Skip to Content
  • Article
  • Open Access

3 February 2026

25 Pages

Transcending the Paradox of Statistical and Value Rationality: A Tripartite Evolutionary Game Analysis of E-Commerce Algorithmic Involution

,
,
and
1
Alibaba Business School, Hangzhou Normal University, Hangzhou 311121, China
2
School of Management, Hangzhou Dianzi University Information Engineering College, Hangzhou 311305, China
3
School of Management and E-Business, Zhejiang Gongshang University, Hangzhou 310018, China
4
School of Computer Science and Technology, Zhejiang Gongshang University, Hangzhou 310018, China

Abstract

The unbridled pursuit of statistical rationality has precipitated a crisis of value rationality in e-commerce ecosystems, leading to algorithmic involution—a dilemma characterized by destructive hyper-competition. To reconcile this theoretical paradox and explore effective governance pathways, this paper constructs a tripartite evolutionary game model involving e-commerce platforms, government regulators, and consumers. Simulation results indicate that high-intensity government deterrence constitutes the necessary stability foundation of hard constraints, while consumer activism acts as the decisive accelerator of the soft environment contingent on high synergistic gains and low information screening costs. Furthermore, a platform’s pivot toward “algorithm for good” is not driven by altruism, but by the rational calibration between short-term extractive gains and long-term benevolent returns. Sensitivity analysis confirms that reducing the ratio of these two factors is the effective lever to speed up system convergence. Finally, effective governance requires restructuring this payoff matrix by establishing dynamic penalty mechanisms and transparent low-cost feedback channels to render ethical algorithmic behavior a dominant strategy in terms of economic rationality. This research aims to guide the e-commerce ecosystem from a zero-sum game of involution toward a sustainable equilibrium of multi-party value co-creation.

1. Introduction

In the digital economy era, algorithms have ascended as the “invisible hand” governing e-commerce ecosystems, fundamentally reshaping resource allocation. While initially designed to optimize matching efficiency between supply and demand—a manifestation of statistical rationality—these algorithms are increasingly deviating from their neutral origins. This tension mirrors Max Weber’s classical sociological distinction between statistical rationality and value rationality [1]. Statistical rationality prioritizes calculability and efficiency, a principle perfectly embodied by metrics such as click-through rates and GMV, whereas value rationality demands adherence to ultimate values including justice, ethics, and human welfare. Driven by the imperative of profit maximization, platforms prioritize calculable metrics over substantive social values. This phenomenon is defined as “algorithmic management”, where code dictates traffic allocation and value distribution, often at the expense of human agency [2,3]. This dominance of instrumental efficiency over value rationality has precipitated what we term “algorithmic involution” (Neijuan). Based on Geertz’s and Huang’s use of involution to describe the intensification of input and social complexity without proportional gains in productivity or structural transformation [4,5], we define algorithmic involution as a pattern in which platforms continuously escalate algorithmic control and data extraction to chase marginal improvements in short-term statistical indicators (such as click-through rates or GMV) under a largely fixed technological and regulatory regime, thereby intensifying competition and extraction without commensurate gains in aggregate welfare or genuine innovation. In this destructive “race to the bottom” equilibrium, e-commerce platforms deploy aggressive strategies such as discriminatory pricing [6], coercive traffic allocation, and addictive recommendation systems to extract short-term surplus, reducing human users and merchants to mere data points in an optimization function [7].
This hegemony of statistical rationality traps the e-commerce ecosystem in a classic prisoner’s dilemma, creating what Weber might describe as an algorithmic “iron cage” where value considerations are suffocated by the relentless pursuit of metrics. Although stakeholders recognize the necessity of value rationality—embodied in “Tech for Good”, fair competition, and consumer welfare—individual platforms face a perverse incentive structure. A platform that unilaterally pursues ethical algorithms risks losing market share to competitors employing predatory, extractive strategies. Consequently, a dynamic of “bad money drives out good” emerges. The conflict between the individual rationality of profit maximization and the collective rationality of sustainable development results in a “tragedy of the commons”, where consumer trust erodes and the industry stagnates in a zero-sum game of efficiency without ethics.
Reconciling this paradox presents a profound challenge due to the “black box” of algorithmic decision-making in e-commerce [8]. Unlike traditional market misconduct, algorithmic violations are concealed within complex codes and dynamic data flows, characterized by high opacity and personalization. Traditional top-down government regulation, which relies on static rules, struggles to penetrate the veil of statistical rationality [9]. Regulators face the difficult trade-off: high-intensity intervention incurs prohibitive enforcement costs and may stifle innovation, while a laissez-faire approach allows the alienation of value rationality to deepen. This governance void underscores the urgent need for a mechanism that can realign the platform’s statistical objectives with societal value constraints.
Existing literature on algorithm governance has predominantly utilized game theory to analyze stakeholder interactions. However, most studies are confined to a binary framework between the government and the platform, largely overlooking the crucial role of users as agents of value rationality [10]. Consumers are the direct victims of algorithmic exploitation, and their feedback serves as “distributed sensors” that can pierce the opacity of algorithms. While recent tripartite studies explore penalty mechanisms for price discrimination [11,12], they predominantly view consumer participation through monetary incentives, overlooking the technical synergy where feedback directly enhances regulatory detection capabilities. Furthermore, current research often treats government regulation and consumer supervision as independent variables, failing to capture the synergistic feedback loop where effective government response empowers consumer activism, and consumer participation lowers the marginal cost of law enforcement.
To explore pathways for transcending the conflict between statistical and value rationality, this study formulates a tripartite evolutionary game model. This dynamic framework, which includes e-commerce platforms, government regulators, and consumers, allows us to examine the evolution of strategic choices over time, in contrast to traditional static equilibrium analysis. The core contributions are as follows: First, we theoretically prove that “government–public collaboration” is the essential catalyst for reconciling efficiency with ethics. Second, we quantify the threshold where “algorithm for good” become a rational economic choice, effectively internalizing value rationality into the platform’s payoff matrix. Third, we demonstrate that while regulatory “hard constraints” are foundational, the “soft environment” of consumer synergy determines the system’s convergence toward a Pareto-optimal equilibrium. This study aims to guide the ecosystem from the involution of statistical rationality toward a symbiotic equilibrium of multi-party value co-creation.

2. Literature Review

2.1. The Statistical Rationality and Algorithm Involution

The evolution of algorithmic strategies on e-commerce platforms can be interpreted as the trajectory of statistical rationality asserting dominance over the ecosystem. Initially, platforms pursue technological efficiency to optimize matching, but this pursuit often alienates substantive value rationality [13], leading to a dynamic trade-off between short-term commercial gains and long-term ecosystem health. This tendency is consistent with international discussions on algorithmic accountability and opacity, which emphasize that metric-driven algorithmic systems often externalize social costs while remaining difficult to scrutinize due to socio-technical limits of transparency [14,15,16,17].
On one hand, driven by the logic of statistical rationality (e.g., maximizing CTR and GMV), platforms have strong incentives to deploy algorithm involution. For example, big data discriminatory pricing boosts short-term revenue through precise calculation but at the expense of consumer fairness [18]. Similarly, bid-ranking distorts search results to generate advertising income, strictly following profit maximization logic while increasing information costs for consumers and undermining fair competition [19]. Furthermore, immersive content feeds create “filter bubbles” that induce irrational consumption and entrench consumer biases [20,21]. Related international evidence on personalization and selective exposure shows that algorithmic curation can measurably shape what users see and reinforce segmented information environments, reinforcing concerns about “filter bubbles/echo chambers” under opaque ranking and recommendation regimes [22,23]. From a platform economics perspective, such metric-driven strategies are amplified by the two-/multi-sided nature of platforms and indirect network effects, which incentivize aggressive optimization of engagement and monetization outcomes [24,25,26,27]. Collectively, these behaviors define the core of algorithmic involution—an arms race based on instrumental efficiency that leads to massive technological investment but diminishes user welfare, manifesting as a severe alienation of value rationality.
On the other hand, choosing “algorithm for good”, which attempts to restore value rationality by prioritizing fairness, transparency, and ethics, entails explicit compliance costs and implicit opportunity costs [28]. Studies widely agree that algorithmic transparency and ethical audits can increase a platform’s operational burden and potentially weaken its core competitiveness [29]. International research further cautions that transparency ideals face structural limitations (seeing without knowing), and that accountability often requires institutional design (e.g., auditing, documentation, and reviewable decision processes) rather than mere disclosure [30]. Even when platforms establish self-governance mechanisms [31], such “self-discipline” often fails to resist the lure of short-term statistical gains without strong external constraints [32]. Consequently, platforms are caught in a classic strategic dilemma: adopting an involution strategy yields excess short-term returns but carries long-term risks of reputational damage and user attrition [33], whereas a pro-social strategy builds long-term trust but may lead to a short-term competitive disadvantage.

2.2. The Traditional Bilateral Governance Framework

In response to the alienation caused by algorithmic involution, external intervention becomes necessary to re-embed value rationality into the market. A significant body of literature has focused on bilateral “government–platform” game models to explore the effectiveness of various regulatory tools, such as fines, audits, and standard-setting [34]. In parallel, international scholarship conceptualizes “algorithmic regulation” and “governance by algorithms” as emerging modes of social ordering, stressing that rule enforcement is complicated by algorithmic opacity, rapid iteration, and scale [35]. Platform governance research also highlights that automation and data-driven control create new enforcement bottlenecks and accountability gaps, making oversight costly and contested [36]. However, these studies consistently find that traditional regulatory models face two major challenges. First, information asymmetry prevents regulators from detecting and verifying platform misconduct in a timely and precise manner [37]. This constraint aligns with international findings that opacity in algorithmic systems can be intentional, technical, and institutional, thereby limiting what regulators and affected parties can know even under transparency requirements [15,16]. Second, the costs of deep technical audits and continuous oversight are prohibitively high [38]. International accountability research similarly argues that scalable auditing and contestability mechanisms are resource-intensive, and that “accountable algorithms” require sustained institutional capacity rather than one-off compliance [9]. This places regulators in their own strategic bind between strong and weak enforcement, often failing to create a consistent deterrent and thus leaving platforms with ample room for opportunistic behavior.
In view of inherent limitations of bilateral models, a stream of forward-looking research has begun to advocate for multi-stakeholder governance, particularly by activating third-party forces to compensate for regulatory shortfalls. The academic community is beginning to recognize that platform governance requires not only top-down regulation, but also bottom-up feedback and checks [39,40]. Nevertheless, these studies predominantly rely on a binary framework or treat third parties as auxiliary. They struggle to explain how to effectively internalize value rationality into the platform’s payoff matrix when regulatory costs are prohibitive and the technical opacity remains high.

2.3. Bottom-Up Supervision and Synergistic Governance

As the direct bearers of algorithmic impact, consumers represent the demand side of value rationality. In traditional algorithm governance research, users were often cast as passive data providers. However, the recent literature increasingly acknowledges the awakening of user agency, viewing them as active “market monitors” [41]. This emphasis also resonates with international discussions of algorithmic accountability, which treat affected users and civil society as essential actors in detecting harms and pressuring platforms to adopt accountable practices [14]. Actions such as platform migration, public criticism, social media exposure, and formal complaints pose a substantial threat to a platform’s reputation and long-term revenue, making them an effective counterbalance to the abuse of statistical rationality [42]. These user behaviors are essentially a strategic choice between active supervision and passive acceptance, a choice that depends on a rational calculation of expected costs and benefits [43]. From the perspective of platform economics, consumer participation and trust are core to platform value creation in two-/multi-sided markets. Thus, consumer exit, voice, and feedback can operate as powerful “market constraints” that reshape platform incentives [24,25,26,27].
While existing research has separately examined the strategic dilemmas of platforms, the limitations of regulation, and the potential role of consumers are often isolated. Few studies have placed these three actors within a dynamic framework to explore how to reconcile statistical and value rationality. Specifically, they have failed to systematically model the synergistic effect between government regulation and consumer supervision [44]. Key questions remain unanswered: How does an efficient consumer feedback mechanism reduce regulatory costs? Conversely, how do government penalties incentivize more consumers to engage in active supervision [45]? International governance scholarship suggests that such synergy is central to closing accountability gaps under opacity: regulatory capacity and bottom-up reporting can be complementary when neither is sufficient alone [35,36]. This reciprocal governance gain is key to breaking the involution stalemate. Therefore, this paper introduces a tripartite evolutionary game model to mathematically demonstrate how the synergy of hard regulation and soft public supervision can restructure the platform’s decision-making logic, guiding the system from the dominance of statistical rationality toward a new equilibrium of value co-creation.

3. Problem Description and Model Assumptions

3.1. Problem Description

As is shown in Figure 1, this study formulates a tripartite evolutionary game framework comprising e-commerce platforms, regulatory agencies, and consumer groups. Under the principle of bounded rationality, each party dynamically adjusts its strategy to maximize expected payoffs.
Figure 1. Tripartite evolutionary interactions in e-commerce algorithm governance.

3.1.1. E-Commerce Platform

The e-commerce platform, as the algorithm’s architect, operates to maximize its profit by choosing between “algorithm involution” and “algorithm for good”. An involution strategy involves deploying extractive algorithms (e.g., price discrimination) to gain short-term excess profits, but this comes with the risk of severe administrative penalties and reputational damage if discovered. Conversely, a pro-social strategy requires upfront investment in compliance and R&D, forgoing short-term gains in exchange for long-term consumer trust and complete risk mitigation. From an evolutionary perspective, the platform’s choice is not a static calculation but a dynamic adjustment, where it continuously weighs the expected net benefits of non-compliance against the long-term net returns of compliance based on the perceived actions of regulators and consumers.

3.1.2. Government

The government, representing the public interest, aims to maximize social welfare through either “strict regulation” or “lax regulation”. Strict regulation involves significant resource investment in proactive oversight, which effectively curbs misconduct and enhances public trust. This effect is amplified by a synergistic collaboration with consumers, which lowers discovery costs and increases enforcement efficiency. Lax regulation, while saving administrative expenses, creates a risk of significant negative social externalities and loss of credibility. Thus, the regulator’s strategic evolution depends on the assessment of enforcement costs, synergistic efficiency, and the potential societal damages from misconduct.

3.1.3. Consumer

The consumer group, as the direct recipient of algorithmic impact, makes decisions based on utility maximization, choosing between “active feedback” and “passive acceptance”. Active feedback entails costs in time and effort, but yields benefits by mitigating direct utility loss and potentially securing compensation, especially under active regulation. Passive acceptance avoids these upfront costs but exposes consumers entirely to the risk of exploitation, leading to the potential loss of their consumer surplus. The evolution of consumer strategy is thus highly dependent on the perceived success rate and net benefits of recourse, which are directly coupled with the regulator’s enforcement posture—a more active regulator increases the expected payoff for consumer feedback.

3.2. Model Assumptions

To ensure analytical rigor and internal consistency, the game model is constructed based on the following assumptions. The strategic behaviors in the evolutionary game model are elaborated as follows, with the evolutionary game tree illustrated in Figure 2 and the related notation defined in Table 1.
Figure 2. Evolutionary game tree.
Table 1. Parameters and definitions.
Assumption 1. 
The game involves three core actors: the e-commerce platform, the government, and consumer. Each actor chooses from a discrete and finite strategy set. The platform selects either “algorithm involution” or “algorithm for good”. Government weights between “strict regulation” and “lax regulation”. Consumer chooses between “active feedback” and “passive acceptance”. No actor can adopt two pure strategies simultaneously.
Assumption 2. 
All actors exhibit bounded rationality. They cannot identify the optimal strategy through a single decision at the outset. Instead, they adjust their choices gradually through imitation, learning, and trial and error. The strategic dynamics follow an evolutionary process in which actors shift toward strategies yielding higher payoffs. The system may eventually converge to an ESS through repeated interactions.
Assumption 3. 
The platform’s behavior—especially actions consistent with algorithm involution—is not fully observable to regulators or consumers at zero cost. Misconduct can only be detected with a probability that depends on both parties’ strategies. In particular, strict regulation by government and active feedback by consumers generate the synergistic effect, increasing the likelihood that involutional behaviors are uncovered.
Assumption 4. 
All parameters in the model (e.g., costs, benefits, losses, and penalties) are exogenously given positive constants. For each actor, the final payoff under any strategy profile equals the sum of benefits minus the sum of costs and losses. All payoff components are assumed to be linearly additive.
Assumption 5. 
Each actor is modeled as a large population. The proportion adopting a particular strategy (e.g., the share x of platforms choosing algorithm for good) represents the fraction of individuals in that group following that strategy. Mixed strategies are therefore allowed and interpreted as population-level distributions.
Assumption 6. 
Technological advancement increases the effectiveness of consumer oversight ( Δ ), while reducing the cost of information verification ( C r ). This reflects the dual benefits of digital governance: enhanced supervisory capability and reduced participation thresholds. In particular, the synergy gain  Δ  is only realized when the regulator actively responds to consumer feedback. Without strict regulation (i.e., under the lax regulation strategy), consumer reports cannot be effectively converted into enforcement capacity, and thus the detection probability remains at its baseline level  β .
Assumption 7. 
This definition implies that while the act of exposure (increased by  Δ ) is necessary for justice, the result (knowing one has been cheated) creates a specific psychological cost (sense of betrayal). This is a realistic trade-off in consumer psychology: “I want to know the truth, but the truth hurts”. However, this hurt is outweighed by the utility of active feedback ( R ) and incentives ( H ) in the active strategy.
In accordance with problem description and theoretic assumptions, this study establishes a three-party evolutionary game model. The resulting payoff matrix is shown in Table 2.
Table 2. Payoff matrix.

4. The Equilibrium Analysis in Three-Party Evolutionary Game Model

4.1. E-Commerce Platform

The expected return U 11 for choosing “algorithm for good”:
U 11 = π − C p + B p
The expected return U 12 for choosing “algorithm involution”:
U 12 = π + E p − L − F ⋅ ( β + y ( α − β ) + y z Δ )
Average expected return of the e-commerce platform:
U ¯ 1 = x U 11 + ( 1 − x ) U 12
Dynamic equation for the replication of the e-commerce platform:
F ( x ) = d x d t = x ( U 11 − U ¯ 1 ) = x ( 1 − x ) ( U 11 − U 12 )
U 11 − U 12 = ( π − C p + B p ) − [ π + E p − L − F ⋅ ( β + y ( α − β ) + y z Δ ) ] = B p − C p − E p + L + F ⋅ ( β + y ( α − β ) + y z Δ )
We can obtain:
F ( x ) = d x d t = x ( 1 − x ) [ B p − C p − E p + L + F ⋅ ( β + y ( α − β ) + y z Δ ) ]
In accordance with the stability theorem of differential equations, the strategy probability of platforms ( x ) constitutes the ESS if it satisfies F ( x ) = 0 and d F ( x ) d x < 0 . For d F ( x ) d x = ( 1 − 2 x ) ( B p − C p − E p + L + F ( β + y ( α − β ) + y z Δ ) ) , let J ( z ) = B p − C p − E p + L + F ( β + y ( α − β ) + y z Δ ) . Since d J ( z ) d z > 0 , J ( z ) is an increasing function of z . Let J ( z ) = 0 , we can obtain z * = − [ B p − C p − E p + L + F ( β + y ( α − β ) ) ] / y Δ F .
When z = z * = − [ B p − C p − E p + L + F ( β + y ( α − β ) ) ] / y Δ F , J ( z ) = 0 , and d F ( x ) d x ≡ 0 , the platforms cannot determine a stable strategy. When 0 < z < z * , J ( z ) < 0 , and d F ( x ) d x x = 0 < 0 , d F ( x ) d x x = 1 > 0 , x = 0 is the ESS for the e-commerce platform. Conversely, when z * < z < 1 , J ( z ) > 0 , and d F ( x ) d x x = 0 > 0 , d F ( x ) d x x = 1 < 0 , x = 1 is the ESS for e-commerce platform.

4.2. Government

The expected return U 21 for choosing “strict regulation”:
U 21 = x G − C g + ( 1 − x ) F ( α + z Δ ) + z W
The expected return U 22 for choosing “lax regulation”:
U 22 = x G + ( 1 − x ) ( β F − S )
Average expected return of the government:
U ¯ 2 = y U 21 + ( 1 − y ) U 22
Dynamic equation for the replication of the government:
F ( y ) = d y d t = y ( U 21 − U ¯ 2 ) = y ( 1 − y ) ( U 21 − U 22 )
U 21 − U 22 = [ x G − C g + ( 1 − x ) F ( α + z Δ ) + z W ] − [ x G + ( 1 − x ) ( β F − S ) ] = − C g + z W + ( 1 − x ) [ ( α − β + z Δ ) F + S ]
We can obtain:
F ( y ) = d y d t = y ( 1 − y ) [ − C g + z W + ( 1 − x ) ( ( α − β + z Δ ) F + S ) ]
When the probability of government’s strategy is in a stable state, it must satisfy F ( y ) = 0 and d F ( y ) d y < 0 . For d F ( y ) d y = ( 1 − 2 y ) [ − C g + z W + ( 1 − x ) ( ( α − β + z Δ ) F + S ) ] , let J ( x ) = − C g + z W + ( 1 − x ) ( ( α − β + z Δ ) F + S ) . Since d J ( x ) d x < 0 , J ( x ) is a decreasing function of x . Let J ( x ) = 0 , we can obtain x * = 1 − ( C g − z W ) / ( S + ( α − β + z Δ ) F ) .
When x = x * = 1 − ( C g − z W ) / ( S + ( α − β + z Δ ) F ) , J ( x ) = 0 , and d F ( y ) d y ≡ 0 , the governments cannot determine a stable strategy. When 0 < x < x * , J ( x ) > 0 , and d F ( y ) d y y = 0 > 0 , d F ( y ) d y y = 1 < 0 , y = 1 is the ESS for the government. Conversely, when x * < x < 1 , J ( x ) < 0 , and d F ( y ) d y y = 0 < 0 , d F ( y ) d y y = 1 > 0 , y = 0 is the ESS for governments.

4.3. Consumer

The expected benefit U 31 for choosing “active feedback”:
U 31 = R − C r + x B u − ( 1 − x ) E u + y H − ( 1 − x ) D ( y ( α + Δ ) + ( 1 − y ) β )
The expected return U 32 for choosing “passive acceptance”:
U 32 = x B u − ( 1 − x ) E u − ( 1 − x ) D ( β + y ( α − β ) )
The average expected return of consumers:
U ¯ 3 = z U 31 + ( 1 − z ) U 32
Dynamic equation for the replication of consumers:
F ( z ) = d z d t = z ( U 31 − U ¯ 3 ) = z ( 1 − z ) ( U 31 − U 32 )
U 31 − U 32 = R − C r + y [ H − ( 1 − x ) Δ D ]
We can obtain:
F ( z ) = d z d t = z ( 1 − z ) [ R − C r + y ( H − ( 1 − x ) Δ D ) ]
When the probability of platforms choosing differentiated strategy is in a stable state, it must satisfy F ( z ) = 0 and d F ( z ) d z < 0 . d F ( z ) d z = ( 1 − 2 z ) ( R − C r + y ( H − ( 1 − x ) Δ D ) ] . let J ( y ) = R − C r + y ( H − ( 1 − x ) Δ D ) . Since d J ( y ) d y > 0 , J ( y ) is an increasing function of y . Let J ( y ) = 0 , we can obtain y * = ( C r − R ) / ( H − ( 1 − x ) Δ D ) .
When y = y * = ( C r − R ) / ( H − ( 1 − x ) Δ D ) , J ( y ) = 0 , and d F ( z ) d z ≡ 0 , the consumers cannot determine a stable strategy. When 0 < y < y * , J ( y ) < 0 , and d F ( z ) d z z = 0 < 0 , d F ( z ) d z z = 1 > 0 , z = 0 is the ESS for consumers. Conversely, when y * < y < 1 , J ( y ) > 0 , and d F ( z ) d z z = 0 > 0 , d F ( z ) d z z = 1 < 0 , z = 1 is the ESS for consumers.

5. Evolutionary Stability Analysis

After deriving the partial derivatives of the replicator dynamic equations, the Jacobian matrix corresponding to the three-party evolutionary system is established. Each element of the matrix represents the marginal effect of one actor’s strategy proportion on the evolutionary payoff of another actor, thereby capturing the interdependent adjustment mechanisms among platforms, regulators and consumers. By evaluating this matrix at specific equilibrium points, we obtain the system’s eigenvalues which serve as the criteria for determining local stability.
J = ∂ F ( x ) ∂ x ∂ F ( x ) ∂ y ∂ F ( x ) ∂ z ∂ F ( y ) ∂ x ∂ F ( y ) ∂ y ∂ F ( y ) ∂ z ∂ F ( z ) ∂ x ∂ F ( z ) ∂ y ∂ F ( z ) ∂ z = J 11 J 12 J 13 J 21 J 22 J 23 J 31 J 32 J 33
The specific values in the Jacobian matrix are as follows:
J 11 = ( 1 − 2 x ) B p − C p − E p + L + β F + y ( α − β ) F + y z Δ F J 12 = x ( 1 − x ) [ ( α − β ) F + z Δ F ] J 13 = x ( 1 − x ) y Δ F J 21 = − y ( 1 − y ) ( ( α − β ) F + z Δ F + S ) J 22 = ( 1 − 2 y ) z W − C g + ( 1 − x ) ( α − β ) F + z Δ F + S J 23 = y ( 1 − y ) [ W + ( 1 − x ) Δ F ] J 31 = z ( 1 − z ) y Δ D J 32 = z ( 1 − z ) [ H − ( 1 − x ) Δ D ] J 33 = ( 1 − 2 z ) R − C r + y H − y ( 1 − x ) Δ D
Table 3 presents the eigenvalue structures and the associated stability conditions of the eight corner equilibria. These equilibria correspond to distinct combinations of platform algorithmic behavior, regulation enforcement, and consumer supervision. Their stability depends on whether the eigenvalues are negative (ESS), mixed in sign (saddle point), or contain at least one positive value (unstable equilibrium).
Table 3. Stability analysis of system equilibrium point.
Overall, the stability patterns across the equilibria reveal that the system’s evolutionary trajectory is highly sensitive to the relative payoffs and costs of responsible algorithm adoption, regulatory enforcement, and consumer participation. Only when the incentive structures of three actors become complementary can the system escape the algorithmic involution trap and sustainably converge toward welfare-enhancing “algorithm for good”.
Most of equilibrium points represent conditional stability, meaning that stability is achieved only under specific parameter configurations. For instance, the equilibrium in which all three actors adopt passive strategies (E1) becomes stable only when the platform lacks incentives to invest in responsible algorithms ( K + β F < 0 ), consumers receive negative net benefits from supervision ( R − C r < 0 ), and the government finds that regulatory costs exceed the revenue from fines and subsidies ( S + ( α − β ) F − C g < 0 ), trapping the system in a persistent vicious cycle. Similarly, the equilibrium where only consumers remain actively engaged (E4) becomes conditionally stable when the government finds the sum of reputation benefits and fines insufficient to cover regulatory costs, resulting in a state of “consumer-driven but ineffective oversight”.
Other equilibria describe alternative governance configurations. Unlike static assumptions, the state where government regulators and platforms cooperate while consumers remain inactive (E5) is identified as an unstable point. Without consumer participation to generate reputational rewards ( W ), the government lacks the incentive to maintain costly regulation ( C g ) once platforms are compliant. In contrast, when platforms and consumers form mutually reinforcing incentives but regulatory agencies withdraw (E6), the system gravitates toward a market self-governance state. Another equilibrium (E7) captures situations in which both regulators and consumers exert pressure on platforms, but platforms remain resistant, leading to a stable yet undesirable scenario of “strong regulation but platform non-compliance”.
When all three actors’ incentives are aligned—platforms gain from adopting responsible algorithms, regulatory enforcement becomes cost-effective ( C g < W ), and consumer supervision generates positive net returns—the system converges to the ideal equilibrium (E8), representing a stable and mutually reinforcing tripartite collaborative governance regime. This state reflects the long-term outcome policymakers seek to achieve in algorithm governance.

6. Numerical Simulation of Evolutionary Games

This study deploys MATLAB R2025a to implement a series of numerical simulations, which are designed to analyze the evolutionary trajectories of system equilibria, the influence of initial strategies on outcomes, and the sensitivity of key parameters. Therefore, parameter values were assigned based on a synthesis of industry reports, case studies, and expert consultation. Importantly, these assigned values are intended to reveal the system’s intrinsic relationships and dynamic trends. Their absolute magnitude does not affect the model’s practical applicability.
The parameter assignments are grounded in the commercial logic and data relationships of the contemporary e-commerce sector. For the parameter representing the extra gains from algorithmic involution ( E p ), we reference the “big data discriminatory pricing” observed in the Online Travel Agency (OTA) sector. Empirical cases reveal that for an identical hotel room, a platform might charge a loyal user CNY 450 while offering it to a new user for CNY 380. This 18% price differential represents the specific algorithmic rent extracted from consumer surplus, justifying a significant positive value for E p in the simulation. In this evolutionary game, the platform’s base revenue ( π ) denotes the transactional revenue of the specific business segment governed by the algorithm (e.g., the specific hotel booking module utilizing discriminatory pricing), not the conglomerate’s total annual revenue. In reality, antitrust fines are often calculated based on the group’s total sales. Consequently, a fine that appears small globally (4% of total revenue) can be astronomically high relative to the specific revenue stream of the infringing business unit. The discovery probability under active governance ( α ) is set significantly higher than under passive governance ( β ). This difference reflects the substantial efficacy gap in detecting covert algorithmic misconduct, such as price discrimination. Furthermore, the synergistic gain from consumer participation ( Δ ) underscores the critical role of consumer feedback in enhancing regulatory precision.
To further clarify the empirical logic for detection probabilities, the assignment of α , β , and Δ captures the structural asymmetry in algorithmic governance. Specifically, β (passive discovery) is calibrated to a low baseline to reflect the “black box” nature of algorithms, where misconduct is rarely self-evident without external cues. In contrast, α (active discovery) is set at a higher tier to model the efficacy of dedicated technical audits, though complete transparency remains theoretically unattainable due to code complexity. The parameter Δ is defined as the marginal gain from “crowdsourced monitoring,” representing how consumer complaints pinpoint specific violations, thereby converting a low-probability search into a high-probability verification. Regarding the sensitivity analysis, the parameter ranges were deliberately selected to span the full spectrum from “nominal supervision” (low α , low Δ ) to “stringent enforcement” (high α , high Δ ). This wide variation is essential to verify the robustness of the equilibrium results, ensuring that the model’s conclusions hold valid across different stages of regulatory maturity and are not artifacts of specific point estimates.
The parameters ensure the economic rationality of active participation for all parties. For e-commerce platform, the cost of implementing algorithm for good ( C p ) is lower than the aggregate benefits of long-term consumer loyalty ( B P ), avoided fines ( F ), and avoided consumer churn ( L ). For the government, governance costs ( C g ) are lower than the societal benefits of a fair market ( G ) and the reputational gains from effective regulation ( W ). For the consumer, while the information screening cost ( C r ) is equivalent to the potential utility loss ( E u ), the active feedback strategy becomes profitable due to the inclusion of enhanced utility ( B u ) and positive incentives ( H ), which collectively outweigh the participation costs.
To illustrate the system’s evolutionary dynamics and establish a baseline for policy analysis, the initial model parameters were configured to satisfy the stability requirements of the ESS. The specific values are set as follows: α = 0.5, β = 0.2, Δ = 0.3, π = 10, B p = 3, C p = 5, E p = 4, L = 4, F = 6, C g = 4, G = 8, S = 6, W = 5, R = 2, C r = 4, B u = 3, E u = 4, H = 3, D = 3. It should be noted that while the specific numerical values are normalized for simulation purposes, they are chosen to preserve the relative magnitude relationships observed in the aforementioned empirical cases and literature, ensuring the robustness of the evolutionary trends.

6.1. ESS Analysis

When the constraints are satisfied K + F ( α + Δ ) > 0 , C g − W < 0 , R − C r + H > 0 , the system converges to the ESS at the point E8(1,1,1). In this stable equilibrium, the e-commerce platform commits to “algorithm for good”, the government maintains “strict regulation”, and consumers engage in “active feedback”. Such strategic alignment is fundamental to cultivate a robust algorithmic governance ecosystem within the digital economy.
The dynamic evolution of this system is illustrated in Figure 3a, which demonstrates that the probability of strategy selection for the platform ( x ), government ( y ), and consumer ( z ) converge to 1 over time. Furthermore, Figure 3b highlights the temporal disparities in reaching this equilibrium. The government stabilizes most rapidly, acting as the primary driver of the system. The e-commerce platform follows with a moderate convergence rate, whereas consumers exhibit the slowest adaptation speed. This sequence suggests that top-down strict regulation is the most immediate force, while platform technical compliance and the cultivation of consumer awareness are lagging processes. Consequently, the government’s decisive action is crucial to trigger the evolutionary chain, sending strong signals that subsequently compel platforms to adjust algorithms and encourage consumers to participate.
Figure 3. Evolutionary path to E8(1,1,1). (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths.
This analysis further underscores the foundational conditions for the optimal evolution of algorithm governance. A stable strategy can be formed, predicated on reasonable parameters and the leading role of government regulation, which triggers the subsequent synergy from platforms and consumers. While consumer commitment to active feedback is the final piece of the puzzle, strict regulation acts as the catalyst that facilitates the sustainable optimization of the algorithmic ecosystem, guiding its transition from “algorithm involution” to “algorithm for good”.

6.2. Initial Probability Analysis

6.2.1. The Influence of x on y and z

We evaluated the system’s response to the e-commerce platform’s initial willingness to adopt “algorithm for good” at values of 0.2, 0.5, and 0.85. Figure 4a indicates that higher initial goodwill expedites the convergence to the ESS. Regarding the government ( y ), an inverse relationship in convergence speed is observed in Figure 4b, reflecting a strategic adjustment in governance intensity. When the e-commerce platform’s autonomous goodwill motivation is strong, the government slows down its intervention to moderately optimize the input of regulatory resources, relying more on market self-discipline avoiding resource misallocation caused by excessive intervention. Conversely, when the e-commerce platform’s initial goodwill motivation is insufficient, the government accelerates active governance to force the e-commerce platform to adjust its strategy, promoting the system to evolve toward a benign equilibrium. Regarding the consumer ( z ), the convergence curves are relatively close, but Figure 4c shows that lower platform goodwill slightly accelerates consumer active feedback. This suggests that algorithmic risks and insufficient platform responsibility are more likely to trigger user vigilance and rights protection behavior than high goodwill.
Figure 4. The influence of algorithm for good probability on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths of x on y. (c) Two-dimensional evolutionary paths of x on z.

6.2.2. The Influence of y on x and z

The government’s initial propensity for “strict regulation” was analyzed at levels of 0.2, 0.5, and 0.85. Figure 5a shows that a more proactive government accelerates the path to the stable equilibrium. As the initial willingness of the government to adopt active governance increases, the faster the evolution to the E8(1,1,1). From Figure 5b, regulatory pressure has a significant impact on the platform’s decision-making ( x ), indicating that high-intensity regulation effectively pushes platforms toward algorithm goodwill. In Figure 5c, strict regulation also positively influences consumer participation in feedback mechanisms ( z ), where the gap between the red and blue curves further illustrates that government leadership is a prerequisite for activating consumer power.
Figure 5. The influence of strict regulation probability on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths of y on x. (c) Two-dimensional evolutionary paths of y on z.
It is noteworthy that the government’s strict regulation positively correlates with the strategy evolution of e-commerce platforms and consumers. This indicates that strong government supervision and guidance can effectively drive the digital ecosystem toward a benign equilibrium. When the government increases strict governance efforts, it sends a clear signal of rule enforcement, which not only prompts e-commerce platforms to prioritize long-term goodwill over short-term involution but also empowers consumers to engage in active feedback behaviors (such as information screening and rights protection) with greater confidence. Conversely, lax regulation willingness may lead to delayed evolution or even deviation from the optimal equilibrium, highlighting the government’s pivotal role in shaping the algorithmic governance landscape.

6.2.3. The Influence of z on x and y

Consumer willingness for “active feedback” was set at 0.2, 0.5, and 0.85. Figure 6a demonstrates that higher initial consumer activity speeds up the overall evolution to E8(1,1,1). As shown in Figure 6b, active feedback has a profound impact on the e-commerce platform ( x ), demonstrating that strong consumer pressure efficiently motivates platforms to adopt algorithm for good. In Figure 6c, high consumer engagement provides significant impetus for the government to sustain strict regulation ( y ).
Figure 6. The influence of active feedback probability on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths of z on x. (c) Two-dimensional evolutionary paths of z on y.
It is noteworthy that consumers’ willingness of active feedback positively correlates with the strategy evolution of e-commerce platforms and governments. This confirms that consumers’ active participation in vigilant decision-making and rights protection can effectively drive the digital ecosystem toward a benign equilibrium. When consumers show higher active feedback, they form a market constraint force that compels e-commerce platforms to abandon algorithmic involution and pursue long-term goodwill. Meanwhile, this collective vigilant behavior also signals to the government that social supervision is effective, thereby encouraging the government to optimize active governance strategies. Conversely, low consumer rationality may lead to delayed evolution or even deviation from the optimal equilibrium, highlighting the pivotal role of consumers as “market supervisors” in shaping the algorithmic regulation landscape.

6.3. Parameter Sensitivity Analysis

6.3.1. Sensitivity Analysis of F

We examined the impact of penalty intensity ( F ) on the tripartite game by simulating values of 4, 6, and 8. Figure 7a depicts the evolutionary trajectories under these varying intensities. Figure 7b illustrates that with the increase in parameter penalty intensity ( F ), the system’s convergence speed accelerates significantly. Crucially, the e-commerce platform ( x ) exhibits the highest sensitivity to this parameter. The blue curve ( F = 4 ) for the platform lags significantly, showing a slow adaptation rate. However, as F increases to 8, the platform is strongly incentivized to adopt algorithm goodwill to avoid high fines, resulting in a steep and rapid convergence to 1.
Figure 7. The influence of F on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths.
The government ( y ) and consumers ( z ) are also more willing to make active choices under higher penalties, but their response curves are more clustered compared to the distinct separation seen in the platform’s curves. This suggests that penalty intensity ( F ) is a decisive factor primarily for shaping platform behavior. The higher and appropriate penalties can effectively force the e-commerce platform to abandon involution, which subsequently aligns the interests of the government and consumers, promoting the system to evolve toward a benign equilibrium more rapidly. Therefore, formulating targeted penalty policies based on e-commerce platform behavior characteristics can significantly enhance the efficacy of algorithmic governance.

6.3.2. Sensitivity Analysis of α

To gauge the effect of regulatory oversight capabilities, the government’s basic detection probability ( α ) was set at 0.4, 0.5, and 0.6. Figure 8a shows the evolutionary paths of the e-commerce platform, government, and consumers under the influence of different α . Figure 8b demonstrates that enhancing α strengthens the government’s ability to identify algorithmic involution. Although the overall trajectories appear clustered, the zoomed-in inset reveals a clear positive correlation. The higher detection probabilities ( α = 0.6 ) lead to a faster convergence compared to lower probabilities ( α = 0.4 ). This heightened detection risk compels platforms to embrace goodwill strategies, while improved regulatory effectiveness boosts consumer confidence in providing feedback. Correspondingly, the evolution speed toward the stable strategy E8(1,1,1) accelerates. The results identify α as a pivotal metric for regulatory efficiency. A higher detection probability creates a robust deterrence mechanism, aligning stakeholder interests and expediting the transition to the benign equilibrium. Therefore, improving the government’s detection capabilities can significantly enhance the efficacy of algorithmic governance.
Figure 8. The influence of α on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths.

6.3.3. Sensitivity Analysis of β

To gauge the effect of regulatory oversight capabilities, the government’s detection probability under the lax regulation β (reflecting minimum regulatory effectiveness) was set at 0.1, 0.2, and 0.3. Figure 9a shows the evolutionary paths of the e-commerce platform, government, and consumers under the influence of different α . Figure 9b demonstrates that, unlike the significant impact of strict regulation ( α ), the evolutionary trajectories under varying β values are highly clustered with minimal separation. As revealed by the zoomed-in inset, increasing β from 0.1 to 0.3 produces almost negligible variation in the convergence speed toward the stable strategy E8(1,1,1). This indicates that the system is relatively insensitive to the detection probability under lax regulation. Since the evolutionary target is the ideal state dominated by strict regulation, the impact of β which functions when y = 0 naturally diminishes as the government strategy evolves. Therefore, while β provides a necessary compliance baseline, merely improving the efficiency of lax supervision is insufficient to drive the system significantly faster toward the benign equilibrium. The core leverage lies in establishing strict regulatory deterrence.
Figure 9. The influence of β on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths.

6.3.4. Sensitivity Analysis of W

The influence of the government’s additional reputation benefit ( W ), obtained when implementing active governance alongside consumers’ active feedback choices on participants is analyzed. Figure 10a shows the evolutionary paths under different W values (3, 5, 7). Figure 10b illustrates that with the increase in parameter W , the government’s willingness to implement active governance is significantly enhanced (driven by higher reputation returns). The government’s curve exhibits the most distinct separation, where the curve ( W = 7 ) converges much faster than the curve ( W = 3 ). This proactive governmental stance further motivates the e-commerce platform to choose algorithmic goodwill and consumers to engage in active feedback strategies. Correspondingly, the evolution speed toward the stable strategy E8(1,1,1) accelerates.
Figure 10. The influence of W on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths.
This result suggests that W serves as a key incentive factor for the government’s active governance behavior. The higher W can effectively strengthen the government’s motivation to participate in algorithmic governance, thereby driving the tripartite strategy system to evolve toward a benign equilibrium more rapidly. Therefore, measures that enhance the government’s reputation gains from collaborative governance can help optimize W and improve the efficacy of algorithmic governance.

6.3.5. Sensitivity Analysis of H

We analyzed the influence of consumers’ additional positive incentive H , obtained when the government implements strict regulation alongside consumers’ active feedback. Figure 11a shows the evolutionary paths under different H values (2, 3, 4). Figure 11b illustrates that with the increase in parameter H , consumers’ willingness to make active feedback is significantly enhanced (driven by stronger positive incentives from collaborative governance). This impact is visually pronounced in the consumer trajectories, where a substantial gap exists between the slow-moving curve ( H = 2 ) and the rapidly converging curve ( H = 4 ). This increased consumer participation further motivates the e-commerce platform to prefer algorithmic goodwill and the government to engage in active governance more proactively. Correspondingly, the evolution speed toward the stable strategy E8(1,1,1) accelerates.
Figure 11. The influence of H on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths.
This result suggests that H serves as a key driving factor specifically for consumers’ active feedback in algorithmic governance. The higher H can effectively strengthen consumers’ motivation to engage in vigilant behaviors (such as information screening and rights protection), thereby pushing the tripartite strategy system to evolve toward a benign equilibrium more rapidly. Therefore, measures that enhance consumers’ positive incentives can help optimize H and improve the efficacy of algorithmic governance by eliminating the lag in consumer participation shown in the low-incentive scenarios.

6.3.6. Sensitivity Analysis of E p / B p

The section explores the platform’s strategic trade-off by analyzing the system’s sensitivity to the ratio of the platform’s short-term involution benefit ( E p ) and long-term goodwill benefit ( B p ). The ratio was simulated at values of 0.67, 1.33, and 1.67. Figure 12a shows the evolutionary paths under the influence of different E p / B p ratios. Figure 12b illustrates that with the increase in E p / B p , the motivation for e-commerce platform to adopt goodwill strategies diminishes. This inverse relationship is clearly visualized by the gap between the curves. The curves ( E p / B p = 0.67 ) representing a focus on long-term benefits converge rapidly to the stable state, whereas the curves ( E p / B p = 1.67 ) representing a dominance of short-term interests lag significantly behind. This shift dampens both consumer feedback and government intervention, causing the overall convergence to E8 to slow significantly
Figure 12. The influence of E p / B p on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths.
This result suggests that E p / B p is a determinant factor in shaping the platform’s strategic trade-off between short-term gains and long-term development. The lower E p / B p can effectively incentivize the platform to abandon involution, align the interests of the government and consumers, and promote the system to evolve toward a benign equilibrium more rapidly. Therefore, policies that reduce the platform’s short-term involution benefits or enhance its long-term goodwill benefits can help optimize E p / B p and improve the efficacy of algorithmic governance.

6.3.7. Sensitivity Analysis of Δ and C r

The study analyzed the combined influence of consumers’ synergy gain Δ (the additional enhancement of consumers’ active feedback) and information screening cost C r (the cost of consumer active feedback), setting different combination levels (e.g., Δ = 0.1, C r = 4, low), ( Δ = 0.3, C r = 3, medium), ( Δ = 0.5, C r = 2, high). Figure 13a shows the evolutionary paths of the e-commerce platform, government, and consumers under different combinations. Figure 13b illustrates that with the increase in Δ increases and reduction in C r , consumers are more willing to choose active feedback, which in turn motivates the e-commerce platform to prefer algorithmic goodwill and the government to engage in strict regulation more proactively. This acceleration is most prominently reflected in the consumer trajectories, where the high-incentive combination represented by the curve ( Δ = 0.5, C r = 2, high) converges to the stable state far more rapidly than the lagging low-incentive combination represented by the curve ( Δ = 0.1, C r = 4, low). As shown in the zoomed-in view, optimizing these parameters not only speeds up individual strategies but also minimizes the evolutionary lag between the three parties.
Figure 13. The influence of Δ and C r on system evolutionary outcome. (a) Three-dimensional evolutionary paths. (b) Two-dimensional evolutionary paths.
This result suggests that increasing (e.g., through consumer education, building collaborative supervision platforms) and reducing (e.g., via government-provided transparency tools, platform-driven simplified information presentation) can effectively incentivize consumers to participate vigilantly, pushing the tripartite strategies to evolve toward a benign equilibrium more rapidly. Therefore, policies and measures that optimize this combination can significantly enhance the efficiency of algorithmic governance.

7. Conclusions and Implications

7.1. Conclusions

This study constructs a tripartite evolutionary game model involving e-commerce platform, government and consumer to explore the dynamic evolution of algorithmic governance. The theoretical contribution lies in mathematically demonstrating how to reconcile the paradox between the platform’s pursuit of statistical rationality and market governance’s need for value rationality.
First, government–public collaboration serves as the core catalyst for bridging the rationality gap, accelerating the system toward the ideal state of “good governance”. Although unilateral government regulation imposes external constraints on statistical rationality, it is costly, reactive, and struggles to penetrate the algorithmic black box. Conversely, consumer resistance represents a demand for value rationality, but lacks sustainability without official backing. Therefore, a synergistic mechanism is essential. Specifically, increasing the government’s reputational gain ( W ) and consumer incentives ( H ) creates a positive resonance. Simulation results confirm that these parameters are highly actor-specific, with W primarily driving the government’s proactivity and H serving as the decisive trigger for consumer participation. This accelerates the system’s convergence toward the ideal state E8(1,1,1), proving that synergy is not merely additive but multiplicative in resolving the governance impasse.
Second, this tripartite framework offers distinct advantages over other multi-actor models because consumer participation introduces a dynamic variable ( Δ ) that scales with the platform’s market reach. Unlike third-party auditors who provide periodic checks, consumers act as a distributed, real-time sensor network. Our findings show that the combination of high synergy gains ( Δ ) and low screening costs ( C r ) creates a “tipping point” for consumer behavior, converting dispersed user experiences into actionable regulatory intelligence. This mechanism creates an “incentive-compatible” loop where the direct victims of algorithmic involution become its most effective monitors, overcoming information asymmetry at a lower social cost.
Third, a platform’s strategic choice is essentially a rational calibration between short-term extractive gains (statistical rationality) and long-term benevolent returns (value rationality). Sensitivity analysis of the E p / B p ratio reveals a stark inverse relationship, whereby a lower ratio signals a strategic shift toward long-term trust and coincides with a markedly faster rate of system equilibrium attainment. Interestingly, the platform’s initial goodwill also shows a strategic substitution effect with other actors. When platform autonomy is high, government regulatory pressure naturally optimizes and eases, whereas low platform goodwill triggers rapid, forced governance.
Finally, hard constraints constitute the necessary foundation for governance, while the soft environment determines the efficiency ceiling. Although detect improvements in lax supervision ( β ) yield negligible gains, high-intensity deterrence—combining severe penalties ( F ) with strict detection ( α )—serves as the pivotal lever for curbing the excesses of statistical rationality. However, once this foundation is laid, the soft environment (e.g., synergistic benefits, user participation costs)—characterized by high synergistic gains ( Δ ) and low information screening costs ( C r )—becomes decisive. When consumers can exercise high-impact supervision at a low cost, platforms face dual pressure from both regulators and the market, forcing them to accelerate the abandonment of involutionary strategies and embrace a sustainable equilibrium.

7.2. Implications

Resolving the dilemma of algorithmic involution in e-commerce is a systemic challenge that requires the reconstruction of interests and the design of new mechanisms. Based on our simulation findings, the cost–benefit structure of platform decision-making can be fundamentally altered only through a combination of strict government regulation, strategic corporate transformation, and collaborative consumer participation. The following practical implications are proposed to guide the e-commerce ecosystem from a zero-sum “involution trap” toward a sustainable “win–win” equilibrium.
First, the government must transition from a passive regulator to a meta-governor that orchestrates both deterrence and collaboration. Our results show that platforms are highly sensitive to financial penalties and detection risks. Therefore, regulators should move beyond static fines and implement dynamic penalty mechanisms anchored to the platform’s excess profits derived from algorithmic misconduct. This ensures that the cost of violation always exceeds the potential illegal gains. Simultaneously, to address the high costs of government monitoring, authorities should invest in regulatory technology infrastructure. By establishing real-time interfaces with platform data, regulators can replace laggy, ex-post punishments with real-time risk warnings. Furthermore, the government should lower the threshold for public supervision by establishing incentivized whistleblower programs and verified complaint channels. These measures effectively convert dispersed consumer complaints into precise leads for law enforcement, creating a closed-loop governance system.
Second, e-commerce platforms must pivot their strategic focus from short-term traffic monetization to long-term reputation management. The study reveals that when platforms prioritize immediate statistical metrics over long-term value, the entire ecosystem deteriorates rapidly. To reverse this, platforms need to reconstruct their internal assessment systems. Metrics reflecting consumer trust, retention rates, and social responsibility should be given higher weight than short-term growth indicators in algorithm optimization objectives. Proactive transparency is also a rational competitive strategy. By voluntarily disclosing the basic logic of recommendation algorithms and establishing external ethics committees, platforms can reduce user suspicion and regulatory pressure. Ultimately, treating “algorithm for good” not as a burden but as a distinctive competitive barrier will allow platforms to accumulate trust capital, securing sustainable advantages in an increasingly compliance-driven market.
Third, consumer agency must be fully activated to serve as the “distributed sensors” of the digital market. Our findings suggest that consumer participation is most effective when the effort required is low and the impact is high. Therefore, the current complex mechanisms for reporting algorithm abuse must be simplified. We advocate for the implementation of “one-click” feedback features embedded directly within user interfaces, allowing consumers to easily flag irrational recommendations or price discrimination. Moreover, the effectiveness of consumer feedback must be visible. Establishing a transparent feedback–response mechanism—where platforms publicly report the handling of user complaints—is crucial. When consumers see that their actions lead to tangible improvements or rewards, their role evolves from passive data providers to active governance participants, forming a powerful market constraint force that compels platforms to return to the essence of service.

7.3. Limitations and Future Research

First, the assumption of linear payoff functions and exogenous parameters, while ensuring tractability, may overlook non-linear dynamics such as threshold effects or diminishing marginal returns. Second, modeling homogeneous populations abstracts from real-world heterogeneity in platform capabilities and consumer digital literacy; thus, the results capture average tendencies rather than group-specific dynamics. Third, as the parameters are numerically calibrated rather than structurally estimated from micro-data, the simulations provide qualitative mechanisms rather than precise quantitative forecasts. Future research should address these constraints by endogenizing key variables, introducing agent heterogeneity and network topologies, and utilizing platform-level empirical data for causal validation to better reflect the complexity of algorithmic governance.

Author Contributions

Conceptualization, Y.L. and D.L.; methodology, Y.L.; software, L.W.; validation, B.C.; formal analysis, D.L.; investigation, B.C.; resources, D.L.; data curation, B.C.; writing—original draft preparation, Y.L. and L.W.; writing—review and editing, Y.L. and L.W.; visualization, B.C.; supervision, D.L.; project administration, D.L.; funding acquisition, Y.L. and L.W. All authors have read and agreed to the published version of the manuscript.

Funding

Zhejiang Office of Philosophy and Social Science (24SSHZ002YB), Zhejiang Province Philosophy and Social Sciences Planning Post-Funding Project (25HQZZ014Z), Pioneer and Leading Goose R&D Program of Zhejiang (2024C03268), Scientific Research Fund of Zhejiang Provincial Education Department (Y202351690), Scientific Research Fund of Hangzhou Dianzi University Information Engineering College (KYP0224001).

Institutional Review Board Statement

Not applicable.

Data Availability Statement

No data was used for the research described in the article.

Acknowledgments

The authors thank the anonymous referees for their invaluable comments on an earlier version of the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Weber, M. Economy and Society: An Outline of Interpretive Sociology; University of California Press: Berkeley, CA, USA, 1978. [Google Scholar]
  2. Kellogg, K.C.; Valentine, M.A.; Christin, A. Algorithms at work: The new contested terrain of control. Acad. Manag. Ann. 2020, 14, 366–410. [Google Scholar] [CrossRef] [Scilit]
  3. Benlian, A.; Wiener, M.; Cram, W.A.; Krasnova, H.; Maedche, A.; Möhlmann, M.; Recker, J.; Remus, U. Algorithmic management: Bright and dark sides, practical implications, and research opportunities. Bus. Inf. Syst. Eng. 2022, 64, 825–839. [Google Scholar] [CrossRef] [Scilit]
  4. Geertz, C. Agricultural Involution: The Processes of Ecological Change in Indonesia; University of California Press: Berkeley, CA, USA, 1963. [Google Scholar]
  5. Huang, P.C. The Peasant Family and Rural Development in the Yangzi Delta, 1350–1988; Stanford University Press: Redwood City, CA, USA, 1990. [Google Scholar]
  6. Dubé, J.P.; Misra, S. Personalized pricing and consumer welfare. J. Political Econ. 2023, 131, 131–189. [Google Scholar] [CrossRef] [Scilit]
  7. Han, G.; Feng, Z.; Chen, S.; Xue, X.; Wu, H. Evaluating differential pricing in e-commerce from the perspective of utility. Electron. Commer. Res. Appl. 2024, 64, 101373. [Google Scholar] [CrossRef] [Scilit]
  8. Durán, J.M.; Jongsma, K.R. Who is afraid of black box algorithms? On the epistemological and ethical basis of trust in medical AI. J. Med. Ethics 2021, 47, 329–335. [Google Scholar] [CrossRef] [Scilit]
  9. Yeung, K. Algorithmic regulation: A critical interrogation. Regul. Gov. 2018, 12, 505–523. [Google Scholar] [CrossRef] [Scilit]
  10. Shi, Y.; Wei, Z.; Shahbaz, M.; Zeng, Y. Exploring the dynamics of low-carbon technology diffusion among enterprises: An evolutionary game model on a two-level heterogeneous social network. Energy Econ. 2021, 101, 105399. [Google Scholar] [CrossRef] [Scilit]
  11. Li, J.; Xu, X.; Yang, Y. Research on the Regulation of Algorithmic Price Discrimination Behaviour of E-Commerce Platform Based on Tripartite Evolutionary Game. Sustainability 2023, 15, 8294. [Google Scholar] [CrossRef] [Scilit]
  12. Guo, Y.; Lin, J.; Zhuang, W. An Evolutionary Game-Based Regulatory Path for Algorithmic Price Discrimination in E-Commerce Platforms. Mathematics 2024, 12, 2774. [Google Scholar] [CrossRef] [Scilit]
  13. Kalberg, S. Max Weber’s types of rationality: Cornerstones for the analysis of rationalization processes in history. Am. J. Sociol. 1980, 85, 1145–1179. [Google Scholar] [CrossRef] [Scilit]
  14. Diakopoulos, N. Algorithmic accountability: Journalistic investigation of computational power structures. Digit. J. 2015, 3, 398–415. [Google Scholar] [CrossRef] [Scilit]
  15. Burrell, J. How the machine ‘thinks’: Understanding opacity in machine learning algorithms. Big Data Soc. 2016, 3, 2053951715622512. [Google Scholar] [CrossRef] [Scilit]
  16. Ananny, M.; Crawford, K. Seeing without knowing: Limitations of the transparency ideal and its application to algorithmic accountability. New Media Soc. 2018, 20, 973–989. [Google Scholar] [CrossRef] [Scilit]
  17. Bracci, E. The loopholes of algorithmic public services: An “intelligent” accountability research agenda. Account. Audit. Account. J. 2023, 36, 739–763. [Google Scholar] [CrossRef] [Scilit]
  18. Hu, Y.; Li, G.; Liu, M.; Qu, S. Information sharing and personalized pricing in online platforms. Prod. Oper. Manag. 2024, 34, 3958–3977. [Google Scholar] [CrossRef] [Scilit]
  19. Yang, J.; Sahni, N.S.; Nair, H.S.; Xiong, X. Advertising as information for ranking e-commerce search listings. Mark. Sci. 2024, 43, 360–377. [Google Scholar] [CrossRef] [Scilit]
  20. Chen, H. Personalized recommendation system of e-commerce based on big data analysis. J. Interdiscip. Math. 2018, 21, 1243–1247. [Google Scholar] [CrossRef] [Scilit]
  21. Rhodes, A.; Zhou, J. Personalized pricing and competition. Am. Econ. Rev. 2024, 114, 2141–2170. [Google Scholar] [CrossRef] [Scilit]
  22. Bakshy, E.; Messing, S.; Adamic, L.A. Exposure to ideologically diverse news and opinion on Facebook. Science 2015, 348, 1130–1132. [Google Scholar] [CrossRef] [Scilit]
  23. Flaxman, S.; Goel, S.; Rao, J.M. Filter bubbles, echo chambers, and online news consumption. Public Opin. Q. 2016, 80, 298–320. [Google Scholar] [CrossRef] [Scilit]
  24. Rochet, J.C.; Tirole, J. Platform competition in two-sided markets. J. Eur. Econ. Assoc. 2003, 1, 990–1029. [Google Scholar] [CrossRef] [Scilit]
  25. Armstrong, M. Competition in two-sided markets. RAND J. Econ. 2006, 37, 668–691. [Google Scholar] [CrossRef] [Scilit]
  26. Rysman, M. The economics of two-sided markets. J. Econ. Perspect. 2009, 23, 125–143. [Google Scholar] [CrossRef] [Scilit]
  27. Hagiu, A.; Wright, J. Multi-sided platforms. Int. J. Ind. Organ. 2015, 43, 162–174. [Google Scholar] [CrossRef] [Scilit]
  28. Mittelstadt, B.D.; Allo, P.; Taddeo, M.; Wachter, S.; Floridi, L. The ethics of algorithms: Mapping the debate. Big Data Soc. 2016, 3, 2053951716679679. [Google Scholar] [CrossRef] [Scilit]
  29. Tironi, M.; Lisboa, D.I.R. Artificial intelligence in the new forms of environmental governance in the Chilean State: Towards an eco-algorithmic governance. Technol. Soc. 2023, 74, 102264. [Google Scholar] [CrossRef] [Scilit]
  30. Wachter, S.; Mittelstadt, B.; Floridi, L. Why a right to explanation of automated decision-making does not exist in the general data protection regulation. Int. Data Priv. Law 2017, 7, 76–99. [Google Scholar] [CrossRef] [Scilit]
  31. Grochowski, M.F.; Jablonowska, A.A.; Lagioia, F.; Sartor, G. Algorithmic price discrimination and consumer protection: A digital arms race? Technol. Regul. 2022, 2022, 36–47. [Google Scholar] [CrossRef] [Scilit]
  32. Xiao, Y.; Zhao, W.; Huang, Y.; Li, T.; Li, Q. A joint learning recommendation model for e-commerce platforms integrating long-term and short-term interests. IEEE Trans. Serv. Comput. 2024, 17, 1326–1339. [Google Scholar] [CrossRef] [Scilit]
  33. Griva, A. “I can get no e-satisfaction”: What analytics say? Evidence using satisfaction data from e-commerce. J. Retail. Consum. Serv. 2022, 66, 102954. [Google Scholar] [CrossRef] [Scilit]
  34. Chen, S.; Qiu, H.; He, W. The information cocoon paradox: Fostering unity or fueling divergence? Humanit. Soc. Sci. Commun. 2025, 12, 859. [Google Scholar] [CrossRef] [Scilit]
  35. Just, N.; Latzer, M. Governance by algorithms: Reality construction by algorithmic selection on the Internet. Media Cult. Soc. 2017, 39, 238–258. [Google Scholar] [CrossRef] [Scilit]
  36. Gorwa, R.; Binns, R.; Katzenbach, C. Algorithmic content moderation: Technical and political challenges in the automation of platform governance. Big Data Soc. 2020, 7, 2053951719897945. [Google Scholar] [CrossRef] [Scilit]
  37. Herani, R. Should we play an unfair game? The roles of regulatory effectiveness, government support and algorithmic bias in e-commerce entrepreneurial readiness. J. Entrep. Emerg. Econ. 2025, 18, 237–265. [Google Scholar] [CrossRef] [Scilit]
  38. Shou, M.; Yu, J.; Dai, R. Identify the effect of government regulations on the live streaming e-commerce. Ind. Manag. Data Syst. 2023, 123, 2909–2928. [Google Scholar] [CrossRef] [Scilit]
  39. Liu, Y.; Sun, X. Towards more legitimate algorithms: A model of algorithmic ethical perception, legitimacy, and continuous usage intentions of e-commerce platforms. Comput. Hum. Behav. 2024, 150, 108006. [Google Scholar] [CrossRef] [Scilit]
  40. Jin, J.; Zhang, D.; Hao, Y. Optimizing the environmental value transmission chain of green agricultural products: An evolutionary game analysis with livestream e-commerce. Comput. Ind. Eng. 2025, 207, 111359. [Google Scholar] [CrossRef] [Scilit]
  41. Are, C. Flagging as a silencing tool: Exploring the relationship between de-platforming of sex and online abuse on Instagram and TikTok. New Media Soc. 2025, 27, 3577–3595. [Google Scholar] [CrossRef] [Scilit]
  42. Reynolds, C.J.; Hallinan, B. User-generated accountability: Public participation in algorithmic governance on YouTube. New Media Soc. 2024, 26, 5107–5129. [Google Scholar] [CrossRef] [Scilit]
  43. Duani, N.; Barasch, A.; Morwitz, V. Demographic pricing in the digital age: Assessing fairness perceptions in algorithmic versus human-based price discrimination. J. Assoc. Consum. Res. 2024, 9, 257–268. [Google Scholar] [CrossRef] [Scilit]
  44. Zhang, G.; Wang, X.; Wang, Y.; Xu, J. A tripartite evolutionary game for the regional green logistics: The roles of government subsidy and platform’s cost-sharing. Kybernetes 2024, 53, 216–237. [Google Scholar] [CrossRef] [Scilit]
  45. Xu, X.; Fan, R.; Wang, D.; Wang, Y.; Wang, Y. The role of consumer reviews in e-commerce platform credit supervision: A signaling game model based on complex network. Electron. Commer. Res. Appl. 2024, 63, 101347. [Google Scholar] [CrossRef] [Scilit]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.