False data content game method and system based on dynamic non-cooperative game model

Abstract

This invention provides a method and system for deceptive data content game theory based on a dynamic non-cooperative game model, relating to the field of network security detection technology. It addresses the limitations of existing technologies, such as delayed attack and defense response and insufficient game targeting. The method constructs a three-party game model comprising a generator, a user, and a detector. Each party defines its own strategy space, which is the set of executable strategies in the generation, targeted application, or detection of deceptive data content. A payoff function is set for each party, deriving a corresponding payoff value based on their current strategies and their interactions. After initialization, the equilibrium strategy in the three-party game is solved iteratively. Each party updates its strategy by selecting the strategy that maximizes its payoff within its own strategy space based on the payoff value. After the iteration terminates, the detector's strategy is determined as the deceptive data content detection strategy. This invention successfully achieves end-to-end, dynamic, and precise deceptive data content game theory results.

Description

Technical Field

[0001] This invention relates to the field of network security detection technology, specifically to a method and system for detecting fake data content based on a dynamic non-cooperative game model. Background Technology

[0002] With the rapid development of generative artificial intelligence technology, the ability to generate fake data now covers various forms, including text, voice, images, videos, and multimodal content, with significantly improved simulation accuracy and generation efficiency. Simultaneously, the methods used to spread fake information are becoming increasingly sophisticated. Through the synergy of user profiling analysis, psychological manipulation strategies, and social networks, content can be precisely delivered and spread in a chain, greatly increasing the social harm and difficulty of controlling fake information. Against this backdrop, fake data detection technology, as a key defensive element, has received widespread attention for its effectiveness and adaptability.

[0003] Current research on detection technologies largely focuses on feature recognition and discrimination of single-modal content. Several specialized detection schemes have been developed for scenarios such as image tampering traces, abnormal speech synthesis, or textual logical contradictions. While these schemes possess a certain level of recognition capability under specific conditions, their design logic often isolates the content generation stage, failing to incorporate the entire chain of fake data from generation and targeted application to dissemination into the technical considerations. There is a significant disconnect between detection strategies and actual attack and defense scenarios, making it difficult to address cross-modal and collaborative content forgery and dissemination.

[0004] Some studies have attempted to incorporate game theory to enhance the policy adaptability of detection systems. However, existing models are mostly built on static assumptions, treating the strategies of both attackers and defenders as fixed parameters, neglecting the essential characteristics of continuous policy evolution and real-time response in technological confrontation. The generator avoids detection through rapid iterative optimization, while the detector's strategy adjustment cycle is relatively long, leading to a significant decline in defensive effectiveness over time. More critically, existing game theory frameworks generally treat the generation and detection stages in isolation, failing to incorporate key variables such as the inducement strategy and propagation path in the application stage into the game variable system. This makes it difficult for the models to reflect the dynamic coupling relationship between "generation-application-detection" in real-world attack and defense scenarios. Furthermore, when facing complex scenarios with concurrent and large-scale propagation of multi-form content, there is a lack of a global decision-making mechanism to coordinate multimodal detection resources and balance detection accuracy and efficiency. Strategies at each stage operate in isolation, making the overall system's robustness and response efficiency insufficient to meet practical needs.

[0005] The aforementioned issues collectively lead to problems such as insufficient adaptability of existing technological systems in dynamic adversarial environments, susceptibility to circumvention of detection strategies, and weaknesses in the defense chain. The continuous evolution of technologies for generating and using fake data further highlights the limitations of static, fragmented, and localized technological approaches. There is an urgent need in this field for a technological approach that closely aligns with the dynamic evolution of attack and defense, connects all stages of content generation, strategy application, and detection response, and possesses global collaborative optimization capabilities, in order to drive the evolution of fake data governance capabilities towards intelligence and systematization. Summary of the Invention

[0006] The purpose of this invention is to address the limitations of existing fake data content game technologies, such as sluggish attack and defense responses, insufficient game targeting, and difficulty in adapting to multi-form, large-scale, dynamic adversarial scenarios. Therefore, this invention proposes a fake data content game method and system based on a dynamic non-cooperative game model. This invention constructs a three-party dynamic non-cooperative game system comprising "fake data generator - fake data user - fake data detector." Through participant modeling, strategy space construction, payoff function design, dynamic equilibrium solving, and iterative strategy optimization, it successfully achieves end-to-end, dynamic, and precise fake data content game effects.

[0007] The present invention employs the following technical solutions to achieve its objective: A method for playing fake data content based on a dynamic non-cooperative game model includes the following steps: S1. Construct a three-party game model that includes the generator, user, and detector of fake data, and define the strategy space for each of the generator, user, and detector. The strategy space is the set of strategies that each party can execute in the generation, targeted use, or detection of fake data content. S2. Based on the three-party game model, set the payoff function for the generator, user and detector respectively. The input parameters of the payoff function include the current strategy of each party and the interaction between strategies, and the output parameter is the payoff value of each party. S3. Initialize the strategies of the generator, user and detector respectively, and solve the equilibrium strategy in the three-party game through iterative method. Each iteration includes: calculating the current payoff value of each party based on the current strategy and corresponding payoff function of each party, and each party selects the strategy that maximizes its own payoff in its own strategy space based on the payoff value to update the current strategy. S4. When the strategy update process has met the preset convergence condition, the iteration is terminated, the fake data content game is completed, and the detection strategy after the iteration is terminated is determined as the fake data content detection strategy.

[0008] Specifically, in step S1, the strategy space of the fake data generator includes a combination of generation technology selection strategy and detection avoidance optimization strategy. The generation technology selection strategy is the configuration of generation technology for text, voice, image, video and / or multimodal fake data. The detection avoidance optimization strategy includes artifact elimination, feature camouflage, multi-technology fusion generation and / or dynamic switching of generation mode. The strategic space of those who use false data includes a combination of target positioning strategies, dissemination inducement strategies, and delivery method strategies. Among them, the target positioning strategy is target screening based on user profiles and / or multi-dimensional psychological characteristics; the dissemination inducement strategy includes social relationship guidance, emotional resonance inducement, and / or cognitive bias inducement; and the delivery method strategy includes covert channel delivery, batch delivery, and / or time-segmented precise delivery. The strategy space for fake data detection includes a combination of detection technology selection strategies, detection optimization strategies, and response and handling strategies. The detection technology selection strategy is a configuration of dedicated detection technologies for text, voice, image, video, and / or multimodal fake data. The detection optimization strategy includes model compression inference, multi-stage pipelined inference, dynamic threshold adjustment, and / or multi-technology fusion detection. The response and handling strategy includes real-time blocking, source tracing, secondary verification, and / or early warning push.

[0009] Specifically, in step S1, a three-party game model is constructed, including: constructing corresponding feature vectors for the generator, user, and detector, respectively. Each feature vector consists of multiple core feature parameters. The core feature parameters include the generator's generation technology type and evasion capability parameters, the user's target group characteristics and induction strategy strength parameters, and the detector's detection technology coverage and efficiency parameters.

[0010] Preferably, in step S2, when calculating the corresponding profit values ​​for each party, the current strategies and corresponding feature vectors of the generator, user, and detector are used as input parameters for their respective profit functions.

[0011] Preferably, in step S2, the payoff function of the fake data generator includes a payoff item for successful generation, a payoff item for generation, a payoff item for avoiding detection, and a payoff item for the risk of being detected; the payoff function of the fake data user includes a payoff item for successful propagation, a payoff item for application, and a payoff item for propagation blocking; the payoff function of the fake data detector includes a payoff item for successful detection, a payoff item for detection, and a payoff item for missed detection / false detection; the weight coefficients of each parameter in the payoff function are dynamically configured according to the game scenario.

[0012] Preferably, in step S3, the Nash equilibrium solution mechanism is used to solve the equilibrium strategy in the three-party game process. During the iteration process, the strategy updates of each party follow the optimal response principle of non-cooperative games. Any participant among the generator, user, and detector chooses the strategy that maximizes its own benefit given the current strategies of other participants.

[0013] Specifically, in step S3, each party selects the strategy that maximizes its own benefit within its own strategy space based on the benefit value to update the current strategy. This includes: for any participant among the generator, user, and detector, calculating the benefit value corresponding to each candidate strategy in the participant's own strategy space based on the participant's current strategy, the current strategies of other participants, and the benefit function, and selecting the candidate strategy with the largest benefit value as the participant's strategy for the next iteration.

[0014] Specifically, in step S4, the preset convergence conditions include: in two consecutive iterations, the change in the payoff value corresponding to the strategies of the generator, user, and detector is less than a preset threshold, or the number of iterations reaches the preset maximum number of iterations; wherein, the change in strategy is determined by calculating the Euclidean distance or Hamming distance between the strategy vectors in adjacent iteration rounds.

[0015] Specifically, in step S4, the fake data content detection strategy is a specific strategy instance in the detection party's strategy space, including the combination configuration of detection technology selection strategy, detection optimization strategy and response and handling strategy; the fake data content detection strategy is used to guide the subsequent identification, analysis and handling of fake data content.

[0016] This invention also provides a fake data content game system for implementing the aforementioned method, comprising the following functional modules: The participant modeling module is configured to construct a three-party game model that includes the fake data generator, the user, and the detector, extract the core feature parameters of each participant, and construct and dynamically update the feature vectors of each participant. The strategy space management module is configured to store, classify, update, and recall the sets of strategies used by the generator, user, and detector in the generation, targeted application, and detection stages of fake data content, respectively. The revenue calculation module is configured to calculate the revenue values ​​of the generator, user, and detector in real time based on the current strategies of each participant, the interaction relationships between strategies, and the corresponding feature vectors through a revenue function. The dynamic equilibrium solution module is configured to initialize the strategies of each participant, and through iterative execution of strategy updates and benefit evaluation, control each participant to select the strategy that maximizes its own benefit within its own strategy space based on the benefit value, and terminate the iteration when the preset convergence condition is met, and output the strategy of the detection party as the fake data content detection strategy. The game execution module is configured to receive false data content detection strategies, drive the detection party to perform detection and blocking operations, and synchronously monitor the strategy execution status of the generator and the user. The data acquisition and feedback module is configured to collect strategy execution data, environmental feedback data, and detection result data in real time during the game process, and feed the collected data back to the participant modeling module and the payoff calculation module to support the updating of feature vectors and payoff functions, as well as subsequent strategy iterations.

[0017] In summary, due to the adoption of this technical solution, the beneficial effects of this invention are as follows: This invention collects game process data in real time, dynamically updates participants' strategies and payoff functions, and iteratively solves the Nash equilibrium, achieving real-time adaptation of the strategies of the generator, user, and detector. It can quickly respond to the iterative upgrades of fake data generation and application technologies, and significantly improve the timeliness of the game.

[0018] This invention is the first to incorporate the use of false data into a game theory system, constructing a three-party collaborative game mechanism of "generation-use-detection". This breaks through the limitations of existing technologies that only focus on the binary game of "generation-detection", and the game strategy is more in line with actual application scenarios and is more targeted.

[0019] This invention quantifies the effects and costs of each party's strategies through a payoff function, and achieves the output of the globally optimal game strategy based on Nash equilibrium. It can be adapted to various forms of fake data such as text, voice, video, and images. When applying the method, it can be combined with technologies such as model compression and multi-stage pipelined inference, which can efficiently cope with the needs of fake data game in large-scale data flow scenarios.

[0020] The system of this invention adopts a modular architecture. When a new generation / detection technology is added, the strategy library managed by the strategy space management module can be directly updated, realizing flexible expansion of the strategy space. The reward function can adjust the weight parameters according to different application scenarios such as social platforms, adapting to the game requirements of different scenarios, and its robustness is significantly better than that of a single technical solution.

[0021] This invention quantifies the implementation costs and risk losses of the generator and the user through a payoff function, guiding the testing party to choose the most cost-effective and efficient game-theoretic testing strategy. While ensuring the game-theoretic effect, it reduces the computing power investment and technical costs of the testing party, thereby improving the practicality and feasibility of the solution. Attached Figure Description

[0022] The present invention is described in detail with reference to the following figures, which include three figures as follows: Figure 1 This is a schematic diagram illustrating the overall process of the false data content game method of the present invention; Figure 2 This is a detailed flowchart illustrating the actual game-playing process of the method of the present invention; Figure 3 This is a schematic diagram showing the module composition and relationship of the false data content game system of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0025] Example 1 A method for playing games with fake data content based on a dynamic non-cooperative game model. Figure 1 The overall process of this method is briefly described below and can be viewed concurrently; the key steps of this method can be summarized as follows: S1. Construct a three-party game model that includes the generator, user, and detector of fake data, and define the strategy space for each of the generator, user, and detector. The strategy space is the set of strategies that each party can execute in the generation, targeted use, or detection of fake data content. S2. Based on the three-party game model, set the payoff function for the generator, user and detector respectively. The input parameters of the payoff function include the current strategy of each party and the interaction between strategies, and the output parameter is the payoff value of each party. S3. Initialize the strategies of the generator, user and detector respectively, and solve the equilibrium strategy in the three-party game through iterative method. Each iteration includes: calculating the current payoff value of each party based on the current strategy and corresponding payoff function of each party, and each party selects the strategy that maximizes its own payoff in its own strategy space based on the payoff value to update the current strategy. S4. When the strategy update process has met the preset convergence condition, the iteration is terminated, the fake data content game is completed, and the detection strategy after the iteration is terminated is determined as the fake data content detection strategy.

[0026] This embodiment will describe the details and preferred methods of each step in the above-described order. The specific game process ultimately formed during the iterative game process can be found in [link to documentation]. Figure 2 The illustration.

[0027] First, in step S1, the roles and core objectives of the three game participants are clarified, thereby constructing a three-way game model as follows: Fake data generators: Their core objective is to generate various forms of fake data and evade detection. Their core characteristic parameters include the type of generation technology and the ability to evade detection. Users of fake data: The core objective is to effectively spread fake data through various strategies. Their core characteristic parameters include the characteristics of the target group, the strength of the inducement strategy, and different application scenarios. Fake data detection: The core objective is to identify various types of fake data and block their spread. Its core characteristic parameters include the coverage and efficiency of detection technology, and can further cover the type of detection technology and accuracy.

[0028] This embodiment constructs corresponding feature vectors for the generator, user, and detector, respectively. These feature vectors are composed of the aforementioned core feature parameters. The feature vectors are used for policy matching and reward calculation. For the first... Feature vectors of each participant Its expression is as follows:

[0029] In the formula, That is, the first The first core feature parameter of each participant, and so on. This represents the total number of core characteristic parameters of the participating party.

[0030] Next, we define the respective policy spaces for the generator, the user, and the detector.

[0031] The strategy space of fake data generators This includes combinations of generation technology selection strategies and detection avoidance optimization strategies, as follows: Generative technology selection strategy Configuration of technologies for generating text, voice, images, videos and / or multimodal fake data, such as copywriting generation models, voice generation models, digital human generation models, image generation models, etc. Detection avoidance optimization strategies Artifact removal, feature camouflage, multi-technology fusion generation, and / or dynamic switching of generation modes.

[0032] Therefore, the generator strategy The expression is as follows:

[0033] The strategic space for those who use fake data This includes combinations of target positioning strategies, dissemination inducement strategies, and delivery method strategies, as follows: Target positioning strategy Target filtering based on user profiles and / or multi-dimensional psychological characteristics, such as target profiling based on multi-dimensional psychological enhancement, group feature matching and positioning, personalized target filtering and other operations; Propagation Induction Strategy Social relationship guidance, emotional resonance induction, and / or cognitive bias induction are used to guide or induce the spread of information flow through these strategies. Delivery method strategy Covert channel delivery, batch delivery, and / or time-segmented precise delivery.

[0034] Therefore, the strategy of using square The expression is as follows:

[0035] The strategy space of fake data detection This includes combinations of detection technology selection strategies, detection optimization strategies, and response and handling strategies, as follows: Detection technology selection strategy : Dedicated detection technology configuration for text, voice, image, video and / or multimodal fake data, which may specifically include cross-domain artifact characterization image detection, keyframe extraction video detection, diffusion reconstruction contrast training general image detection, visual scene graph video source tracing, position game message consistency detection and other methods; Detection optimization strategy Model compression inference, multi-stage pipelined inference, dynamic threshold adjustment and / or multi-technology fusion detection; Response and handling strategies Real-time blocking, source tracing, secondary verification, and / or early warning push notifications.

[0036] Therefore, the detection strategy The expression is as follows:

[0037] In step S2, when calculating the corresponding payout values ​​for each party, the current strategy and corresponding feature vector of the generator, user, and detector are used as input parameters for their respective payout functions.

[0038] In this preferred embodiment, the generator's revenue = revenue from successful generation - generation cost - cost of avoiding detection - loss due to the risk of being detected, i.e., the generator's revenue value. The expression is as follows:

[0039] In the formula, To generate success reward weights, The probability of generating data to evade detection is determined by the generating strategy. Detection strategy Decide; To generate cost weights, To generate the cost of implementing the technology; To avoid weighting testing costs, To avoid detection and optimize costs; As a risk loss weight, Losses after detection include data corruption, technology exposure, and other losses.

[0040] In this preferred embodiment, the user's benefit = successful propagation benefit - application cost - propagation disruption loss, i.e., the user's benefit value. The expression is as follows:

[0041] In the formula, To disseminate the weighting of successful returns, The probability of successful dissemination of false data is jointly determined by the strategies of the three participating parties; In order to apply cost weighting, The cost of implementing the strategy; To block loss weights, The losses incurred due to the disruption of transmission include delivery failures and loss of targets.

[0042] In this preferred embodiment, the testing party's revenue = revenue from successful detection - testing cost - loss from missed / false detections, i.e., the testing party's revenue value. The expression is as follows:

[0043] In the formula, To detect the successful return weight, The probability of detecting and blocking fake data is jointly determined by the strategies of the three participating parties; As a weight for detection cost, The cost of implementing a detection strategy may include computing power costs, technology investment, etc., and is related to the detection strategy. related; Weight loss due to errors This includes losses resulting from missed or false detections, such as the spread of false data and trust crises.

[0044] In step S3, this embodiment uses the Nash equilibrium solution mechanism to solve the equilibrium strategy in the three-party game process. During the iteration process, the strategy updates of each party follow the optimal response principle of non-cooperative games. Any participant among the generator, user, and detector chooses the strategy that maximizes its own benefit given the current strategies of other participants.

[0045] In this embodiment, a Nash equilibrium solution model for dynamic non-cooperative game is constructed based on the strategy space and payoff function of the three participants. The core objective is to find the optimal solution for the strategies of the three participants. This prevents any single participant from increasing their own profits when adjusting their strategy individually.

[0046] The specific iterative solution process is as follows: First, initialize the game by setting the maximum number of iterations. Convergence threshold Initialize the initial strategies of the three participants. With feature vectors ; To calculate the payoff of a single game, substitute the current strategy into the payoff functions of the three players to calculate the current payoff for each player. ; Each participant, based on its current payoff and the strategies of the other two participants (excluding itself), searches for the optimal strategy within its own strategy space to generate the strategy for the next round. Therefore, the... The iteration rule for each participant can be expressed as:

[0047] In the formula, Representing the The participating party in the first The optimal strategy for round-robin; Representative at the The strategic space of each participant In the middle, find the value that can make the target profit. The strategy that achieves the maximum value; among which Representing the The participating party in the first In a round, if the strategies of the other two participants are fixed, the player adopts its current strategy. Corresponding profit value .

[0048] In step S4 of this embodiment, the preset convergence conditions include: in two consecutive iterations, the change in the reward value corresponding to the strategies of the generator, the user, and the detector is less than a preset threshold, or the number of iterations reaches the preset maximum number of iterations.

[0049] By calculating the difference in returns between the current strategy and the previous strategy. Determine the Is it less than the convergence threshold? Determine whether to terminate the iteration. Convergence threshold. Defined as the critical value of the payoff difference used to determine whether an iteration has converged, it can be set based on factors such as the accuracy requirements and computing costs of the fake data game scenario. For example, in high-precision scenarios... The value range is from 0.001 to 0.01, but in scenarios where real-time performance is emphasized, this value can be relaxed to 0.05 to 0.1. If the generator, user, and detector... All less than If the iteration converges, the current strategy is the Nash equilibrium strategy; otherwise, the iteration process continues.

[0050] Finally, in step S4, based on the dynamic equilibrium solution, the full-link game process can be executed to realize the flow of "strategy output - execution monitoring - state feedback - strategy iteration", as follows: Strategy Output: Based on the Nash equilibrium strategy, the optimal game strategy is output to the three participants (or the game platform dominated by the detection party), where the detection party's strategy explicitly adapts to the detection technology and response scheme of the current generator and user strategies.

[0051] Execution monitoring: Real-time collection of strategy execution data from the three participants, including the form and characteristics of fake data generated by the party, the propagation path and effect of the user, and the detection accuracy and blocking efficiency of the detection party. Status feedback: Based on the collected data, update the feature vectors of the three participants and determine whether there is a strategy deviation, such as the generator adopting a new avoidance detection technology or the user adjusting the propagation inducement strategy. Strategy iteration: If a strategy deviation occurs, the dynamic equilibrium solution process is restarted to update the optimal game strategy, thereby realizing real-time dynamic adjustment of the game strategy and ensuring the continuity and adaptability of the game effect.

[0052] In this process, the detection strategy after the iteration is terminated is determined as the fake data content detection strategy. It is a specific strategy instance in the detection strategy space, including the combination and configuration of detection technology selection strategies, detection optimization strategies, and response and handling strategies. Therefore, the fake data content detection strategy can be used to guide subsequent operations in the actual identification, analysis, and handling of fake data content during network security testing.

[0053] Example 2 Based on Example 1, this example provides a fake data content game system, which is used to implement the fake data content game method in Example 1. For example... Figure 3 As shown, the system includes the following functional modules: The participant modeling module is configured to construct a three-party game model that includes the fake data generator, the user, and the detector, extract the core feature parameters of each participant, and construct and dynamically update the feature vectors of each participant. The strategy space management module is configured to store, classify, update, and recall the sets of strategies used by the generator, user, and detector in the generation, targeted application, and detection stages of fake data content, respectively. The revenue calculation module is configured to calculate the revenue values ​​of the generator, user, and detector in real time based on the current strategies of each participant, the interaction relationships between strategies, and the corresponding feature vectors through a revenue function. The dynamic equilibrium solution module is configured to initialize the strategies of each participant, and through iterative execution of strategy updates and benefit evaluation, control each participant to select the strategy that maximizes its own benefit within its own strategy space based on the benefit value, and terminate the iteration when the preset convergence condition is met, and output the strategy of the detection party as the fake data content detection strategy. The game execution module is configured to receive false data content detection strategies, drive the detection party to perform detection and blocking operations, and synchronously monitor the strategy execution status of the generator and the user. The data acquisition and feedback module is configured to collect strategy execution data, environmental feedback data, and detection result data in real time during the game process, and feed the collected data back to the participant modeling module and the payoff calculation module to support the updating of feature vectors and payoff functions, as well as subsequent strategy iterations.

Claims

1. A false data content game method based on a dynamic non-cooperative game model, characterized in that, Includes the following steps: S1. Construct a three-party game model that includes the generator, user, and detector of fake data, and define the strategy space for each of the generator, user, and detector. The strategy space is the set of strategies that each party can execute in the generation, targeted use, or detection of fake data content. S2. Based on the three-party game model, set the payoff function for the generator, user and detector respectively. The input parameters of the payoff function include the current strategy of each party and the interaction between strategies, and the output parameter is the payoff value of each party. S3. Initialize the strategies of the generator, user and detector respectively, and solve the equilibrium strategy in the three-party game through iterative method. Each iteration includes: calculating the current payoff value of each party based on the current strategy and corresponding payoff function of each party, and each party selects the strategy that maximizes its own payoff in its own strategy space based on the payoff value to update the current strategy. S4. When the strategy update process has met the preset convergence condition, the iteration is terminated, the fake data content game is completed, and the detection strategy after the iteration is terminated is determined as the fake data content detection strategy. In step S1, the strategy space of the fake data generator includes a combination of generation technology selection strategy and detection avoidance optimization strategy. The generation technology selection strategy is the configuration of generation technology for text, voice, image, video and / or multimodal fake data. The detection avoidance optimization strategy includes artifact elimination, feature camouflage, multi-technology fusion generation and / or dynamic switching of generation mode. The strategic space of those who use false data includes a combination of target positioning strategies, dissemination inducement strategies, and delivery method strategies. Among them, the target positioning strategy is target screening based on user profiles and / or multi-dimensional psychological characteristics; the dissemination inducement strategy includes social relationship guidance, emotional resonance inducement, and / or cognitive bias inducement; and the delivery method strategy includes covert channel delivery, batch delivery, and / or time-segmented precise delivery. The strategy space for fake data detection includes a combination of detection technology selection strategy, detection optimization strategy and response and handling strategy. The detection technology selection strategy is a configuration of dedicated detection technologies for text, voice, image, video and / or multimodal fake data. The detection optimization strategy includes model compression inference, multi-stage pipeline inference, dynamic threshold adjustment and / or multi-technology fusion detection. The response and handling strategy includes real-time blocking, source tracing, secondary verification and / or early warning push. In step S1, a three-party game model is constructed, including: constructing corresponding feature vectors for the generator, user, and detector, each feature vector consisting of multiple core feature parameters; the core feature parameters include the generator's generation technology type and avoidance ability parameters, the user's target group characteristics and induction strategy strength parameters, and the detector's detection technology coverage and efficiency parameters. In step S2, the payoff function for the fake data generator includes a payoff for successful generation, a cost for generation, a cost for avoiding detection, and a loss due to the risk of being detected; the payoff function for the fake data user includes a payoff for successful propagation, a cost for use, and a loss due to propagation blocking; the payoff function for the fake data detector includes a payoff for successful detection, a cost for detection, and a loss due to missed detection / false detection; the weight coefficients of each parameter in the payoff function are dynamically configured according to the game scenario.

2. The false data content game method of claim 1, wherein: In step S2, when calculating the corresponding payout values ​​for each party, the current strategies and corresponding feature vectors of the generator, user, and detector are used as input parameters for their respective payout functions.

3. The false data content game method of claim 1, wherein: In step S3, the Nash equilibrium solution mechanism is used to solve the equilibrium strategy in the three-party game. During the iteration process, the strategy updates of each party follow the optimal response principle of non-cooperative games. Any participant among the generator, user, and detector chooses the strategy that maximizes its own payoff given the current strategies of other participants.

4. The false data content game method of claim 3, wherein, In step S3, each party selects the strategy that maximizes its own benefit within its own strategy space based on the benefit value to update the current strategy. This includes: for any participant among the generator, user, and detector, calculating the benefit value corresponding to each candidate strategy in the participant's own strategy space based on the participant's current strategy, the current strategies of other participants, and the benefit function, and selecting the candidate strategy with the largest benefit value as the participant's strategy for the next iteration.

5. The false data content game method of claim 1, wherein, In step S4, the preset convergence conditions include: in two consecutive iterations, the change in the payoff value corresponding to the strategies of the generator, user, and detector is less than a preset threshold, or the number of iterations reaches the preset maximum number of iterations.

6. The false data content game method of claim 1, wherein: In step S4, the fake data content detection strategy is a specific strategy instance in the detection party's strategy space, including the combination configuration of detection technology selection strategy, detection optimization strategy and response and handling strategy; the fake data content detection strategy is used to guide the subsequent identification, analysis and handling of fake data content.

7. A false data content game system implementing the method of any of claims 1-6, characterized by Includes the following functional modules: The participant modeling module is configured to construct a three-party game model that includes the fake data generator, the user, and the detector, extract the core feature parameters of each participant, and construct and dynamically update the feature vectors of each participant. The strategy space management module is configured to store, classify, update, and recall the sets of strategies used by the generator, user, and detector in the generation, targeted application, and detection stages of fake data content, respectively. The revenue calculation module is configured to calculate the revenue values ​​of the generator, user, and detector in real time based on the current strategies of each participant, the interaction relationships between strategies, and the corresponding feature vectors through a revenue function. The dynamic equilibrium solution module is configured to initialize the strategies of each participant, and through iterative execution of strategy updates and benefit evaluation, control each participant to select the strategy that maximizes its own benefit within its own strategy space based on the benefit value, and terminate the iteration when the preset convergence condition is met, and output the strategy of the detection party as the fake data content detection strategy. The game execution module is configured to receive false data content detection strategies, drive the detection party to perform detection and blocking operations, and synchronously monitor the strategy execution status of the generator and the user. The data acquisition and feedback module is configured to collect strategy execution data, environmental feedback data, and detection result data in real time during the game process, and feed the collected data back to the participant modeling module and the payoff calculation module to support the updating of feature vectors and payoff functions, as well as subsequent strategy iterations.

Images