]

Ferrara, E., Varol, O., Davis, C., Menczer, F., & Flammini, A. (2016). The rise of social bots. Communications of the ACM , 59(7), 96–104. Touches on artificial engagement and its manipulation—relevant to our model of reward optimization

work page 2016
[5]

J., & Rubinstein, A

Osborne, M. J., & Rubinstein, A. (1994). A Course in Game Theory . MIT Press. A rigorous introduction to game theory concepts, including Stackelberg games

work page 1994
[6]

Bakshy, E., Messing, S., & Adamic, L. A. (2015). Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239), 1130–1132. Supports the idea that algorithmic curation affects viewer behavior and engagement dynamics

work page 2015

Show all 24 references

[9]

Hou, M. (2023). Cultural heuristics in online video creation: How creators adapt to plat- form dynamics. Journal of Digital Media & Policy , 14(1), 45–61. Discusses heuristic and adaptive behavior by creators under platform constraints, sup- porting bounded rationality in deci...

2023
[10]

folk theories

Choi, Y., Kang, E. J., Lee, M. K., & Kim, J. (2023). Creator-Friendly Algorithms: Behaviors, Challenges, and Design Opportunities in Algorithmic Platforms. In Proc. of CHI ’23 , 22 pages. Mixed-methods study of how YouTube creators form “folk theories” of opaque algorithms and...

2023
[11]

The algorithm is like a mercurial god

Verwiebe, R., Buder, C., Weissmann, S., Osorio-Krauter, C., & Philipp, A. (2024). “The algorithm is like a mercurial god”: Exploring content creators’ perception of algorithmic agency on YouTube. New Media & Society , 26(1), 123–142. Qualitative interviews uncover how YouTube ...

2024
[12]

F., Ho, T.-H., & Chong, J.-K

Camerer, C. F., Ho, T.-H., & Chong, J.-K. (2004). A cognitive hierarchy model of games. The Quarterly Journal of Economics , 119(3), 861–898. Introduces level- k thinking and bounded rationality in strategic behavior, relevant to limited lookahead in beef scenarios

2004
[13]

Falk, A., Fehr, E., & Fischbacher, U. (2006). Neural correlates of retaliation. Science, 314(5806), 1141–1144. Shows behavioral basis for retaliatory dynamics and conflict escalation, illuminating beef- ing behavior

2006
[14]

Zheng, L., Huang, B., Qiu, H., & Bai, H. (2023). The role of social media followers’ agency in influencer marketing: A study based on the heuristic–systematic model of information processing. International Journal of Advertising , 42(5), 789–810. Applies the heuristic–systemat...

2023
[15]

Covington, P., Adams, J., & Sargin, E. (2016). Deep neural networks for YouTube recommendations. Proceedings of the 10th ACM Conference on Recommender Systems , 191–198. Describes algorithmic feedback loops and creators’ experimental adaptations

2016
[16]

F., Marder, B., Reich, J., & Brooks, C

Kizilcec, R. F., Marder, B., Reich, J., & Brooks, C. (2021). Algorithms and creators: Feedback and adjustment on YouTube. Journal of Learning Analytics , 8(2), 1–20. Analyzes creator responses to audience feedback under algorithmic uncertainty

2021
[17]

Davies, R. (2024). Good intent, or just good content? Assessing MrBeast’s philan- thropy. Journal of Philanthropy and Marketing, 29(2), 185–200. Analyzes MrBeast’s philanthropic approach, discussing its ethical, economic, and cultural implications

2024
[18]

Burgess, J., & Green, J. (2018). YouTube: Online Video and Participatory Culture (2nd ed.). Polity Press. 18 Discusses beefing and participatory content culture on YouTube, contextualizing creator conflict

2018
[19]

Krishna, V. (2009). Game Theory: Analysis of Conflict . Academic Press. Introduces foundational models in game theory, including dynamic and auction games, relevant to modeling strategic creator behavior

2009
[20]

Fudenberg, D., & Tirole, J. (1991). Game Theory. MIT Press. Covers equilibrium concepts and strategic interaction, useful for understanding multi- agent dynamics in creator-algorithm games

1991
[21]

Gibbons, R. (1992). A Primer in Game Theory . Pearson Education. A concise and accessible introduction to game theory, including Stackelberg competition and mixed strategies

1992
[22]

D., & Green, J

Mas-Colell, A., Whinston, M. D., & Green, J. R. (1995). Microeconomic Theory. Oxford University Press. Provides formal tools for modeling decision-making and strategic interaction in economics and algorithmic environments

1995
[23]

Young, H. P. (2015). Strategic Learning and Its Limits . Oxford University Press. Explores learning in games and bounded rationality—relevant to how creators adapt to algorithmic feedback

2015
[24]

Liu, C., Wang, Z., & Zhang, J. (2020). Cross-platform content strategy: Evidence from social media creators. Journal of Interactive Marketing , 52, 26–43. Analyzes how creators adjust their content strategies across multiple platforms, support- ing the need for multi-platform ...

2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [7]

Tang, J., Jiang, M., Zhang, Y., & Yan, X. (2019). A Stackelberg Game Model for Con- tent Promotion in Social Media Platforms. IEEE Transactions on Computational Social Systems, 6(2), 278–287. Models content promotion using Stackelberg games to analyze platform and creator in- teractions

work page 2019

[2] [8]

exposure game

Hron, J., Krauth, K., Jordan, M. I., Kilbertus, N., & Dean, S. (2023). Modeling Content Creator Incentives on Algorithm-Curated Platforms. In Proc. of ICLR 2023 . Formalizes an “exposure game” to study how recommender-system choices shape creator incentives. 17

work page 2023

[3] [1]

Stackelberg, H. von. (1934). Marktform und Gleichgewicht . Vienna: Springer-Verlag. Foundational work introducing Stackelberg competition, forming the basis for leader- follower strategic models

work page 1934

[4] [2]

Tufekci, Z. (2015). Algorithmic harms beyond Facebook and Google: Emergent chal- lenges of computational agency. Colorado Technology Law Journal, 13(203), 203–218. Discusses algorithmic influence on user behavior, which inspired our treatment of algo- rithms as strategic actors

work page 2015

[5] [3]

Goldfarb, A., & Tucker, C. (2011). Online Display Advertising: Targeting and Obtru- siveness. Marketing Science, 30(3), 389–404. Explores the relationship between engagement metrics and platform incentives

work page 2011

[6] [4]

Ferrara, E., Varol, O., Davis, C., Menczer, F., & Flammini, A. (2016). The rise of social bots. Communications of the ACM , 59(7), 96–104. Touches on artificial engagement and its manipulation—relevant to our model of reward optimization

work page 2016

[7] [5]

J., & Rubinstein, A

Osborne, M. J., & Rubinstein, A. (1994). A Course in Game Theory . MIT Press. A rigorous introduction to game theory concepts, including Stackelberg games

work page 1994

[8] [6]

Bakshy, E., Messing, S., & Adamic, L. A. (2015). Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239), 1130–1132. Supports the idea that algorithmic curation affects viewer behavior and engagement dynamics

work page 2015