A prospect-theoretic evolutionary game approach for disinformation governance in social-media ecosystems Authors: Song Xiaoyu Published: 2026 ABSTRACT The viral spread of disinformation on participatory media platforms threatens public opinion integrity and social stability. To support effective governance decision-making, we propose a prospect-theoretic evolutionary game model involving three bounded-rational actors: regulators, social media platforms, and self-media operators. By incorporating a prospect-theoretic value function to capture loss aversion and diminishing sensitivity ignored by traditional utility models, we derive the stability conditions for the system’s ideal cooperative equilibrium where regulators disclose information truthfully, platforms moderate proactively, and self-media operators verify content. Numerical simulations show that initial cooperation propensity, reward–penalty intensity, and loss aversion significantly influence system convergence. The findings provide actionable and incentive-compatible policy guidelines for curbing online disinformation.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-53697-9. Introduction The rise of self-media has fundamentally transformed the information landscape, democratizing content creation while simultaneously creating unprecedented channels for the spread of disinformation1, 2. This proliferation of false information erodes public trust, distorts social discourse, and poses significant threats to social stability and national security3, making its effective governance a critical and urgent challenge. Addressing this challenge requires an understanding of disinformation propagation as a complex, strategic interaction involving multiple autonomous actors. Evolutionary Game Theory (EGT) has emerged as a powerful analytical framework for modeling such dynamic, multi-agent systems where strategies evolve through imitation, learning, and adaptation4, 5. Unlike classical game theory, EGT does not assume hyper-rationality but rather explores how population-level behaviors emerge and stabilize over time, providing key insights into phenomena like cooperation and norm evolution in structured groups6, 7. A substantial body of research has applied EGT to dissect online public opinion dynamics. Studies have built tripartite models involving regulators, media, and the public to reveal the interactive mechanisms of disinformation spread8. Analyses using Jacobian matrices have shown that increasing regulator penalties can curb false information, though a balance with regulatory costs is necessary9. Other work has integrated network effects and herd behavior into game-theoretic models to explain multidimensional opinion dissemination10, with findings corroborated in specific contexts like university settings11. Research has further elucidated the strategic interdependence between regulator regulation and media reporting tendencies12, modeled the acceleration of negative emotions via self-media “traffic” incentives13, and demonstrated the vicious cycles triggered by regulator inaction14. Collectively, these studies provide a strong foundation by quantifying stakeholder interests and simulating equilibrium outcomes15, 16, 17, 18, 19. However, two significant gaps persist in this literature. First, a predominant assumption of perfect or bounded rationality in standard EGT often overlooks the descriptive insights of behavioral economics. Real-world decision-makers in high-uncertainty environments like disinformation governance are not purely rational optimizers. Their choices are heavily influenced by cognitive biases, particularly those related to risk perception under uncertainty. Prospect Theory (PT), the seminal descriptive model of decision-making under risk24, posits that individuals evaluate outcomes relative to a reference point, exhibit loss aversion (losses loom larger than equivalent gains), and demonstrate diminishing sensitivity to changes in value. While PT has been successfully integrated with EGT in other domains such as environmental governance and supply chain management20, 21, its application to model the risk-sensitive, psychologically-grounded strategic choices in the tripartite disinformation governance ecosystem remains notably underexplored. Second, while the roles of regulators and content creators (self-media) are frequently examined, the pivotal and legally-increasingly-responsible role of social media platforms is often under-theorized or treated passively22, 23. Platforms are not neutral conduits but active governance actors with their own strategic incentives concerning content moderation, user engagement, and regulatory compliance. A comprehensive model must explicitly account for their strategic choice between active governance and passive response. To bridge these gaps, this paper constructs a tripartite evolutionary game model that integrates the precepts of Prospect Theory. We focus on the strategic interactions between the regulator, social media platforms, and self-media operators. Our core contribution is to ground the agents’ payoff calculations in PT’s value function, thereby capturing how subjective perceptions of gains and losses, driven by loss aversion and diminishing sensitivity, fundamentally alter the evolutionary dynamics of the system compared to traditional utility-based models. Through stability analysis and systematic numerical simulation, we investigate how key parameters—including the intensity of regulator rewards and penalties, initial willingness to cooperate, and the agents’ psychological parameters (loss aversion λ, sensitivity coefficients α and β)—influence the trajectory and efficiency of the system’s convergence toward an ideal collaborative governance state. Notably, the strategic interaction between regulators and platforms is driven by legal obligations, reputational incentives, and content moderation costs, all of which have been systematically incorporated into the parameter settings of our proposed model. The remainder of this paper is structured as follows: Sect. "Problem description and model construction" details the problem description and the construction of the PT-based tripartite evolutionary game model. Section "Stability analysis of equilibrium point" provides a stability analysis of the system’s equilibrium points. Section "Numerical simulation analysis" presents numerical simulations and sensitivity analyses. Section "Research conclusions" concludes with findings, policy implications, and directions for future research. Problem description and model construction Problem description As the direct producer and disseminator of information, the behavior of self-media operators plays a decisive role in the formation and development of online public opinion. The regulator plays the role of regulator and standardizer in the field of public opinion, formulates perfect laws and regulations, strengthens the regulation of self-media behavior, standardizes the behavior of self-media accounts, and maintains the clarity of cyberspace. Social platforms, as content publishing and dissemination channels, need to establish a sound content review mechanism to deal with inaccurate information in a timely manner. Social platforms should also cooperate with the regulator to combat online rumors, improve the reach of disinformation, and take measures such as banning and closing down offending self-media accounts. Among them, the regulator and social platforms need to strengthen the supervision of self-media operators, while self-media should consciously abide by laws and regulations. Through the mutual constraint game of the three main parties, a stable and healthy public opinion environment is gradually formed. It can not only enhance the clarity of the network environment, but also effectively prevent and reduce the phenomenon of rumor spreading and protect the legitimate rights and interests of citizens. Accordingly, this paper constructs a tripartite evolutionary game model of “regulator-social platform-self-media operators”. Basic assumptions Assumption 1 Prospect Theory. The process of public opinion regulation is influenced by factors such as information policy and network regulations, and is characterized by strong uncertainty. All parties involved in public opinion governance, including the regulator, social platforms and self-media, are often unable to accurately assess the value of different decisions and have subjective cognitive biases. Therefore, this paper starts from the perspective of prospect theory, and measures the gains and losses of different strategic choices by subjective perceived value rather than actual utility. According to prospect theory, when the game parties are uncertain about the gain or loss, the perceived value is generated with the following formula:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:V\left({\Lambda\:}\pi\:\right)=\left\{\begin{array}{c}\varDelta\:{\pi\:}^{\alpha\:},\varDelta\:\pi\:\ge\:0\\\:{-\lambda\:\left({\Delta\:}\pi\:\right)}^{\beta\:},\varDelta\:\pi\:<0\end{array}\right.$$\end{document} The aforementioned formulas depict the ∆π profit and loss margin between the player’s behavioral strategy and the reference point. To enable the evaluation of the designated reference point, it is typically established at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\alpha ,\beta \:(0{ \leqq }\:\alpha \:,\beta \:{ \leqq }\:1)$$\end{document}.The risk preference coefficient, indicative of the perceived value function, suggests that a higher value corresponds to an increased risk preference by the decision-maker \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\lambda\:(\lambda\:\ge\:1)$$\end{document}. The loss avoidance coefficient, indicative of the perceived value function, suggests that a higher value correlates with an increased propensity for loss avoidance by the decision maker. The aforementioned formula enables the quantification of the subjective perceived value of the regulator, social platforms, and self-media regarding public opinion governance decisions, facilitating the analysis of their strategic choice behavior. Assumption 2 Selection of strategy. Public opinion management is a complex multi-party game process. When facing a public crisis, the regulator can choose to “actively publicize” the details of the incident or “respond in silence”. Positive disclosure can enhance public trust and improve the image of the regulator, but it entails a certain amount of time and investigation costs. Silent response may result in a loss of credibility and an erosion of public trust. Self-media can choose to “report truthfully” or “report falsely”. Truthful reporting refers to reporting information after verifying its authenticity, which can help to enhance the personal image of self-publishers and attract attention, but it requires time and information costs. False reports publish unverified or fabricated information to gain traffic and cause reputational damage, but in the heat of an event, they are likely to be used because of their urgency and importance. Social platforms may choose to “actively manage” or “passively respond”. Proactive management of inaccurate information can help to maintain stability and order in the online environment, and requires vetting costs. Negative responses may lead to disorder and regulatoral penalties, damaging the platform’s reputation. Strategies indicate the possible action paths of each subject in information disclosure and dissemination. The probability of the regulator releasing information is\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\:x$$\end{document}, the probability of self-media reporting truthfully is y, and the probability of the platform taking the initiative to govern is z, where 0