Rumor propagation prediction method based on adversarial behavior and evolutionary game

Abstract

The invention belongs to the technical field of internet application, and relates to a rumor propagation prediction method based on adversarial behaviors and evolutionary games, which comprises the following steps: acquiring data of users and messages in a social network, and constructing real-time preference degrees of the users for rumor messages and rumor refuting messages according to the data of the users and the messages; using an evolutionary game theory to calculate the driving force of the user for forwarding the rumor message and the rumor refuting message according to the real-time preference degree; using a Rosenzweig-MacArthur model to calculate coexistence and confrontation coefficients of the rumor message and the rumor refuting message according to the willingness driving force; constructing a rumor propagation dynamics model SIPOR according to the willingness driving force to predict a rumor propagation situation; according to the method, the states of the users are divided into the common infection state, the rumor state and the rumor refuting state in the rumor propagation process, the propagation dynamics model SIPOR better conforming to rumor propagation facts is constructed by considering the interaction between the users in different states, and the rumor propagation prediction accuracy is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of Internet applications and relates to a rumor propagation prediction method based on adversarial behavior and evolutionary game. Background Art

[0002] Internet rumors refer to intentional and aggressive false information spread through the Internet. Such rumors seriously mislead the public's judgment and cause extremely bad social impact. As the threshold for rumor propagation is lowered, the spread of Internet rumors has become a phenomenon that cannot be ignored. Therefore, it is very important to study the mechanism of Internet rumor propagation and control the spread of rumors.

[0003] At present, the research on rumors in social networks can be divided into three major categories: disease transmission models, mathematical models, and machine learning algorithm models. In the direction of disease transmission models, infectious disease models are widely used in the study of rumor propagation dynamics. The rumor propagation mechanism is studied by dividing users in the propagation process into different behavioral states and combining the constructed state transition equation. In the direction of mathematical models, many scholars have constructed various classic rumor propagation dynamics models based on infectious disease models with the help of mathematical methods. In the direction of machine learning algorithm models, scholars use algorithm models to extract social network information features, detect rumor information and study its popularity, and then analyze and predict the spread of rumors.

[0004] With the deepening of social network research, scholars have realized that they can no longer use a single user awareness or rumor topic as a factor affecting the spread of rumors. Xiao et al. (Y.Xiao, W.He, T.Yang and Q.Li, "A Dynamic Information Dissemination Model Based on User Awareness and Evolutionary Games," in IEEE Transactions on Computational Social Systems, vol.10, no.5, pp.2837-2846, Oct.2023) proposed a dynamic information dissemination model based on user perception and evolutionary games. The model divides users into two states: low awareness and high awareness, and then analyzes the interaction between users at different cognitive levels through the evolutionary game mechanism, and then analyzes the dynamic information dissemination model. This paper shows that classifying users and then exploring their interactions can effectively analyze the dynamics of information dissemination.

[0005] Inspired by this paper, the present invention adopts a rumor propagation prediction method based on adversarial behavior and evolutionary game, introduces an intentional rumor state and an intentional rumor refuting state, and solves the problem that the existing technology does not fully consider the comprehensive impact of the topic (i.e., rumor message) itself and its derivative information (i.e., rumor refuting message) on the topic propagation, and cannot analyze the coexistence and confrontation of rumor message and rumor refuting message and the important factors affecting the spread of rumors by analyzing the adversarial behavior between the two groups. Summary of the invention

[0006] In order to solve the above-mentioned problems in the prior art, the present invention adopts a rumor propagation prediction method based on adversarial behavior and evolutionary game, comprising:

[0007] S1. Obtaining data of users and messages in the social network, the messages including rumor messages and rumor-refuting messages; calculating the real-time preference of users for rumor messages and rumor-refuting messages through a multivariate linear regression model based on the data of users and messages;

[0008] S2. Using the evolutionary game model, the driving force of users' willingness to forward rumor messages and rumor-refuting messages is calculated based on the users' real-time preference for rumor messages and rumor-refuting messages; using the Rosenzweig-MacArthur model, the coexistence and confrontation coefficients of rumor messages and rumor-refuting messages are calculated based on the users' willingness to forward rumor messages and rumor-refuting messages;

[0009] S3. Based on the driving force of users' willingness to forward rumor messages and refute rumor messages, a rumor propagation dynamics model SIPOR is constructed to predict the rumor propagation trend.

[0010] Calculating the user's real-time preference for messages includes: calculating user factors based on user data, and calculating topic factors based on the data of rumor messages and rumor-refuting messages; using a multivariate linear regression model to analyze the weights of user factors and topic factors; and weightedly combining user factors and topic factors according to the weights to obtain the user's real-time preference for rumor messages and rumor-refuting messages.

[0011] The evolutionary game model includes two game decisions, namely forwarding rumor messages and forwarding rumor-refuting messages; calculating the user's willingness driving force to forward rumor messages and rumor-refuting messages includes: calculating the user's behavioral benefit function for the two game decisions based on the user's real-time preference for rumor messages and rumor-refuting messages; calculating the user's willingness driving force for the two game decisions based on the behavioral benefit function.

[0012] Users include: ordinary infected users, rumor users, and rumor-refuting users; the calculation of the user's behavioral benefit function for each game decision includes: calculating the proportion of "forwarding rumor messages" among the user's neighboring users x i The ratio of "forwarding rumor-refuting messages" is 1-x i, i∈{I,P,O}, obtain the purpose-driven gain factor ε when the rumor user and rumor-refuting user forward the rumor message i and the purpose-driven gain factor ω when forwarding rumor-refuting messages i ; Among them, I, P, and O are common infected users, rumor users, and rumor-refuting users, respectively;

[0013] If the user is a common infected user, then according to the ratio x I , 1-x I Calculate the behavioral benefit functions of ordinary infected users for the decisions of "forwarding rumor messages" and "forwarding rumor-refuting messages" respectively;

[0014] If the user is a rumor user, then according to the ratio x P , 1-x P and the drive gain factor ε P ,ω P Calculate the behavioral benefit function of the decision of "forwarding rumor messages" and the decision of "forwarding rumor-refuting messages";

[0015] If the user is a rumor-refuting user, then according to the ratio x O , 1-x O and the drive gain factor ε O ,ω O Calculate the behavioral benefit function of the decision to "forward rumor messages" and the decision to "forward rumor-refuting messages".

[0016] Calculating the coexistence and confrontation coefficients of rumor information and rumor refutation includes:

[0017]

[0018] Among them, Coexistence(t) and Contrast(t) represent the coexistence coefficient and confrontation coefficient of rumor messages and rumor-refuting messages at time t, respectively. 1 represents the growth rate of rumor information at time t, α 2 represents the growth rate of rumor-refuting messages at time t, k R It represents the maximum average driving force of rumor message, and υ is the interaction coefficient between rumor message and rumor-refuting message.

[0019] According to the driving force of users' willingness to forward rumors and refute rumors, the rumor propagation dynamics model SIPOR is constructed to predict the rumor propagation trend, including:

[0020] S31, classifying users into susceptible state S, common infection state I, rumor state P, rumor refutation state O and immune state R according to the forwarding state of the users for the message in the social network;

[0021] S32. Construct a multi-user type rumor propagation mechanism, and use the mean field theory to construct a rumor propagation dynamics model SIPOR based on the multi-user type rumor propagation mechanism;

[0022] S33. According to the driving force of users' willingness to forward rumor messages and refute rumor messages, the rumor propagation dynamics model SIPOR is used to predict the rumor propagation trend.

[0023] The multi-user type rumor propagation mechanism includes:

[0024] Users in susceptible state S transfer to common infection state I with probability α, users in common infection state I transfer to rumor state P with probability β, and transfer to rumor refutation state O with probability γ; users in rumor state P and rumor refutation state O transfer to rumor state P and rumor refutation state O with probability γ, respectively. η moves to the common infection state I;

[0025] When a user forwards a rumor message more than k times in a row, his state changes from the common infection state I to the rumor state P; when a user forwards a rumor-refuting message more than q times in a row, his state changes from the common infection state I to the rumor-refuting state O. When a user in rumor state P forwards a rumor-refuting message, his state changes to the common infection user I; when user O in rumor-refuting state forwards a rumor message, his state changes to the common infection user I; users in common infection state I, rumor state P, and rumor-refuting state O eventually change to the immune state R with probabilities ε, μ, and ω, respectively.

[0026] Among them, α, β, γ∈[0,1], α+β+γ<1, k, q, ε, μ, and ω are adjustable parameters.

[0027] The rumor propagation dynamics model SIPOR is:

[0028]

[0029] in, Represents the dynamic changes of the susceptible state S; Indicates the dynamic changes of the common infection state I, represents the dynamic change of rumor state P, Indicates the dynamic changes of rumor-refuting state O, represents the dynamic changes of the immune state R, S(t), I(t), P(t), O(t), and R(t) respectively indicate that the user is in the susceptible state S, the common infection state I, the rumor state P, the rumor-refuting state O, and the immune state R at time t.

[0030]

[0031]

[0032] Among them, ξR (t) is the probability of a user forwarding a rumor message at time t, ξ Anti-R (t) is the probability that the user forwards the rumor-refuting message at time t.

[0033] Calculating the probability of a user forwarding a rumor message and the probability of a rumor-refuting message at time t includes: calculating the probability θ of an ordinary infected user forwarding a rumor message at time t based on the willingness driving force of the ordinary infected user R (t) and the probability of rumor-refuting news θ Anti-R (t), the probability ρ of the rumor user forwarding the rumor message at time t is calculated based on the rumor user’s willingness driving force R (t) and the probability of rumor-refuting news ρ Anti-R (t), the probability ψ of the rumor-refuting user forwarding the rumor message at time t is calculated based on the rumor-refuting user’s willingness driving force R (t) and the probability of rumor-refuting news ψ Anti-R (t), according to the probability θ R (t),ρ R (t),ψ R (t) Calculate the probability of a user forwarding a rumor message at time t, according to the probability θ Anti-R (t),ρ Anti-R (t),ψ Anti-R (t) Calculate the probability that a user forwards a rumor-refuting message at time t.

[0034] Beneficial effects:

[0035] 1. The present invention introduces the Rosenzweig-MacArthur equation to construct the game change trend between the rumor group and the rumor refuting group, thereby obtaining the coexistence and confrontation of rumor messages and rumor refuting messages in the entire rumor propagation life cycle. According to the coexistence and confrontation of rumor messages and rumor refuting messages, the development direction of rumor propagation can be analyzed more accurately; 2. In the process of rumor propagation, the present invention divides the user's state into ordinary infection state I, rumor state P, and rumor refuting state O, considers the interaction between users in different states, and uses evolutionary game theory to construct a propagation dynamics model SIPOR that is more in line with the facts of rumor propagation. The model can more accurately analyze and predict the behavior of users in each state during the rumor propagation process, and then accurately obtain the development direction of rumor propagation, thereby improving the accuracy of rumor propagation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flowchart of a rumor propagation prediction method based on adversarial behavior and evolutionary game provided by an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of a rumor propagation prediction method based on adversarial behavior and evolutionary game provided by an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of measuring user message real-time preference provided by an embodiment of the present invention;

[0039] Figure 4 A schematic diagram of the analysis of user confrontation behavior and rumor-refuting game situation provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] like Figure 1 , Figure 2 As shown, the present invention proposes a rumor propagation prediction method based on adversarial behavior and evolutionary game. The input of the present invention is basic information of users in social networks, user historical behavior information, user relationship network, and rumor topic information. The output is important factors affecting the spread of rumors, user momentary state set, user state transition set, and the coexistence and adversarial evolution relationship between rumor messages and rumor-refuting messages.

[0042] The specific steps include:

[0043] S1. Obtaining data of users and messages in the social network, the messages including rumor messages and rumor-refuting messages; calculating the real-time preference of users for rumor messages and rumor-refuting messages through a multivariate linear regression model based on the data of users and messages;

[0044] Measuring the real-time preference of user messages requires considering user factors and topic factors. Among them, user factors include user activity, user information acquisition ability, and user topic matching. For topic factors, it is necessary to consider the real-time popularity of topic messages; the regression coefficient is fitted through past data through the multivariate linear regression model, and then the real-time preference of user messages is calculated based on user factors and topic factors.

[0045] like Figure 3 As shown, specifically, calculating the user's real-time preference for the message includes:

[0046] S11: Calculate user factor Var based on user data in (u i ), calculate the topic factor Var based on the message data msg (u i );

[0047] User real-time preference is used to describe whether users decide to participate in rumor topics or forward which type of messages based on their own factors after being influenced by rumor topics.

[0048] Var user (u i )=Act(u i )×Obt(u i )×Pair(u i )(1)

[0049]

[0050] Among them, Act(u i ) is the user activity, Obt(u i ) is the user information acquisition capability, Pair(u i ) is the user interest-topic matching degree, Hot(t) is the real-time heat of the message, w(u i ,u j ) represents user u i The weight of the source of the received message, w(u i ,u j )>0 indicates rumor news, w(u i ,u j )<0 indicates rumor-refuting news, u j is the source user of the message, and n is the number of source users of the message.

[0051] S12: Regression coefficients of fitted multivariate linear regression model

[0052] Considering the obvious advantages of the multivariate linear regression model in processing multivariate models, the present invention uses the multivariate linear regression model to analyze how user factors and topic factors affect the user's real-time preference for the two types of messages:

[0053]

[0054]

[0055] in, Represents user u i The preference for rumor information at time t, Represents user u i The preference for rumor-refuting news at time t, β 1 , β 2 , β 3 Var is the regression coefficient obtained by fitting the historical real data. msg (u i ) R The topic factor of the rumor message, Var msg (u i) Anti-R Indicates the topic factor of rumor-refuting news.

[0056] S2. Using the evolutionary game model, the driving force of users' willingness to forward rumor messages and rumor-refuting messages is calculated based on the users' real-time preference for rumor messages and rumor-refuting messages; using the Rosenzweig-MacArthur model, the coexistence and confrontation coefficients of rumor messages and rumor-refuting messages are calculated based on the driving force of users' willingness to forward rumor messages and rumor-refuting messages;

[0057] like Figure 4 As shown, according to the forwarding status of messages by users in social networks, users are divided into: ordinary infected users, rumor users and rumor refuting users; among them, ordinary users are topic participants without obvious bias and purpose, rumor users are rumor users with obvious bias, and rumor refuting users are rumor refuting users with obvious bias.

[0058] The present invention defines two game strategies for users, namely "forwarding rumor messages" and "forwarding rumor-refuting messages".

[0059] For ordinary infected users, when exposed to multiple types of rumor messages, users will play a psychological game based on their own cognition and the mutual influence of various messages, and finally make a decision. I , 1-x I Indicates the proportion of "forwarding rumor messages" and "forwarding rumor-refuting messages" among the user's neighboring nodes. Among them, 0≤x I ≤1.

[0060] For rumor users, use x P , 1-x P Indicates the proportion of "forwarding rumor messages" and "forwarding rumor-refuting messages" among the user's neighboring nodes. Among them, 0≤x P ≤1.

[0061] For rumor-refuting users, use x O , 1-x O Indicates the proportion of "forwarding rumor messages" and "forwarding rumor-refuting messages" among the user's neighboring nodes. Among them, 0≤x O ≤1.

[0062] The profit function of the three game entities, ordinary users, rumor users and rumor-refuting users, is shown in Table 1.

[0063] Table 1 Profit function

[0064]

[0065] Among them, pay R(I) 、pay Anti-R(I)They represent the revenue functions of ordinary infected users forwarding rumors and forwarding rumor refutation, pay R(P) 、pay Anti-R(P) They represent the revenue function and payoff of rumor users for forwarding rumors and refuting rumors respectively. R(H) 、pay Anti-R(H) They represent the revenue functions of rumor-refuting users forwarding rumors and refuting rumors, respectively, and ε P , ε O They represent the purpose-driven gain factors when rumor users and rumor-refuting users forward rumor messages, ε P >1,0<ε O <1.ω P ,ω O Respectively represent the purpose-driven gain factors when rumor users and rumor-refuting users forward rumor-refuting messages, 0<ω P <1,ω O >1.

[0066] Preferably, ε P The value is 1.6, ε O The value is 0.3, which can be adjusted according to the actual situation; ω P Take the value 0.6, ω O The value is 2.1 and can be adjusted accordingly based on actual conditions.

[0067] The three types of users invisibly generate three-party game behavior when participating in rumor topics. Considering the advantages of evolutionary game theory in studying multi-party game behavior and enabling gamers to obtain maximum benefits, the present invention uses the maximum benefits obtained by each type of user through the game to quantify their forwarding behavior. Generally, the sigmoid function is used to map the user's forwarding behavior to (0,1). Among them, the closer the mapping result is to 1, the stronger the user's willingness to forward the topic. The details are as follows:

[0068] For ordinary infected users, the driving force of the willingness to forward rumors and forward rumor refutation is Tend R(I) 、Tend Anti-R(I) for:

[0069]

[0070]

[0071] For rumor users, the driving force behind the willingness to forward rumors and forward rumor refutation is Tend R(P) 、Tend Anti-R(P) for:

[0072]

[0073]

[0074] For rumor-refuting users, the driving force behind the willingness to forward rumors and to forward rumor-refuting is Tend R(H) 、Tend Anti-R(H) for:

[0075]

[0076]

[0077] Since there is a confrontation between rumor messages and rumor-debunking messages during the spread of rumor topics, some ordinary users will be transformed into rumor users or rumor-debunking users. This situation is similar to the relationship between predators and prey in the biological field. Specifically, rumor users are analogized to predators, and rumor-debunking users are analogized to prey, and there is a natural confrontation between them. Therefore, the present invention calculates the coexistence and confrontation coefficients Coexistence(t) and Contrast(t) of rumor messages and rumor-debunking by introducing the Rosenzweig-MacArthur equation:

[0078]

[0079]

[0080] Among them, α 1 represents the growth rate of rumor information at time t, α 2 represents the growth rate of rumor-refuting messages at time t, k R It represents the maximum value (saturated average driving force) of the average willingness to forward rumor messages among ordinary infected users, rumor users and rumor-refuting users, and υ is the interaction coefficient between rumor messages and rumor-refuting messages.

[0081] The growth rates of rumor messages and rumor-refuting messages are defined as follows:

[0082]

[0083]

[0084] Among them, RumorCount represents the number of rumor messages in a certain time slice, AntiRumorCount represents the number of rumor-refuting messages in a certain time slice, and t 1 With t 2 Represent different time slices (t 2 >t 1 ).

[0085] The effect coefficient between rumor news and rumor-refuting news:

[0086]

[0087] Among them, Hot R (t), Hot Anti-R (t) represents the popularity of rumors and rumor-refuting messages at time t.

[0088] S3. According to the driving force of users’ willingness to forward rumors and refute rumors, a rumor propagation dynamics model SIPOR is constructed to predict the rumor propagation trend.

[0089] In the process of rumor topic propagation, in order to make the propagation trend of rumor topics more specific, and taking into account the rumor status and rumor refutation status formed by different user groups with certain intentions or due to cognitive differences, the present invention constructs a rumor propagation dynamics model SIPOR based on user confrontation behavior and evolutionary game on the basis of the classic SIR model, so as to more accurately analyze the rumor propagation trend.

[0090] Specifically, the steps to predict the spread of rumors include:

[0091] S31. The users in the model are divided into five states: susceptible state S, common infection state I, rumor state P, rumor refuting state O, and immune state R.

[0092] S32. Construct a multi-user type rumor propagation mechanism, and use the mean field theory to construct a rumor propagation dynamics model SIPOR based on the multi-user type rumor propagation mechanism;

[0093] In order to make the model analysis more accurate and reliable, some factors that have little impact on the model results and may cause unnecessary complexity in the analysis process are excluded when establishing the model, and the following assumptions are obtained:

[0094] Assuming that the total number of users participating in the rumor topic propagation model remains unchanged, S(t)+I(t)+P(t)+O(t)+R(t)=1.

[0095] Since the spread of rumor messages takes time and is affected by the network environment and rumor-refuting messages, the number of user nodes in the rumor state will no longer increase after reaching a certain number, that is, the rumor state nodes are saturated.

[0096] When a user node continuously forwards rumor messages or rumor-refuting messages and reaches a certain threshold, its status changes from ordinary infection status to rumor status or rumor-refuting status, and at the same time, the benefit of forwarding the corresponding message is increased by the purpose-driven gain factor.

[0097] As the popularity of the rumor topic changes, users in the topic network will eventually change from various states to an immune state.

[0098] Based on the above assumptions, the criteria for formulating the multi-user type rumor propagation mechanism of the present invention are as follows:

[0099] The susceptible state S transfers to the common infection state I with a probability of α, and the common infection state I transfers to the rumor state P with a probability of β and to the rumor refutation state O with a probability of γ; the rumor state P and rumor refutation state O are respectively The probability of η transferring to the common infection state I; among them, α, β, γ∈[0,1]; since some nodes have never participated in the topic propagation during the entire topic propagation process, therefore, α+β+γ<1.

[0100] When a user forwards a rumor message more than k times in a row, its status changes from a common infected node to a rumor node; when a user forwards a rumor-refuting message more than q times in a row, its status changes from a common infected node to a rumor-refuting node. When the rumor user forwards a rumor-refuting message again, its status changes to a common infected user; the same applies to the rumor-refuting user; k and q are adjustable parameters.

[0101] As the popularity of various types of messages changes, common infection nodes, rumor nodes, and rumor-refuting nodes will eventually transform into immune nodes with probabilities of ε, μ, and ω, respectively.

[0102] Based on the multi-user type rumor propagation mechanism principle, the following dynamic equations can be constructed:

[0103]

[0104] Since the probability of a user changing from a rumor state or rumor-refuting state to a common infection state and then changing to a rumor state or rumor-refuting state is extremely low, all users will eventually become immune. i There are n neighbor users, and the probability that m of them will transition to another state follows a binomial distribution:

[0105]

[0106] From formula (23), we can see that for an ordinary infected user, the probability θ of forwarding a rumor message and refuting a rumor message at time t is R (t),θ Anti-R (t) are:

[0107]

[0108]

[0109] For rumor users, the probability of forwarding rumor messages and refuting rumor messages at time t is ρ R (t),ρ Anti-R (t) are:

[0110]

[0111]

[0112] For rumor-refuting users, the probability of forwarding rumor messages and rumor-refuting messages at time t is ψ R (t),ψ Anti-R (t) are:

[0113]

[0114]

[0115] In summary, the probability that any user forwards a rumor message or refutes a rumor message at time t is R (t), ξ Anti-R (t) are:

[0116] ξ R (t)=I(t)×θ R (t)+P(t)×ρ R (t)+O(t)×ψ R (t) (30)

[0117] ξ Anti-R (t)=I(t)×θ Anti-R (t)+P(t)×ρ Anti-R (t)+O(t)×ψ Anti-R (t) (31)

[0118] In the analysis process, the interaction between the user and all other users is relatively complex, and if the interaction between all other users is regarded as an average force, the analysis of each user becomes relatively simple and clear. Therefore, the present invention uses the mean field theory to simplify the interaction between users, and the dynamic equations are as follows:

[0119]

[0120] S33. According to the driving force of users' willingness to forward rumor messages and refute rumor messages, the rumor propagation dynamics model SIPOR is used to predict the rumor propagation trend.

[0121] The output of the rumor propagation dynamics model SIPOR is the user's momentary state set and the user's state transition set; the relevant departments can understand the propagation trend of the rumor topic based on the user's momentary state set, the user's state transition set, and the coexistence and confrontation coefficients of rumor messages and rumor debunking, and choose the appropriate time to release rumor debunking information through the user's state sets at different times.

[0122] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A rumor propagation prediction method based on adversarial behavior and evolutionary game, characterized in that: include: S1. Obtaining data of users and messages in the social network, the messages including rumor messages and rumor-refuting messages; calculating the real-time preference of users for rumor messages and rumor-refuting messages through a multivariate linear regression model based on the data of users and messages; S2. Using the evolutionary game model, the driving force of users' willingness to forward rumor messages and rumor-refuting messages is calculated based on the users' real-time preference for rumor messages and rumor-refuting messages; using the Rosenzweig-MacArthur model, the coexistence and confrontation coefficients of rumor messages and rumor-refuting messages are calculated based on the users' willingness to forward rumor messages and rumor-refuting messages; S3. According to the driving force of users' willingness to forward rumor messages and refute rumor messages, a rumor propagation dynamics model SIPOR is constructed to predict the rumor propagation trend.

2. The rumor propagation prediction method based on adversarial behavior and evolutionary game according to claim 1 is characterized in that: Calculating the user's real-time preference for messages includes: calculating user factors based on user data, and calculating topic factors based on the data of rumor messages and rumor-refuting messages; using a multivariate linear regression model to analyze the weights of user factors and topic factors; and weightedly combining user factors and topic factors according to the weights to obtain the user's real-time preference for rumor messages and rumor-refuting messages.

3. The rumor propagation prediction method based on adversarial behavior and evolutionary game according to claim 1 is characterized in that: The evolutionary game model includes two game decisions, namely forwarding rumor messages and forwarding rumor-refuting messages; Calculating the driving force of users' willingness to forward rumor messages and rumor-refuting messages includes: calculating the user's behavioral benefit function for two game decisions based on the user's real-time preference for rumor messages and rumor-refuting messages; The user's willingness driving force for the two game decisions is calculated based on the behavioral benefit function.

4. The rumor propagation prediction method based on adversarial behavior and evolutionary game according to claim 3 is characterized in that: Users include: Ordinary infected users, rumor users, and rumor-refuting users; Calculating the user's behavioral benefit function for each game decision includes: calculating the proportion of "forwarding rumor messages" among the user's neighboring users x i The ratio of "forwarding rumor-refuting messages" is 1-x i , i∈{I,P,O}, obtain the purpose-driven gain factor ε when the rumor user and rumor-refuting user forward the rumor message i and the purpose-driven gain factor ω when forwarding rumor-refuting messages i ; Among them, I, P, and O are common infected users, rumor users, and rumor-refuting users, respectively; If the user is a common infected user, then according to the ratio x I , 1-x I Calculate the behavioral benefit functions of ordinary infected users for the decisions of "forwarding rumor messages" and "forwarding rumor-refuting messages" respectively; If the user is a rumor user, then according to the ratio x P , 1-x P and the drive gain factor ε P ,ω P Calculate the behavioral benefit function of the decision of "forwarding rumor message" and the decision of "forwarding rumor-refuting message"; If the user is a rumor-refuting user, then according to the ratio x O , 1-x O and the drive gain factor ε O ,ω O Calculate the behavioral benefit function of the "forward rumor message" decision and the "forward rumor-refuting message" decision.

5. The rumor propagation prediction method based on adversarial behavior and evolutionary game according to claim 1 is characterized in that: Calculating the coexistence and confrontation coefficients of rumor information and rumor refutation includes: Among them, Coexistence(t) and Contrast(t) represent the coexistence coefficient and confrontation coefficient of rumor messages and rumor-refuting messages at time t, respectively. α1 represents the growth rate of rumor messages at time t, α2 represents the growth rate of rumor-refuting messages at time t, and k R It represents the maximum average driving force of rumor message, and υ is the interaction coefficient between rumor message and rumor-refuting message.

6. The rumor propagation prediction method based on adversarial behavior and evolutionary game according to claim 1 is characterized in that: According to the driving force of users' willingness to forward rumors and refute rumors, the rumor propagation dynamics model SIPOR is constructed to predict the rumor propagation trend, including: S31, classifying users into susceptible state S, common infection state I, rumor state P, rumor refutation state O and immune state R according to the forwarding state of the users for the message in the social network; S32. Construct a multi-user type rumor propagation mechanism, and use the mean field theory to construct a rumor propagation dynamics model SIPOR based on the multi-user type rumor propagation mechanism; S33. According to the driving force of users' willingness to forward rumor messages and refute rumor messages, the rumor propagation dynamics model SIPOR is used to predict the rumor propagation trend.

7. The rumor propagation prediction method based on adversarial behavior and evolutionary game according to claim 6 is characterized in that: The multi-user type rumor propagation mechanism includes: Users in susceptible state S transfer to common infection state I with probability α, users in common infection state I transfer to rumor state P with probability β, and transfer to rumor refutation state O with probability γ; users in rumor state P and rumor refutation state O transfer to rumor state P and rumor refutation state O with probability γ, respectively. α moves to the normal infection state I; When a user forwards a rumor message more than k times in a row, his state changes from the common infection state I to the rumor state P; when a user forwards a rumor-refuting message more than q times in a row, his state changes from the common infection state I to the rumor-refuting state O. When a user in rumor state P forwards a rumor-refuting message, his state changes to the common infection user I; when user O in rumor-refuting state forwards a rumor message, his state changes to the common infection user I; users in common infection state I, rumor state P, and rumor-refuting state O eventually change to the immune state R with probabilities ε, μ, and ω, respectively. Among them, α, β, γ∈[0,1], α+β+γ<1, k, q, ε, μ, and ω are adjustable parameters.

8. A rumor propagation prediction method based on adversarial behavior and evolutionary game according to claim 7, characterized in that: The rumor propagation dynamics model SIPOR is: in, Represents the dynamic changes of the susceptible state S; Indicates the dynamic changes of the common infection state I, represents the dynamic change of rumor state P, Indicates the dynamic changes of rumor-refuting state O, represents the dynamic changes of the immune state R, S(t), I(t), P(t), O(t), and R(t) respectively indicate that the user is in the susceptible state S, the common infection state I, the rumor state P, the rumor-refuting state O, and the immune state R at time t.

9. The rumor propagation prediction method based on adversarial behavior and evolutionary game according to claim 8 is characterized in that: Among them, ξ R (t) is the probability of a user forwarding a rumor message at time t, ξ Anti-R (t) is the probability that the user forwards the rumor-refuting message at time t.

10. The rumor propagation prediction method based on adversarial behavior and evolutionary game according to claim 9, characterized in that: Calculating the probability of a user forwarding a rumor message and the probability of a rumor-refuting message at time t includes: calculating the probability θ of an ordinary infected user forwarding a rumor message at time t based on the willingness driving force of the ordinary infected user R (t) and the probability of rumor-refuting news θ Anti-R (t), the probability ρ of the rumor user forwarding the rumor message at time t is calculated based on the rumor user’s willingness driving force R (t) and the probability of rumor-refuting news π Anti-R (t), the probability ψ of the rumor-refuting user forwarding the rumor message at time t is calculated based on the rumor-refuting user’s willingness driving force R (t) and the probability of rumor-refuting news ψ Anti-R (t), according to the probability θ R (t),ρ R (t),ψ R (t) Calculate the probability of a user forwarding a rumor message at time t, according to the probability θ Anti-R (t),ρ Anti-R (t),ψ Anti-R (t) Calculate the probability that a user forwards a rumor-refuting message at time t.

Images