CN-121998631-A - Social simulator cognitive intervention method and system

CN121998631ACN 121998631 ACN121998631 ACN 121998631ACN-121998631-A

Abstract

The invention relates to the technical field of network security and discloses a social simulator cognitive intervention method and a social simulator cognitive intervention system, wherein the method comprises the steps of generating group topology based on user interaction data of a real social network; the method comprises the steps of carrying out initial evaluation signal and Bayesian optimization on the basis of a configuration proportion of agent nodes, carrying out dynamic adjustment on the configuration proportion of the agent nodes, reasoning the edge probability of the information of the failure of collecting information of each agent node in the group topology on the basis of a probability graph model, screening candidate intervention nodes according to the edge probability, deploying intervention type agent nodes on the basis of the adjusted configuration proportion, carrying out intervention on the candidate intervention nodes, evaluating the intervention result to generate an evaluation signal, and feeding back the evaluation signal to the dynamic adjustment step to replace the initial evaluation signal. The invention can effectively organize intervention subjects with different roles based on large-scale data of a complex network environment, dynamically optimize the throwing proportion, and carry out simulation prediction on the nodes to be intervened based on probabilistic graph reasoning.

Inventors

ZHANG YONGDONG
LIU WU
ZHOU XUAN

Assignees

中国科学技术大学

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (10)

1. A method of social simulator cognitive intervention, comprising: generating a group topology based on user interaction data of a real social network, wherein the group topology comprises a plurality of agent nodes playing different roles, and the agent nodes comprise at least one intervention agent node capable of negatively acting on the propagation of the unrealistic information; dynamically adjusting the configuration proportion of at least one intervention type agent node in the group topology based on an initial evaluation signal and a Bayesian optimization method; Reasoning the edge probability of the information of the inauthentic acquisition of each agent node in the group topology based on a probability graph model, screening candidate intervention nodes according to the edge probability, deploying intervention type agent nodes based on the adjusted configuration proportion, and performing intervention on the candidate intervention nodes, wherein the intervention type agent nodes are configured to generate intervention text content; The intervention result is evaluated to generate an evaluation signal, and the evaluation signal is fed back to the dynamic adjustment step to replace the initial evaluation signal.
2. The social simulator cognitive intervention method of claim 1, wherein the generating a group topology based on user interaction data of a real social network specifically comprises: Acquiring user interaction data, wherein the user interaction data comprises the connection relationship and state information of user nodes in a real social network, and the state information comprises structural features, behavior features, content features and credibility features of the user nodes; clustering the user interaction data by using a Gaussian mixture model to obtain feature distribution of a plurality of role types and an initial role duty ratio vector; determining the number of the intelligent agent nodes playing each role according to the total number of the intelligent agent nodes and the initial role duty ratio vector, and generating node characteristics of the corresponding intelligent agent nodes according to characteristic distribution sampling of different roles; And constructing the connection edge relationship among the intelligent agent nodes according to a group connection preference mechanism to form the group topology, wherein the group connection preference mechanism is used for enabling the topological characteristics of the intelligent agent nodes of each role type to accord with corresponding characteristic distribution.
3. The cognitive intervention method of claim 2, wherein the constructing the edge relationship between the agent nodes according to a group connection preference mechanism to form the group topology, the group connection preference mechanism is used to make the topology characteristics of the agent nodes of each character type conform to the corresponding characteristic distribution, specifically comprises: For any agent node Its role to be allocated is The population connection preference mechanism is defined as a function of scoring the following Is to be solved: ; Wherein, the And (3) with Is an agent node The output degree and the input degree of the model (1), And (3) with Is a role In structural features The probability distribution mean of the outbound degree and the probability distribution mean of the inbound degree, And (3) with Is a role In structural features Standard deviation of the outgoing degree and standard deviation of the incoming degree, Is a constant set.
4. The social simulator cognitive intervention method of claim 1, wherein the dynamically adjusting the configuration ratio of at least one intervention type agent node in the population topology based on the initial evaluation signal and the bayesian optimization method specifically comprises: Defining a system feedback function , wherein, A proportional role duty vector is configured for the agent node including the intervention, In order to evaluate the signal(s), For intervention cost function, λ is a trade-off parameter; The system feedback function is optimized by Bayesian optimization method Modeling the prior probability distribution of (a) as a gaussian process, computing based on a historical observation dataset Posterior probability distribution of (2); constructing an expected gain function based on a system feedback function and a corresponding posterior probability distribution by maximizing As a function of the expected gain of the variable And updating.
5. The cognitive intervention method of claim 4, wherein the system feedback function is optimized by a bayesian optimization method Modeling the prior probability distribution of (a) as a gaussian process, computing based on a historical observation dataset The posterior probability distribution of (2) specifically includes: Based on Bayesian optimization theory, firstly, a system feedback function is performed Is approximated as a gaussian process: ; Wherein, the The process of gaussian is represented by the expression, And (3) with Representing two different role duty cycle vectors respectively, As a function of the mean value of the function, As a covariance function; Based on Bayesian theory pair under the assumption of the Gaussian process Is estimated at the posterior probability of (2) In the intervention simulation of the wheel, observation data are obtained , wherein, Representing the role duty cycle vector adopted in the t-th intervention simulation of the system, Representing character duty ratio vector used in t-th round of intervention simulation The feedback value after that will The posterior probability distribution of (a) is estimated as Assume that Mean function in prior probability Zero, there is a mean of posterior probability distribution , It is indicated that the desire is to be met, The transpose is represented by the number, Representing a custom observed noise variance of the image, Is that Covariance vector with history role duty vector, Wherein As a function of the kernel, In the form of a matrix of nuclei, Elements of (a) , Is a matrix of units which is a matrix of units, In the case of an observed vector of values, Variance of posterior probability distribution , Representing the variance of the observed noise, Is that Is a self-covariance of (c).
6. The cognitive intervention method of claim 4, wherein the constructing the expected gain function based on the system feedback function and the corresponding posterior probability distribution is performed by maximizing As a function of the expected gain of the variable The updating method specifically comprises the following steps: Based on a system feedback function Defining an expected gain function to be optimized , wherein, It is indicated that the desire is to be met, , The i-th feedback value in the observed data is represented, and the analysis expression form of the expected gain function is obtained based on the Bayesian optimization theory: ; Intermediate variable , And (3) with A probability distribution function and a probability density function of a standard normal distribution respectively, Is that The average value and standard deviation of posterior probability distribution are optimized and solved to obtain the expected gain function Role duty cycle vector of time-step agent node : , ; Wherein, the Indicating that The dimensional unit probability simplex space, Representing the total number of roles of an agent node, i.e , ; Is given by expert restriction The maximum put upper bound of the configuration proportion of the role-like agent node, An index representing a class r character.
7. The social simulator cognitive intervention method of claim 1, wherein the reasoning of the edge probability of the inauthentic information acquired by each agent node in the group topology based on the probability graph model specifically comprises: training a deep neural network model based on a real event data set to calculate conditional propagation probability between two connected agent nodes according to state information of the two agent nodes as a weight of a corresponding directed edge in the group topology; iterative calculation is carried out on iterative information transmitted by neighbor nodes of each intelligent agent node based on a belief propagation algorithm and the conditional propagation probability so as to calculate the edge probability of the information of the failure of the intelligent agent node to acquire the information; The iteration information is calculated based on the priori probability of believing unreal information of the intelligent agent node, an edge function and iteration information transmitted by other neighbor nodes, wherein the edge function is defined by the conditional propagation probability.
8. The social simulator cognitive intervention method of claim 1, wherein the screening candidate intervention nodes according to the edge probability and deploying intervention agent nodes based on the adjusted configuration proportion, the intervention agent nodes configured to generate intervention text content, specifically comprises: screening agent nodes with edge probability higher than a set probability threshold from the group topology to form a high risk node set; Determining the deployment quantity of at least one type of intervention type intelligent agent node according to the configuration proportion of the adjusted intelligent agent node, wherein the intervention type intelligent agent node comprises an intelligent agent node of an infirm information doubter, an intelligent agent node of an infirm information propagator and an intelligent agent node of an information blocker; Hierarchical ordering is carried out according to the edge probability size and structural characteristics of candidate intervention nodes in a high risk node set, wherein the structural characteristics comprise the degree of departure, the degree of incidence, pageRank values and network distances between the candidate intervention nodes and an unreasonable information transmission source, a set number of candidate intervention nodes with earlier ordering are determined to be high-harm or unreasonable information source types, a set number of candidate intervention nodes with later ordering are determined to be high-susceptibility or potential transmission types, other candidate intervention nodes are determined to be high-influence or core transmission types, and intervention type intelligent nodes of corresponding types are matched and deployed according to the types of the candidate intervention nodes, wherein: For candidate intervention nodes classified as highly susceptible or potentially propagating, deploying the non-real information suspects agent node configured to generate intervention text content comprising questioning or alerting content based on a first alert word template; For candidate intervention nodes classified as high-impact or core propagation, deploying the anti-infirm information propagator agent node configured to generate intervention text content comprising a quote source for anti-refuting based on a second prompt word template; for candidate intervention nodes classified as high jeopardy or unreasonable information sources, deploying the information blocker agent node configured to generate intervention text content comprising account numbers or content restriction cues based on a third cue word template.
9. The cognitive intervention method of claim 1, wherein the evaluating the intervention result to generate an evaluation signal and feeding the evaluation signal back to the dynamic adjustment step to replace the initial evaluation signal comprises: Agent node index set with successful intervention ; Represent the first Individual agent nodes are at From the standpoint of the value of the infonnation at the time step, Representing that the information is extremely not believed to be in-true, Indicating that the information is extremely believed to be not real, For the start time of the system simulation, Is an analog time interval; An index representing an ith agent; the evaluation index includes: Intervention propagation rate : ; Depth of intervention propagation : Wherein Indicating the propagation path lengths of the intervening successful agent node and the intervening source agent node, Index of the interfered agent node and the interfered source agent node respectively; Intervention propagation breadth : ; The total number of all agent nodes; Viewpoint conversion rate : ; Emotion score tendency : ; Is an index set of all agent nodes; final evaluation signal The method comprises the following steps: ; Index for evaluation index; Wherein, the Representing the evaluation index Is used for the purpose of determining the value of the normalization, Is that And (5) corresponding weight.
10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 9 when the computer program is executed.

Description

Social simulator cognitive intervention method and system Technical Field The invention relates to the technical field of network security, in particular to a social simulator cognitive intervention method and system. Background Traditional cognitive intervention studies have significant limitations in practical applications. The traditional intervention modeling method mainly comprises four types of numerical modeling frameworks, namely an independent cascade Model (INDEPENDENT CASCADE Model), a linear threshold Model (Linear Threshold Model), a Game theory Model (Game-Theoretic Model) and an epidemic Model (EPIDEMICS MODEL). The models can theoretically describe information transmission and intervention mechanisms, but generally adopt an oversimplified transmission environment and participant behavior assumption, abstract the intervention process into numerical calculation, and cannot fully reflect complex interactions in a real social network. Meanwhile, the number of super parameters of the model is large, the over fitting phenomenon is easy to generate, and the generalization capability and practicability of the model are reduced. In recent years, social simulation systems based on large language models (Large Language Model, LLM) have begun to be applied to network event propagation and information intervention research, with information delivery by generating text content to simulate different intervention scenarios. However, the intervention means of the method are often limited to simple information injection, lack of pertinence and strategic, and are difficult to adapt to the intervention requirements of multi-role, multi-level and multi-strategy collaboration in the social network. Meanwhile, the simulation system generally ignores proper intervention proportion in the intervention process and predictability of believing unreal information (usually putting intervention contents after each agent finishes speaking) aiming at an unguarded node in a propagation path, and cannot realize dynamic optimization and accurate prediction of an intervention effect. In a complex social network environment, the effectiveness of an intervention strategy is affected by a variety of factors, including group structural features, individual cognitive states, information propagation dynamics, and the like. The existing research lacks automatic and scientific support in terms of strategy evaluation, so that intervention decisions often depend on experience or manual setting, and the repeatability and the quantifiability are lacking. The situation is urgent to need a comprehensive cognitive intervention method and system capable of being regulated, dynamically optimized and accurately predicted, and theoretical basis and data support are provided for policy design and implementation. Disclosure of Invention In order to solve the technical problems, the invention provides a social simulator cognitive intervention method and a social simulator cognitive intervention system, which can effectively organize intervention subjects with different roles based on large-scale data of a complex network environment, dynamically optimize the throwing proportion, and perform simulation prediction on nodes to be intervened based on probabilistic graph reasoning, and feed back various intervention indexes in real time, so that efficient, controllable and strategic cognitive intervention simulation is realized. In order to solve the technical problems, the invention adopts the following technical scheme: In a first aspect, the present invention provides a social simulator cognitive intervention method, comprising: generating a group topology based on user interaction data of a real social network, wherein the group topology comprises a plurality of agent nodes playing different roles, and the agent nodes comprise at least one intervention agent node capable of negatively acting on the propagation of the unrealistic information; dynamically adjusting the configuration proportion of at least one intervention type agent node in the group topology based on an initial evaluation signal and a Bayesian optimization method; Reasoning the edge probability of the information of the inauthentic acquisition of each agent node in the group topology based on a probability graph model, screening candidate intervention nodes according to the edge probability, deploying intervention type agent nodes based on the adjusted configuration proportion, and performing intervention on the candidate intervention nodes, wherein the intervention type agent nodes are configured to generate intervention text content; The intervention result is evaluated to generate an evaluation signal, and the evaluation signal is fed back to the dynamic adjustment step to replace the initial evaluation signal. In one embodiment, the generating the group topology based on the user interaction data of the real social network specifically includes: Acquiring user interaction data, wh