Unreal public opinion information spreading and control method based on multi-platform competition

Abstract

The invention discloses an unreal public opinion information dissemination and control method based on multi-platform competition, and relates to the technical field of network public opinion information dissemination. In order to solve the technical problems that an optimization strategy giving consideration to policy guidance and a traffic competition effect is lacked in the prior art, and platform users and traffic shares are difficult to maintain on the premise of ensuring that information is well supervised, the technical scheme provided by the invention comprises the following steps: establishing a double-layer directed social network model; based on the double-layer social network, establishing an information spreading rule and a state transition equation between the nodes; and based on the competition relationship among the target social network platforms, performing optimization solution on the control strategy of each target platform. And applying the solved management and control strategy of each target platform to a management system of each target social network platform, and optimizing and adjusting the management and control strategy of each target platform through numerical simulation analysis. The method is suitable for monitoring and controlling the unreal information in the social network platform.

Description

Technical Field

[0001] It involves the technical field of online public opinion information dissemination, specifically the dissemination and control of false public opinion information in multi-platform competition. Background Art

[0002] In today's information society, online social networking platforms have become an important channel for information dissemination. Especially with the popularization of mobile Internet and social media, the speed and breadth of information dissemination have been greatly improved. Existing research mainly focuses on how to monitor and control false information in social networking platforms to reduce its negative impact on society. In the study of the dissemination of false public opinion information, existing technologies usually adopt single-platform information monitoring and control methods, aiming to suppress the spread of false information through means such as information tagging, propagation path tracking and user behavior analysis. The application of these methods has achieved a certain degree of control over the dissemination of public opinion information within a single social networking platform. However, with the increase in the cross-platform dissemination of information on multiple platforms, the traditional single-platform information monitoring and control methods have gradually become ineffective.

[0003] Research shows that multi-platform competition leads to frequent migration of users between different platforms, which increases the complexity of information control. For example, after users are subject to strict content restrictions on one platform, they may turn to another platform to continue spreading false information. This fluidity between multiple platforms makes it difficult for existing control strategies to respond effectively. In addition, there are different management rules and user characteristics between different platforms, which also leads to differences in information dissemination and control strategies. Therefore, it is difficult for existing technologies to comprehensively and effectively control false information disseminated on multiple platforms in real time.

[0004] The main technical problems existing in the existing technology include the following: First, the public opinion control strategy of a single platform cannot adapt to the characteristics of cross-propagation of information between multiple platforms, and lacks effective means to control the flow of information between multiple platforms; second, the existing static control method cannot cope with the dynamic propagation of false public opinion information, and it is difficult to achieve real-time adjustment and optimization of control strategies; finally, due to the competitive relationship between multiple platforms, each platform often faces the risk of user loss and traffic share decline when suppressing false information. The existing technology lacks an optimization strategy that takes into account both policy guidance and traffic competition effects, and it is difficult to maintain platform users and traffic share under the premise of ensuring good supervision of information. The existence of these problems limits the application effect and applicability of the existing technology in the context of complex multiple platforms. Summary of the invention

[0005] In order to solve the technical problems that the existing technology lacks an optimization strategy that takes into account both policy guidance and traffic competition effects, it is difficult to maintain platform users and traffic share under the premise of ensuring good information supervision, and limits the application effect and applicability of the existing technology in a complex multi-platform context, the technical solution provided by the present invention is:

[0006] A method for solving the dissemination and control strategy of false public opinion information based on multi-platform competition, including:

[0007] Steps for building a two-layer directed social network model with at least two social network platforms;

[0008] Based on the two-layer social network, a step of establishing information propagation rules and state transfer equations between nodes;

[0009] Based on the competitive relationship between the target social network platforms, the steps of optimizing and solving the control strategy of each target platform are carried out.

[0010] Furthermore, a preferred embodiment is provided, wherein the two-layer directed social network model is constructed based on user nodes and connection relationships of the at least two social network platforms.

[0011] Furthermore, a preferred implementation is provided, which uses the SHILR propagation model to describe the propagation process of false public opinion information and establishes information propagation rules and state transfer equations between nodes.

[0012] Furthermore, a preferred implementation is provided, which optimizes and solves the control strategies of each target platform based on a zero-sum differential game model and through an iterative approximate dynamic programming algorithm.

[0013] A device for solving false public opinion information dissemination and control strategies based on multi-platform competition, comprising:

[0014] A module for building a two-layer directed social network model with at least two social network platforms;

[0015] Based on the two-layer social network, a module for establishing information propagation rules and state transfer equations between nodes is established;

[0016] A module that optimizes and solves the management and control strategies of each target social network platform based on the competitive relationship between the target social network platforms.

[0017] A method for disseminating and controlling false public opinion information based on multi-platform competition, comprising:

[0018] The obtained control strategies of each target platform are applied to the management system of each target social network platform, and the control strategies of each target platform are optimized and adjusted through numerical simulation analysis.

[0019] A false public opinion information dissemination and control device based on multi-platform competition, comprising:

[0020] A module that applies the solved control strategies of each target platform to the management system of each target social network platform, and optimizes and adjusts the control strategies of each target platform through numerical simulation analysis.

[0021] A computer storage medium is used to store a computer program. When the computer reads the computer program, the computer executes the method described.

[0022] A computer includes a processor and a storage medium. When the processor reads a computer program stored in the storage medium, the computer executes the method described.

[0023] The computer program product, as a computer program, implements the method described when the computer program is executed.

[0024] Compared with the prior art, the technical solution provided by the present invention is beneficial in that:

[0025] The construction of a two-layer directed social network model provides the ability to accurately model the process of false public opinion information dissemination. By establishing a competitive relationship between multiple platforms, it is possible to simulate the behavioral changes of users between different platforms and capture the information interaction process between platforms, thereby providing accurate basic data for subsequent public opinion information management and control. Compared with a single platform model, this multi-layer modeling method can better reflect the complex competition scenarios in reality and improve the application breadth and accuracy of the model.

[0026] The establishment of the dynamic model of false public opinion information propagation has achieved the dynamic analysis effect of the public opinion information propagation process. Through the dynamic description of the node state, the propagation path and speed of false public opinion between different platforms and different nodes can be accurately analyzed. Compared with the traditional static analysis method, this model can reflect the dynamic characteristics of public opinion propagation in real time, helping platforms and relevant departments to better understand the diffusion mechanism of false information, so as to formulate more effective response measures.

[0027] The design of the zero-sum differential game model based on platform competition significantly enhances the optimization effect of the control strategy between platforms. By considering the game relationship between platforms, the model enables each platform to maximize its own interests while suppressing false information. Compared with the strategy of only considering a single platform, it can better balance the contradiction between control and platform activity, and improve the overall competitiveness and user stickiness of the platform.

[0028] The use of iterative approximate dynamic programming algorithms to solve the game model brings about high efficiency and practical effects in strategy optimization. By using neural networks to approximate the value function, it can effectively deal with the high-dimensional and nonlinear problems of the system and ensure the efficiency of solving the game model in a complex environment. Compared with traditional analytical solution methods, iterative approximate algorithms have higher convergence and practicality, and can find the optimal strategy combination between platforms more quickly.

[0029] The application and optimization of information control strategies have achieved accurate labeling and dynamic adjustment of public opinion information. Through numerical simulation to verify the application effect of the strategy, the platform can dynamically adjust the control intensity according to the spread trend of false information, which not only avoids the situation of excessive labeling leading to user loss, but also ensures that false information is effectively suppressed. Compared with static control strategies, this dynamic adjustment mechanism is more flexible and targeted, and improves the overall effect of information control.

[0030] It is suitable for monitoring and controlling false information on social networking platforms to reduce its negative impact on society. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of a two-layer directed online social network structure;

[0032] Figure 2 It is a state transition diagram of the false public opinion information propagation model in a two-layer online social network;

[0033] Figure 3 It is a multi-platform false public opinion information edge state transition graph based on edge partitioning;

[0034] Figure 4 Schematic diagram of the solution process. DETAILED DESCRIPTION

[0035] The technical solution provided by the present invention is described in conjunction with the accompanying drawings, specifically:

[0036] Implementation method 1. The technical solution provided in this implementation method designs an effective multi-platform false public opinion information control strategy by constructing a two-layer directed social network model, based on the false public opinion information propagation dynamics model and the zero-sum differential game model. Each step is closely connected, and the output of the previous step becomes the input of the next step, so as to ensure the effectiveness and integrity of the entire control system. After implementing the above technical solution, effective control of false public opinion information can be achieved in the complex background of multi-platform competition, and theoretical support can be provided for the coordinated control between relevant departments and platforms.

[0037] Specific:

[0038] 1. Construction of a two-layer directed social network model

[0039] Step output: Two-layer directed social network model

[0040] 1.1 Platform nodes and structure definition: Two social network platforms A and B are established. The user nodes of each platform and their connection relationships constitute a two-layer directed social network. Each user has an account on both platforms, and platform A and platform B constitute directed networks respectively. By setting the in-degree and out-degree of the nodes, the degree distribution of platform A and platform B is constructed to ensure the complexity of the two-layer structure and the competitive relationship between the platforms.

[0041] 1.2 Assumptions of the two-layer structure: Assume that the number of users in each layer of the network is equal, but the node degree distribution may be different, and the number of followers and fans of users on different platforms is different. Therefore, each user has a corresponding node in platform A and platform B, that is, in the two-layer network, there is a one-to-one mapping relationship, but the degree distribution of user nodes on different platforms is different.

[0042] 1.3 Information transmission rules: Set five states for each node in the two-layer social network, including unknown state, hesitant state, propagation state, marked state, and immune state. By setting the node state transfer rules in the two-layer network, the information propagation path between different nodes is defined.

[0043] 2. Establishment of a dynamic model for the spread of false public opinion information

[0044] Step output: Dynamic model of false public opinion information propagation

[0045] 2.1 Introducing a two-layer network structure: Based on the two-layer social network constructed above, the SHILR propagation model is used to describe the propagation process of false public opinion information. The information propagation rules between platform A and platform B are defined, including the conversion relationship between propagation accounts, marked accounts, and immune accounts.

[0046] 2.2 Propagation state dynamics model: Establish the dynamics equation of the propagation of false public opinion information in the two-layer social network, including the dynamic description of the state change of each node. For each platform, establish the transfer equation between node states to describe the diffusion process of public opinion information between nodes.

[0047] 2.3 State transfer process modeling: According to the propagation of accounts in different states, the dynamic equations of in-degree and out-degree changes are constructed to describe the dynamic characteristics of the state transfer of accounts on each platform and the propagation process of false public opinion information.

[0048] 3. Design of zero-sum differential game model under platform competition

[0049] Step Output: Zero-Sum Differential Game Model

[0050] 3.1 Analysis of control costs: Analyze the costs that social network platforms may face in controlling false public opinion information, including the loss of active users, potential penalties from relevant departments, and the social costs of information control. These costs are important factors that platforms need to consider when formulating strategies.

[0051] 3.2 Game model construction: Based on the competitive relationship between platforms, a zero-sum differential game model is established to describe the dynamic strategy optimization process between platforms. When making decisions, each platform will consider the behavior of the opponent platform and formulate the optimal false speech marking strategy in order to maximize its own benefits while suppressing the spread of false public opinion.

[0052] 3.3 Solving the differential game model: By introducing the solution method of zero-sum differential game, the iterative approximate dynamic programming algorithm is used to optimize the strategies of each platform and find the strategy combination that minimizes the communication cost. Specifically, the neural network is used to approximate the value function and obtain the optimal control strategy of the platform.

[0053] 4. Application and optimization of information control strategies

[0054] Step output: Information control strategy

[0055] 4.1 Strategy output and platform application: The optimal false speech marking strategy for each platform is obtained through the above game model and applied to the management system of each social network platform. The platform marks false information in real time according to the optimal strategy to achieve effective control of public opinion information.

[0056] 4.2 Strategy effect analysis and optimization: Through numerical simulation and actual case analysis, the application effect of the proposed strategy in different scenarios is verified. Furthermore, by adjusting the parameters in the game model (such as the punishment intensity of relevant departments, user migration rate, etc.), the public opinion control strategy between platforms is optimized to balance the platform user activity and the clarity of cyberspace.

[0057] Implementation Method 2: Combination Figure 1-4 This embodiment is described, specifically:

[0058] Considering that online social network users can be active on multiple social network platforms at the same time in reality, this embodiment introduces a model of the propagation dynamics of false public opinion information on online social networks based on the double-layer network theory. While describing the intra-layer structure of the double-layer online social network, this model also characterizes the inter-layer structure of the double-layer online social network, and on this basis, associates the false public opinion information propagation dynamics of the two. This embodiment will first use the double-layer network theory to characterize the conceptual model of multiple online social network platforms. Secondly, this embodiment will introduce the network node state of the marked account to describe the information propagation rules based on the speech censorship system of the online social network platform; the marked account refers to the account identified by the platform as spreading false information, and these accounts will be subject to platform-specific restrictions, thereby affecting the further spread of information. Finally, this embodiment constructs a corresponding model of the propagation dynamics of false public opinion information on the Internet.

[0059] Construction of a two-layer directed online social network:

[0060] like Figure 1 As shown, the multi-platform online social network constructed in this embodiment is a duplex system with two-layer subnetworks; a duplex system refers to a system that can simultaneously transmit information in both directions, which refers to the relationship between two platforms where information dissemination can exist simultaneously. Each layer of subnetwork represents an online social network platform. Each platform is a directed information dissemination network consisting of accounts and the relationships between accounts. For simplicity, the two subnetworks are platform A and platform B.

[0061] Let the directed networks of platform A and platform B be and in, and represents the account set of online social network platform A and platform B, ε A and ε B They represent the directed edges between accounts on platform A and platform B. The two-layer network must satisfy the following assumptions:

[0062] (1) Both online social networking platforms A and B have N online accounts, namely

[0063] (2) Each layer of the subnetwork has a different intra-layer structure, that is, the degree distribution within the layer is different, which means that the number of followers and fans of the same user's account on different online social network platforms may be different;

[0064] (3) Each node in the two-layer subnetwork is one-to-one corresponding, which means that any network user has accounts on both online social networking platforms and can be active on either platform;

[0065] (4) Other assumptions about the network structure of online social network platforms A and B follow the following basic assumptions:

[0066] Based on the Barabási-Albert model, the topological structure of the directed network remains approximately stable during the propagation of false public opinion information in social networks;

[0067] There are no self-loops or repeated connections between any nodes in this directed network.

[0068] Based on the above, the nodes in the directed graph corresponding to the online social network can be divided into four states:

[0069] Susceptible (S): Online social network users who have not been exposed to false public opinion information so far and are still unaware of the public opinion topic.

[0070] Hesitating (H): Online social network users who have learned about false public opinion information but have not yet decided whether to express their opinions or views.

[0071] Infected (I): Online social network users who spread false public opinion information, opinions or views.

[0072] Recovered (R): Online social network users who are aware of false public opinion information but eventually lose interest in it and stop spreading it.

[0073] To simplify the model description, let S, H, I, and R represent the proportion of the above four types of nodes in the entire group and satisfy the following formula:

[0074]

[0075] Although there must be some disseminators in this directed social network to initiate the discussion and dissemination of false public opinion information, their group size is assumed to be negligible at time t = 0, that is, Based on this assumption, the initial value of the node state ratio of the directed online social network public opinion information dissemination system can be approximated as S(0)=1, H(0)=I(0)=R(0)=0.

[0076] For any online social network user i, both platform A and platform B should have corresponding node accounts. The probability that an online social network user has (j, k) degree in the account of platform A, that is, has j following accounts and k fan accounts, is p jk , where j∈{1,...,J}, k∈{1,...,K}; Similarly, let the probability that the account of the online social network user i on platform B has (x,y) degree be p xy, where x∈{1,...,X},y∈{1,...,Y}. Furthermore, the degree distribution of a two-layer online social network can be represented by a joint distribution probability. Let p jk，xy It represents the probability that a randomly selected network user i has (j, k) degrees on platform A and (x, y) degrees on platform B. It can be observed that the number of followers and fans of online social network user i on different accounts on platforms A and B are two completely independent events. Therefore, the joint distribution probability p jk，xy The following formula should be satisfied:

[0077] p jk，xy =p jk ·p xy (1)

[0078] According to the above definition of p jk 、p xy With p jk，xy We can further obtain the joint probability distribution of the two-layer online social network model. Based on the fact that most online social network users have multiple independent platform accounts, the joint probability generation function of the two-layer online social network model is as follows:

[0079]

[0080] Where a, b, c, d are placeholders for constructing PGF.

[0081] According to equation (2), the average in-degree 〈j〉, average out-degree 〈k〉 of online social network platform A and the average in-degree 〈x〉, average out-degree 〈y〉 of online social network platform B can be expressed as:

[0082]

[0083] Propagation rules:

[0084] Based on the two-layer online social network model, this implementation method will introduce in detail the propagation rules of false public opinion information on online social networks on the two-layer network model. The recently popular online social network control method - false speech marking is introduced into the information propagation process of the two-layer online social network, and the node status of the two-layer online social network is further refined. A user who uses an online social network can open an account on each network platform, but the account status of all network platforms is not necessarily the same. For example, an online social network user may choose to spread false public opinion information on platform A and remain silent on platform B. According to the nodes in the above directed graph, the platform account status in the two-layer online social network is as follows:

[0085] (1) Unknown account (Susceptible, S): An online social network platform account that has not received any relevant topic information pushed by platform A or platform B.

[0086] (2) Hesitating (H): A platform account owned by an online social network user who receives topic information pushed by platform A or platform B and decides to log in to the account of platform A or platform B for further review, but has not yet decided whether to spread the information, opinions or views related to the false topic.

[0087] (3) Infected (I): An online social networking platform account that spreads information, opinions, or views related to the topic.

[0088] (4) Labeled account (L): An online social network platform account that has spread topic-related information, opinions, or views but has been labeled as false speech by the online social network platform, and the platform account in this state is still contagious.

[0089] (5) Recovered (R): An online social network user loses interest in a topic and chooses to ignore or delete the relevant information posted by a platform account. The platform account whose information is ignored or deleted is called a recovered account.

[0090] To simplify the representation, let S A (t), H A (t), I A (t), L A (t), R A (t) represents the proportion of the number of accounts in each state in platform A to the total number of accounts, which satisfies the following formula:

[0091]

[0092] After dividing the network platform account states of the two-layer online social network model, this embodiment further describes the transition rules of the five states of the platform account. Figure 2 It shows the state transfer process of any online social network user's accounts on different platforms. The online social network user's account on platform A has j following accounts and k fan accounts, while on platform B, it has x following accounts and y fan accounts. As can be seen from the figure, the propagation rules of platform A and platform B in the two-layer online social network model are roughly similar. Therefore, in this embodiment, platform A will be taken as an example to describe in detail the propagation process of online social network false public opinion information in the two-layer online social network model, as follows:

[0093] (1) Most of the push notifications on the client side of the online social networking platform are content or topics posted by accounts followed by the account. Based on this, this implementation assumes that the opinions on a certain public opinion topic pushed by platform A are from the node account. Posted by the followed spreading account (I) or the tagged account (L).

[0094] (2) When online social network users access After receiving the information pushed by platform A, the user will decide whether to enter the client of platform A to further view the pushed related information based on the attractiveness of the pushed information. The probability that an online social network user receives false public opinion information posted by a propagation account (I) or a marked account (L) in a mobile terminal and decides to log in to the platform client to further view it is or

[0095] (3) Once the account receives and opens the platform A client to view the specific information, the status of the account will quickly change to Furthermore, the time required for any online social network user to decide whether to spread false public opinion information in the account of platform A follows the parameter β A The probability that an online social network user decides to spread relevant information, opinions, or views in his account on platform A after consulting the platform client is The probability that he chooses to remain silent and ignore the topic in his account on platform A is

[0096] (4) Online social networking platforms need to help relevant departments quickly suppress the outbreak of false public opinion, which is usually achieved by marking and punishing the accounts that spread false opinions. If an online social network user spreads information, opinions or views related to false public opinion in an account on platform A, the account will be punished by online social network platform A. of the platforms were marked as false statements, and the proportion of false statements marked on the platform changed over time.

[0097] At the same time, online social networking platform A will dynamically optimize its own management decisions according to the strategy of platform B. This optimization to obtain certain appropriate label ratios can be derived from the results of zero-sum differential games.

[0098] The decision sequence of the online social network platform A will determine the H A How to convert the account into the corresponding status interval (I A , L A or R A ).

[0099] (5) Online social network users may completely lose interest in the false public opinion topic and

[0100] (or ) at a rate that deletes topic-related information, opinions, or views (marked with L) posted by its platform A account A or unmarked A ), which is then converted into R A state.

[0101] The diffusion process of the online social network platform B can also be defined similarly.

[0102] The willingness of online social network users to spread information, opinions or views related to false public opinion depends on the strength of the corresponding information control of the network platform. If the online social network platform implements stricter information control, the platform users will consider whether the information sent by their own account will be marked as false speech and affect the account's attention, and then choose not to spread information related to the topic. Taking online social network platform A as an example, as false public opinion gradually develops, the user's willingness to spread on platform A is related to the control strength of platform A. Inversely proportional to the control strength of the rival platform B At the same time, the coefficient k A and k B Respectively represent the risk aversion of online social network users to platform A and platform B. In summary, an online social network user has a (j, k) account on platform A and an (x, y) account on platform B. The probability that the user chooses to spread this type of information on platform A at time t is

[0103] Considering the joint degree distribution of online social network platform A and platform B, the propagation probability of platform A and platform B should satisfy the following formula:

[0104]

[0105] Propagation dynamics model construction:

[0106] This implementation constructs a group of dynamic equations for the state changes of each account and the state changes of directed edges on the online social network platform A and platform B, respectively.

[0107] Taking platform A as an example, if we regard online social network platform A and platform B as two independent platforms, then at time t The reduction can be described as Among them, θ A (t) represents the percentage of S-state accounts with in-degree 1 in the online social network platform A at time t; and Then they respectively indicate that at time t, θ is satisfied. A (t) is defined as the proportion of directed edges whose predecessor nodes are in the states of I and L respectively. Figure 2 , the dynamic equations of the state changes of each account on the online social network platform A are:

[0108]

[0109] The state change process of each account on the online social network platform B can be derived in a similar way, and its dynamic equations are as follows:

[0110]

[0111] Where θ B --The percentage of S-status accounts with an in-degree of 1 in the online social network platform B at time t;

[0112] and ——Indicates that the condition is satisfied at time t Define the proportion of directed edges whose predecessor node states are I and L respectively.

[0113] The state transfer process is as follows Figure 3 shown.

[0114] The above is the derivation of the static value of the directed edge state in the two-layer online social network model.

[0115] Game Model Overview

[0116] The zero-sum differential game model between online social networking platforms should satisfy the following assumptions:

[0117] Hypothesis 1: Both online social network platforms A and B can rationally decide the proportion of false speech marking in the dissemination of false public opinion information;

[0118] Hypothesis 2: Online social networking platforms A and B have roughly the same enterprise size and strategic planning, and both platforms have the same prior knowledge of the topic;

[0119] Hypothesis 3: Online social networking platforms A and B attempt to dynamically optimize their own strategies based on guessing each other's decisions and adapt to the opponent's current strategy, trying to minimize their own costs;

[0120] Hypothesis 4: Strategies of Online Social Networking Platforms A and B and Can be further optimized over time.

[0121] Based on the above assumptions, this embodiment defines the cost representation generated by platform control in a zero-sum differential game model between online social network platforms.

[0122] This implementation method divides the costs of online social network platforms into three parts: loss of active users, potential penalties from relevant departments, and social costs of information control.

[0123] (1) Loss of active users

[0124] From the above, we can see that the user group is more willing to be active on rival platforms with less control, which has led to a large number of active users of online social network platforms being lost. The severity of the loss of active users of online social networks is related to the number of false speech marks ( and Without loss of generality, this embodiment uses a quadratic function to represent the expected active user loss of online social network platform A and platform B, that is, and

[0125] (2) Potential penalties from relevant departments

[0126] The loose control of online social networking platforms will lead to the rapid spread of false public opinions in cyberspace and social groups, which will not only have a huge negative impact on society, but also undermine the credibility and satisfaction of relevant departments, and bring great harm to national security and social stability. This implementation assumes that relevant departments and institutions will monitor online social networking platforms during the spread of a false public opinion topic and impose real-time penalties on affiliated companies that fail to control it. Without loss of generality, the potential punishment of the relevant departments of the company to which the online social networking platform belongs is proportional to the amount of false information on the platform. This implementation uses a quadratic function to model the penalty cost. and Potential penalties imposed on the companies that own online social networking platforms A and B by relevant authorities.

[0127] (3) Social costs of information control

[0128] In recent years, with the rapid development of automation and artificial intelligence functions such as deep learning and reinforcement learning in the field of natural language processing in social networks, some tasks of manual review of online social network information have been shared. However, most companies that own online social network platforms still retain manual review methods to conduct real-time supervision of platform content, and recognition functions such as automation and artificial intelligence can only serve as an auxiliary form of manual review. At present, Twitter's risk speech identification review is usually based on the following mechanism: problematic content is screened out through algorithms and artificial intelligence, and then manually read. This implementation assumes that the false speech marking mechanism of online social network platforms A and B follows Twitter's risk speech identification review mechanism, and that the marking costs of platforms A and B increase with time. and Monotonically increasing. Let online social network platform A and platform B make false statements

[0129] On the labeling cost function and

[0130] In summary, this embodiment assumes that online social network platform A and platform B are competing platforms. Online social network platform A hopes that more active users of platform B can be transferred to platform A, while the purpose of online social network platform B is just the opposite. Online social network platform A and platform B will guess each other's false public opinion information control strategy and formulate their own response strategies accordingly. Taking into account the characteristics of the competition between platform A and platform B, such as opposite goals, time-varying strategies, independent decisions, and confrontational decisions, this embodiment adopts zero-sum differential game theory to construct the game between platforms. According to the above platform control cost analysis, the following cost function is introduced:

[0131]

[0132] Where T is the end time of the spread of false public opinion information on online social networks;

[0133] α1 – the aversion coefficient of online social networking platforms to potential punishment by relevant departments;

[0134] α2——The aversion coefficient of online social networking platforms to the social costs of information control.

[0135] It can be found that the loss of active users is not the same as the potential punishment of relevant departments and the social cost of information control, but the cost variables in equation (10) have been expressed in the form of the same dimension. The choice of dimension is determined by the specific problem and has no direct relationship with the maximization of the cost function, which can be regarded as unit 1. At any time t, the online social network platform A aims to use its optimal strategy u A* Minimize the cost function J, where

[0136] In contrast, online social network platform B hopes that its optimal strategy u B* It is possible to maximize the cost function J.

[0137] in

[0138] According to the different status of online social network platform A and platform B,

[0139]

[0140] is the up-value cost function, let

[0141]

[0142] is the down-value cost function. According to equations (11) and (12), we can get the inequality:

[0143]

[0144] like This indicates that the optimal solution of the zero-sum differential game on the online social network platform exists, and the saddle point value (u A* ,u B* ) is called the optimal strategy pair for the zero-sum differential game on online social networking platforms.

[0145] Based on the upper cost function (11) and lower cost function (12) of the zero-sum differential game of the online social network platform, The upper cost function and lower cost function of the game should satisfy the condition

[0146]

[0147] and

[0148]

[0149] Based on equation (13) and equation (14), the following HJB functions of the upper value cost function and the lower value cost function can be obtained respectively:

[0150]

[0151] in

[0152]

[0153] is the upper Hamiltonian function,

[0154]

[0155] in,

[0156]

[0157] is the lower Hamiltonian function.

[0158] Equilibrium state analysis:

[0159] Solving the optimal strategy for multiple platforms in online social networks is very challenging. First, since the upper and lower cost function values ​​are not necessarily equal, this also means that there may be a situation where there is no saddle point. The saddle point refers to the equilibrium point in the game where neither party can reduce the loss by unilaterally adjusting the strategy; the optimal strategy for (u A* ,u B ) * does not necessarily exist. Secondly, even if the optimal strategy is (u A* ,u B ) *However, considering the platform competition, the spread of false public opinion information Z(t) is a complex nonlinear system, and the HJB equation is and The first-order partial derivative of is a nonlinear equation, and the optimal strategy is still difficult to solve. Therefore, the decision u of each platform on the intensity of false speech marking A and u B The expression cannot be displayed directly.

[0160] In order to optimize the false public opinion information control strategy of online social network platforms, this implementation method introduces an iterative approximate dynamic programming algorithm (ADP, Approximate Dynamic Programming) to solve the optimal strategy pair of zero-sum differential game on online social network platforms. Its main purpose is to use adaptive judgment design technology to solve the optimal strategy pair (u A* ,u B* ), so that any initial value Z(0) is positive definite to the origin, and at the same time the cost function reaches a saddle point. is the optimal strategy pair for the up-value cost function, ( u A , u B ) is the optimal strategy pair for the down-value cost function. According to the Min-max theorem, any control strategy pair (u A , u B ) are satisfied If the upper and lower cost function value intervals are 0, that is, both the upper and lower cost functions have strong convergence properties, then J 0 is a saddle point; if the upper and lower cost function value intervals are not 0, there is no saddle point, and the steady state of the system may exist between J and J Therefore, the optimal strategy pair of upper and lower value cost functions can be obtained by solving HJB equations (15) and (17).

[0161] The technical solution provided by the present invention is further described in detail above through several specific implementation modes in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the several specific implementation modes described above are not intended to be used as limitations on the present invention. Any reasonable modification and improvement of the present invention, combination of implementation modes and equivalent substitution within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for solving false public opinion information dissemination and control strategies based on multi-platform competition, characterized in that: include: Steps for building a two-layer directed social network model with at least two social network platforms; Based on the two-layer social network, a step of establishing information propagation rules and state transfer equations between nodes; Based on the competitive relationship between the target social network platforms, the steps of optimizing and solving the control strategy of each target platform are carried out.

2. According to claim 1, a method for solving false public opinion information dissemination and control strategies based on multi-platform competition is characterized in that: The two-layer directed social network model is constructed according to the user nodes of the at least two social network platforms and their connection relationships.

3. According to claim 1, a method for solving false public opinion information dissemination and control strategies based on multi-platform competition is characterized in that: The SHILR propagation model is used to describe the propagation process of false public opinion information, and the information propagation rules and state transfer equations between nodes are established.

4. According to claim 1, a method for solving false public opinion information dissemination and control strategies based on multi-platform competition is characterized in that: Based on the zero-sum differential game model and through the iterative approximate dynamic programming algorithm, the control strategies of each target platform are optimized and solved.

5. A device for solving false public opinion information dissemination and control strategies based on multi-platform competition, characterized in that: include: A module for building a two-layer directed social network model with at least two social network platforms; Based on the two-layer social network, a module for establishing information propagation rules and state transfer equations between nodes is established; A module that optimizes and solves the management and control strategies of each target social network platform based on the competitive relationship between the target social network platforms.

6. A method for disseminating and controlling false public opinion information based on multi-platform competition, characterized in that: include: The step of applying the control strategies of each target platform obtained by solving claim 1 to the management system of each target social network platform, and optimizing and adjusting the control strategies of each target platform through numerical simulation analysis.

7. A device for disseminating and controlling false public opinion information based on multi-platform competition, characterized in that: include: The control strategies of each target platform obtained by solving claim 1 are applied to the management system of each target social network platform, and a module is provided for optimizing and adjusting the control strategies of each target platform through numerical simulation analysis.

8. A computer storage medium for storing a computing program, characterized in that: When the computer reads the computer program, the computer executes the method of claim 6.

9. A computer, comprising a processor and a storage medium, characterized in that: When the processor reads the computer program stored in the storage medium, the computer executes the method of claim 6 .

10. A computer program product, being a computer program, characterized in that When the computer program is executed, the method according to claim 6 is implemented.

Images