Prediction method of intentional topic propagation situation

Abstract

The invention relates to a method for predicting an intentional topic propagation situation, which comprises the following steps of: acquiring topic data on line in real time from a social network, preprocessing to obtain user basic information, user historical behaviors and topic basic information, extracting factors influencing intentional topic propagation from the information, and predicting the intentional topic propagation situation; constructing the message influence of the intentional topic according to the factor; combining the message influence with a Bayesian linear regression model to dynamically quantify the emotional states of the users to the topic to obtain the proportion of the users presenting different emotional states to the topic; the message influence is combined with an Isin model, different emotional state conversion prediction of the user on the topic is carried out, the probability of different emotional state conversion of the topic by the user is obtained, and according to the probability, a kinetic equation is constructed by combining an SVICR model; and solving the kinetic equation to obtain an emotional state set of the user to the topic at different moments in the intentional topic propagation process, and obtaining a propagation situation result of the intentional topic according to the emotional state set. According to the method, the propagation situation of the intentional topic can be accurately predicted.

Description

Technical Field

[0001] The present invention belongs to the field of Internet application technology, and in particular relates to a method for predicting the dissemination trend of intentional topics. Background Art

[0002] Topics play an important role in communication and can help people establish connections, exchange ideas, share information and build common interests. With the rapid development of the Internet, a large number of topics have been spread rapidly on the Internet, and the topics posted by users are also various, and intentional topics have also emerged. Intentional topics refer to specific topics that are intended to guide and influence users' thinking, attitudes and behaviors. On social media and online platforms, the spread of intentional topics can have a wide range of effects, including shaping public opinion, influencing social dynamics and promoting business goals. Therefore, studying how to control and guide the spread of intentional topics has important theoretical and practical value. In recent years, experts in the academic community have continued to study the spread of topics based on social psychology.

[0003] The current research on topics is mainly focused on two major directions: First, based on the classic infectious disease dynamics model, the infectious disease model is improved by dividing the user status twice to study the spread of topics on social networks. Second, by using machine learning or deep learning, information data from social networks is extracted to effectively model and predict the spread of topics.

[0004] In recent years, researchers have conducted research on different psychological states of users and their characteristics on social networks, mainly based on the traditional infectious disease model, proposing different states for improvement. The user's trust psychology is a factor worth considering. et al. (Urena R, Kou G, Dong Y, et al. A review on trustpropagation and opinion dynamics in social networks and group decision making frameworks [J]. Information Sciences, 2019, 478: 461-475) found that trust, reputation and influence are closely related, and divided trust into trust, distrust, hesitation and conflict. In social networks, users establish various explicit or implicit social relationships and use the network as a display platform to spread and share their information, opinions and suggestions. In these scenarios, the spread of information depends on the trust and influence between users. Summary of the invention

[0005] In view of the above challenges, the present invention proposes a method for predicting the intentional topic propagation trend, which includes:

[0006] S1: Topic data is obtained from social networks in real time online. After preprocessing, basic user information, user historical behavior and basic topic information are obtained.

[0007] S2: Extract factors that affect the spread of intended topics from basic user information, user historical behavior, and basic topic information, and construct the message influence of the intended topic based on the factors;

[0008] S3: combining the message influence with a Bayesian linear regression model to dynamically quantify the user's emotional state towards the topic, and obtain the proportion of users with different emotional states towards the topic;

[0009] S4: combining the message influence with the Ising model to predict the user's different emotional state transitions to the topic, and obtaining the probability of the user's different emotional state transitions to the topic;

[0010] S5: Based on the proportion of users who present different emotional states to the topic, the probability of users switching to different emotional states to the topic and the SVICR model, a dynamic equation is constructed;

[0011] S6: Solve the dynamic equation to obtain the set of users' emotional states towards the topic at different times during the process of spreading the intended topic, and obtain the spreading trend result of the intended topic based on the set of emotional states.

[0012] Beneficial effects of the present invention: The present invention focuses on the relationship between users and users, and between users and topics, and combines the user's own characteristics, the information of the topic itself, and the dissemination ability of the topic network. It can accurately simulate the dissemination trend of intentional topics in social networks, as well as the changes in the dissemination trend of intentional topics after control measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a flowchart of steps of an embodiment of the present invention;

[0014] Figure 2 A schematic diagram of the process of intentional topic propagation in the present invention;

[0015] Figure 3 Schematic diagram of the user emotional response mechanism in the present invention;

[0016] Figure 4 A schematic diagram of user emotion conversion in the present invention;

[0017] Figure 5 Schematic diagram of optimized control of intentional topic propagation in the present invention. DETAILED DESCRIPTION

[0018] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances. This is just a way of distinguishing objects with the same attributes when describing the embodiments of this application.

[0019] 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.

[0020] In the process of studying the spread of intentional topics, the study found an interesting phenomenon: in the early stages of the spread of intentional topics, each user will have his or her own initial thoughts and behaviors, but due to the influence of friends and the topic itself, as well as the psychological biases and distrust mechanisms of users, users will distrust the topic and seek verification. In the process of topic spread, users may become topic spreaders because they find that the information about the topic is true after the participation and verification of their friends, or they may become users who suppress the topic because they find that the topic carries false information. Although many Chinese and foreign scholars have conducted a series of studies on the topic spread model and the control of topic spread, the research on intentional topics still faces the following challenges:

[0021] 1. Diversity and dynamism of user status. In social networks, when users are exposed to information about intended topics, they will make different choices due to their different personal emotions towards the topic, resulting in different user status. At the same time, during the spread of the topic, the user's status will change at any time, so how to dynamically quantify the user status becomes a challenge.

[0022] 2. User trust issues. Intentional topics are purposeful, and users will have trust issues with purposeful topics, thus becoming skeptical and verifying the topics. Traditional information dissemination models are not perfect in describing such topics, so introducing appropriate states into the dissemination model becomes a challenge.

[0023] 3. Uncertainty of positive and negative feedback of intentional topics. The impact of intentional topics after dissemination is uncertain, and dynamic control strategies need to be used for control optimization. It is difficult to control the impact of topics with the optimal control cost.

[0024] The embodiment of the present invention proposes a method for predicting the intentional topic propagation situation, the method comprising:

[0025] S1: Topic data is obtained online in real time from social networks. After preprocessing, basic user information, user historical behavior and basic topic information are obtained.

[0026] Specifically, we obtain the original topic data from the public data sets of platforms such as Weibo and Twitter. The topic data is unstructured and cannot be directly used for data analysis. The original topic data can be structured through data preprocessing (such as simple data cleaning). Data preprocessing includes deleting duplicate data, cleaning invalid information, etc.

[0027] The user basic information includes the number of user fans and the number of user friends.

[0028] The user's historical behavior includes the user's historical participation in the dissemination of topics (such as liking, collecting, forwarding, blocking, etc.) and the user's own behavior in posting topics.

[0029] The basic information of the topic includes the content bias of the topic, the keywords of the topic, the source of the topic, the popularity of the topic (such as the number of likes, reposts, favorites, etc.).

[0030] In social networks, the behavior of users in the process of topic propagation is affected by many factors, such as: the user's emotional matching, the influence of related users and the popularity of the topic, etc. Based on this, the present invention extracts factors that affect user behavior from two aspects: user and topic factors.

[0031] S2: Extract factors that affect the spread of intended topics from user basic information, user historical behavior and topic basic information, and construct the message influence of the intended topic based on the factors.

[0032] Furthermore, the factors affecting the spread of the intended topic include user factors and topic factors. The user factors include user emotion matching Fit (U i ), User Positivity (U i )、User Influence(U i ) and user credibility (U i ), the topic factors include topic popularity (t) and topic origin (t).

[0033] The extracting of user factors specifically includes:

[0034] S211: Calculate user emotion matching Fit (U i )

[0035] The user's emotion matching reflects whether the user has an initial emotion on the topic. In the process of topic propagation, the user's behavior of participating in the propagation is affected by the initial emotion. The user's loyalty reflects his trust and satisfaction with the e-commerce platform. The higher the degree of matching between the user's emotion and the emotion conveyed by the current topic information, the greater the probability of the user participating in the propagation of the topic. By extracting the user's emotion label and the keywords in the topic content, the Jaccard similarity coefficient is used to measure the matching degree between the user's emotion and the topic label, and the user's emotion matching degree Fit (U i )for:

[0036]

[0037] Among them, Label(Ui) represents the high-frequency words used in the user's historical behavior, and Words(msg) represents the set of keywords in the intended topic.

[0038] S212: Calculate user activity Positivity (U i )

[0039] The user's positivity includes the number of microblogs posted by the user on the social network and the number of microblogs forwarded, which affects the probability of the user participating in the topic propagation to a certain extent. Generally, the possibility of a user participating in the intentional topic propagation is proportional to the user's positivity. The definition of user positivity (U i )for:

[0040] Positivity(U i )=α*Num[orig(U i )]+Num[forw(U i )] (2)

[0041] Among them, Positivity (U i ) represents the enthusiasm of the ith user to participate in the topic, Num[orig(U i )] represents user U i The number of microblogs published before the topic appeared, Num[forw(U i )] represents user U i The number of forwarded microblogs before the topic appears. Since users who frequently forward microblogs are more likely to forward topics, and the number of original microblogs of ordinary users is far less than the number of forwarded microblogs, a weakening factor α is defined before the original microblog to reduce the impact of the original microblog on this feature, where α∈[0,1].

[0042] S213: Calculate User Influence (U i )

[0043] User influence refers to the degree to which a user's participation in a topic influences his friends' participation in the topic. The greater the user's influence, the higher the possibility of influencing his friends to participate in the topic. User influence includes the average number of reposts and likes of the user's original microblogs. i )for:

[0044]

[0045] Among them, Influence (U i ) represents the i-th user U i The influence of participating in this topic, represents the average number of reposts of the original microblog of the i-th user, Represents the average number of likes for the original microblog of the i-th user.

[0046] S214: Calculate the user's credibility (U i )

[0047] In social networks, there are usually users such as zombie fans and water army, and these users need to be identified. The user's credibility can be judged by whether it is an official account, the level of fans, and the user's interaction behavior with friends.

[0048] Defining User Credibility (U i )for:

[0049]

[0050] Among them, Credibility(U i ) represents the i-th user U i The credibility of participating in this topic, baseCre represents the basic credibility of all users, baseCre∈(0,1), s authGra(U i ) indicates the official account level of the user, fansGra(U i ) represents the fan level of the user, α represents the attenuation factor; Num[Interact(U i )] represents the i-th user U i The total number of interactions (including likes, favorites, forwarding, blocking, etc.) with messages posted by their friends, Num[Interacted(U i )] represents the i-th user U i The total number of times the published message is interacted with by friends (including likes, favorites, forwarding, blocking, etc.), β represents the first attenuation coefficient, γ represents the second attenuation coefficient, β, γ∈(0,1), the smaller the values ​​of β and γ, the more likely the i-th user U iThe smaller the credibility, the smaller its role in the topic dissemination process; conversely, the greater its role.

[0051] The emotional attitude of high-quality topic commentators is very important for the identification of intended topics, and high-quality users often represent official accounts or celebrity accounts, but official accounts often have higher credibility than celebrity accounts. Therefore, the attenuation factor α∈[0,1] is used to weaken the impact of celebrity accounts on the overall credibility of users.

[0052] The topic factors include topic popularity (t) and topic origin (t).

[0053] S221: Calculate the topic popularity (t)

[0054] Topic popularity is defined as the traffic volume of a topic, that is, the popularity of a topic in the network. After any topic is released, it will decay over time and reach a low value after reaching its peak. This decay process is similar to the half-life property of physical elements, so the present invention introduces a half-life function to describe the decay process of topic popularity.

[0055] Define the topic popularity(t) as:

[0056]

[0057] Among them, t represents the current time of the topic propagation heat decay process, t0 represents the starting time of the topic propagation heat decay process, w represents the regularization factor, and w is set to 100.

[0058] S222: Topic Origin (t)

[0059] The source of a topic refers to the event, behavior, news, opinion, or other factor that originally proposed or triggered the topic. The source of a topic can be varied, such as media reports, social media, public discussions, or personal opinions. Among these sources, the more professional and authoritative the topic proposer is, the greater the influence on users.

[0060] Define the topic source Origin(t) as:

[0061] Origin(t) = {M, S, D, P} (6)

[0062] Among them, t represents the current moment of the topic spread heat decay process, M represents media reports, S represents social media, D represents public discussion, and P represents personal opinions.

[0063] The influence of intentional topic messages refers to the spread ability of topics in social networks and the degree of influence on users.

[0064] When a topic spreads, the factors that determine the influence of the intended topic message include topic factors and user factors. Topic factors include the popularity of the topic and the source of the topic, and user factors include the matching degree between the user's emotion and the topic, the user's enthusiasm for participating in the topic, the user's influence, and the user's credibility.

[0065] The message influence of the intentional topic is constructed based on the extracted factors that affect the spread of the intentional topic.

[0066] The message influence of the intended topic is defined as:

[0067]

[0068] Among them, fac topic represents the topic factor in the message influence, fac user represents the user factor in the message influence, Popularity(t) represents the topic popularity, Origin(t) represents the topic source, Fit(U i ) represents the user emotion matching degree, Positivity (U i ) represents user activity, Influence (U i ) represents user influence and Credibility (U i ) indicates the user’s reputation.

[0069] Figure 3 Schematic diagram of the user emotional response mechanism in the present invention.

[0070] S3: The message influence is combined with a Bayesian linear regression model to dynamically quantify the user's emotional state towards the topic, and obtain the proportion of users who present different emotional states towards the topic.

[0071] Reference Figure 3 As shown, considering that the Bayesian linear regression model can more flexibly handle the situation of incomplete data or more noise, combined with user factors and topic factors, the calculation formula for dynamically quantifying user emotions is:

[0072] ERM(U i )=ω1*fac topic +ω2*fac user +ε (8)

[0073] Among them, ERM(Ui) represents the emotional response effect of the ith user on the intended topic, ω1 represents the first weight coefficient obtained by training and fitting the Bayesian linear regression model, and fac topic Indicates topic factors, fac userrepresents the user factor, ω2 represents the first weight coefficient obtained by training and fitting the Bayesian linear regression model, and ε is the error term. In the prior setting of parameters, the prior distribution of the weight coefficients ω1 and ω2 of the present invention is a Gaussian distribution ω~N(0,σ 2 ), σ 2 is the variance of the weight coefficient.

[0074] After users are exposed to the intended topic message, they will decide whether to spread the intended topic according to formula (8). Users will present five states, namely: users who have not been exposed to the topic information are in a susceptible state (Susceptible, S); users who verify the topic due to the suspicion mechanism after being exposed to the topic are in a verification state (Verify, V); users who believe in the topic information and spread it are in an infected state (Infective, I); users who do not trust the topic after verifying it and inhibit its spread are in a restrained state (Constrain, C); users who are immune to this type of information after being exposed to too much information are in an immune state (Recovery, R).

[0075] The SIR (Susceptible-Infected-Recovered Model) propagation model is a classic infectious disease model and an abstract description of the information propagation process. In the SIR model, people are divided into three categories: susceptible (S), infected (I), and recovered (R).

[0076] In the process of intentional topic propagation, based on the SIR model and the psychological suspicion mechanism, the verification state (Verify, V) and the inhibition state (Constrain, C) generated by the questioning are introduced to construct an optimized intentional topic propagation control dynamics model SVICR, and apply it to the prediction of the propagation situation of intentional topics. In the SVICR model, users present five states: susceptible state (S), verification state (Q), infection state (I), inhibition state (C) and immune state (R).

[0077] Figure 4 Schematic diagram of user emotion conversion in the present invention.

[0078] S4: Combining the message influence with the Ising model to predict the user's different emotional state transitions to the topic, and obtain the probability of the user's different emotional state transitions to the topic.

[0079] Considering that the changes in the emotional states of adjacent users have mutual influence and the states are updated over time, the present invention introduces the Ising model in statistical physics to simulate the transition of user states and construct a user state transition mechanism.

[0080] Reference Figure 4 As shown, the process of using the user state conversion mechanism to convert the user state specifically includes:

[0081] S401: Define state space: define the user's emotional state on the topic as a state space, which includes susceptible state (S), verification state (Q), infection state (I), inhibition state (C) and immune state (R).

[0082] According to the idea of ​​the Ising model, an energy value is assigned to each state, which reflects the stability of the user in that state.

[0083] S402: Define the energy function as:

[0084] E(state)=w1*ERM S +w2*ERM V +w3*ERM I +w4*ERM C +w5*ERM R (9)

[0085] Among them, E(state) represents the emotional state energy, ERM S The energy value indicating the user's susceptibility to topic information, ERM V Indicates the energy value of the user's verification status of the topic information, ERM I Indicates the energy value of the user's infectious state to the topic information, ERM C The energy value indicating the user's suppression of topic information, ERM R It represents the energy value of the user's immune state to the topic information, w1 represents the weight corresponding to the susceptible state energy, w2 represents the weight corresponding to the verification state energy, w3 represents the weight corresponding to the infection state energy, w4 represents the weight corresponding to the inhibition state energy, and w5 represents the weight corresponding to the immune state energy.

[0086] According to the characteristics of the Ising model, the user's transition between different states is driven by the probability of energy difference.

[0087] S403: Based on the energy function, define the transition probability of the user from the current state to other states, which can be in the form of Boltzmann distribution:

[0088]

[0089] Among them, P represents the probability of the user's mentality towards topic information shifting from the current state to the new state, state_current represents the user's current state towards the topic information, state_new represents the user's new state towards the topic information, E(state_new) represents the energy corresponding to the user's new state towards the topic information, E(state_current) represents the energy corresponding to the user's current state towards the topic information, and T represents the temperature parameter, which is used to control the randomness of the transition; to make the model more deterministic, so the initial value of the temperature parameter T set in the present invention is 1.0.

[0090] S404: Perform state transfer: Based on the defined transfer probability, use a random sampling method to perform state transfer, and obtain the probability of the user's transition to different emotional states on the topic.

[0091] At each time step, a new state is randomly selected as the next state of the user based on the current state and the transition probability. In the Ising model, each user can be regarded as a spin and can be in different states. The interaction between users can be represented by the energy between adjacent users. The definition of this energy can be based on the social network relationship between users or other factors. The Ising model takes into account the mutual influence between users. When a user state changes, it may be affected by the state of adjacent users.

[0092] The annealing cooling algorithm is introduced to improve the propagation probability and recovery probability during topic propagation, the optimal control problem is established, and an optimized intentional topic propagation control dynamics model SVICR is proposed.

[0093] The construction of the SVICR model is based on the following assumptions:

[0094] 1. Since the spread of topics is explosive and time-limited, the present invention assumes that during the spread of topics, the number of user nodes in the social network will always remain constant at any time, so the sum of the state ratios in the model at any time is S+V+I+C+R=1.

[0095] 2. In social networks, the spread of topic information is a special kind of infectious disease transmission, so users in different states will have a certain probability of infection after contact. When susceptible users come into contact with infected users or verification users, there is a risk of infection.

[0096] 3. During the dissemination process, the popularity of information will continue to weaken with the change of time, and users will gradually forget the topic information and become immune users. Therefore, infected users and verification users have a certain recovery rate to become immune users.

[0097] Based on the above assumptions, the propagation rules of intentional topics in social networks of the present invention are as follows:

[0098] 1. During the spread of a topic, a susceptible node will become an infected node with a probability of α, a verification node with a probability of β, or an inhibitory node with a probability of γ.

[0099] 2. As time goes by, the verification node will become an infected node with a probability of δ, an immune node with a probability of ε, or an inhibited node with a probability of Ф; at the same time, due to the influence of various factors, the infected node will become an immune node with a probability of η.

[0100] 3. Under certain circumstances, an immune node will also become an inhibitory node with a probability of θ.

[0101] S5: Based on the proportion of users showing different emotional states on the topic, the probability of users switching to different emotional states on the topic, and the SVICR model, a dynamic equation is constructed. The dynamic equation satisfies the propagation rules of intentional topics in social networks.

[0102] The kinetic equation is specifically:

[0103]

[0104] Among them, I(t) represents the proportion of users in the infected state I at any time t during the topic propagation process, S(t) represents the proportion of users in the susceptible state S at any time t during the topic propagation process, V(t) represents the proportion of users in the verification state V at any time t during the topic propagation process, C(t) represents the proportion of users in the inhibited state C at any time t during the topic propagation process, R(t) represents the proportion of users in the immune state R at any time t during the topic propagation process, α represents the probability of users transforming from the susceptible state S to the infected state I during the topic propagation process, and β represents the probability of users transforming from the susceptible state S to the infected state I during the topic propagation process. The probability that a user transforms from a susceptible state S to a verification state V, γ represents the probability that a user transforms from a susceptible state S to a suppressed state C during the topic propagation process, δ represents the probability that a user transforms from a verification state V to an infected state I during the topic propagation process, η represents the probability that a user transforms from an infected state I to an immune state R during the topic propagation process, φ represents the probability that a user transforms from a verification state V to an suppressed state C during the topic propagation process, ε represents the probability that a user transforms from a verification state V to an immune state R during the topic propagation process, and θ represents the probability that a user transforms from an immune state R to an suppressed state C during the topic propagation process.

[0105] Figure 5 Schematic diagram of optimized control of intentional topic propagation in the present invention.

[0106] S6: Solve the dynamic equation to obtain the set of users' emotional states towards the topic at different times during the process of spreading the intended topic, and obtain the spreading trend result of the intended topic based on the set of emotional states.

[0107] Since the spread of intentional false information may have a negative impact on social networks and the real world, the propagation probability P' and the recovery probability R' are important indicators to measure the impact of rumor propagation. A higher propagation probability P' means that the information is more likely to be widely spread; a lower recovery probability R' means that the information has a more lasting impact on the audience. Therefore, certain measures need to be taken to control the spread of intentional topics. The present invention adjusts the propagation probability and the recovery probability based on the annealing strategy, such as Figure 5 shown. Figure 5 In the above, Natural propagation means natural propagation, Propagation with control means controlled propagation, Optimization means optimization, Propagation probability means propagation probability, Recovery probability means recovery probability. S means that the user is susceptible to the topic, Q means that the user is seeking verification, I means that the user is infected, C means that the user is inhibited, and R means that the user is immune. The annealing strategy can allow a certain degree of state change while achieving the global optimal solution to avoid falling into the local optimal solution. After the annealing strategy is implemented, users in the infected state will become immune users or inhibited users to a certain extent, which is represented by the variable V1(t) in the present invention; and users in the susceptible state will become immune users to a certain extent, which is represented by the variable V2(t) in the present invention.

[0108] In order to obtain the best control effect based on the control cost and price, the present invention defines the system loss function when implementing the annealing strategy, and then establishes the optimal control problem. When the control strategy is implemented, it will inevitably lead to the loss of users participating in the topic. Therefore, the loss of the present invention is divided into two parts: the user part and the cost part of implementing the control strategy. The definition of the system loss function is as follows:

[0109]

[0110] Among them, L represents the system loss in the state transition process, V1(t) represents the cost of converting users in the infected state at different times t to immune users or suppressed users, V2(t) represents the cost of converting users in the susceptible state at different times t to immune users, c1 represents the first weight of the cost part and the user churn part, c2 represents the second weight of the cost part and the user churn part, c3 represents the third weight of the cost part and the user churn part, and I(t) represents the proportion of users in the infected state I at any time t during the topic propagation process.

[0111] Furthermore, based on the annealing strategy, the control variables are introduced into the kinetic equation to obtain the following:

[0112]

[0113] Where V1 = m + n. Therefore, there exists an optimal control V1 * (t),V2 * (t) and the corresponding solution S * , I * , V * , C * , R * , so that L takes the minimum value, that is, the optimal solution is L(V1,V2)={L(V1 * (t),V2 * (t))}, and at this time V1 and V2 reach their minimum values.

[0114] In order to solve the optimal control problem, we first construct the Lagrangian function L of the problem and derive the optimal solution of the problem:

[0115] L=c1V1 2 (t)+c2V2 2 (t)+c3I(t) (14)

[0116] The Hamiltonian function H for this problem is defined as:

[0117]

[0118] In the formula, λ S (t) represents the accompanying variable of the susceptible state S, λ I (t) represents the accompanying variable of infection state I, λ V (t) represents the accompanying variable of the state V to be verified, λ C (t) represents the accompanying variable of the suppression state C, λ R (t) represents the accompanying variable of immune status R.

[0119] The final state of the system is not fixed, so it is subject to the following constraints:

[0120] S(0)≥0,I(0)≥0,V(0)≥0,C(0)≥0,R(0)≥0

[0121] S(t n ),I(t n ),V(t n ),C(t n ),R(t n ) Final state freedom

[0122] Taking the partial derivative of the Hamiltonian function with respect to the state variable, the adjoint function can be obtained as follows:

[0123]

[0124] The ultimate goal of this optimization control problem is to reduce or eliminate the spread of intentional topics, so the final result is:

[0125] λ S (t n )=λ I (t n )=λ V (t n )=λ C (t n )=λ R (t n )=0 (21)

[0126] In the formula, λ S (t n ) indicates t n The value of the accompanying variable of the susceptible state S at the moment, λ I (t n ) indicates t n The accompanying variable value of the infection state I at the moment, λ V (t n ) indicates t n Always verify the value of the accompanying variable of state V, λ C (t n ) indicates t n The accompanying variable value of the state C is suppressed at all times, λ R (t n ) indicates t n The value of the accompanying variable of the immune state R at that moment.

[0127] According to the Pontryagin extremum principle, the expression of the solution to the optimal control problem is solved by We can get the solution:

[0128]

[0129] Combined with the control conditions, the simplification can be obtained:

[0130]

[0131] Where V1 * (t) represents the optimal value of the cost lost by users in the infected state becoming immune users or inhibited users to a certain extent after the annealing strategy is implemented, V2 * (t) represents the optimal value of the cost lost by a user in a susceptible state to become an immune user to a certain extent after the annealing strategy is implemented, V 1max V represents the maximum cost of an infected user becoming an immune user or a suppressed user to a certain extent. 2max It indicates the maximum cost lost by a user in a susceptible state to become an immune user to a certain extent.

[0132] Since the propagation of intentional false information is unidirectional, the state transition of each user is also unidirectional. That is, the user state can only change from the susceptible state to the verification state, the infected state, or the suppressed state, and finally to the immune state. Assume that user U i A node has n neighbor nodes, and the probability P(X=m) that m neighbor nodes forward the topic at time t follows a binomial distribution:

[0133]

[0134] Then any user U i The probability of forwarding an intention topic at time t is ψ I (t) is:

[0135]

[0136] Similarly, any user U i The probability of verifying the intention topic and suppressing the spread of the intention topic at time t is:

[0137]

[0138] In the formula, ψ V (t) represents any user U i The probability of verifying the intention topic at time t, ψ C (t) represents any user U i The probability of suppressing the spread of intentional topics at time t.

[0139] Combining the mean field theory, by substituting formulas (26)-(27) and (29)-(31) into formula (15), we can obtain the optimized dynamic equations, which are as follows:

[0140]

[0141] Among them, V1 * (t) represents the optimal value of the cost lost by users in the infected state becoming immune users or inhibited users to a certain extent after the annealing strategy is implemented, V2 * (t) represents the optimal value of the cost lost by a user in a susceptible state to become an immune user to a certain extent after the annealing strategy is implemented. represents the probability of the user becoming infected at time t, represents the probability that the user switches to the verification state at time t, represents the probability of the user converting to the suppressed state at time t, m represents the cost of the infected user becoming an immune user, n represents the cost of the infected user becoming a suppressed user, and m+n=V1. Solving the above dynamic equation (32), we can obtain the state set {St}, {Vt}, {It}, {Ct}, {Rt} of the user at different times, that is, the user's communication status results for the intentional topic. According to the output results of the intentional topic propagation dynamics model in the present invention, the communication status of the intentional topic and the communication effect after adding control optimization in the topic propagation process can be obtained.

[0142] Based on the SIR model and annealing strategy, the present invention constructs the intentional topic propagation control model SVICR, and then constructs the optimized dynamic equation and solves the dynamic equation, so as to describe and predict the propagation trend of the intentional topic. It can help governments, organizations and individuals achieve goals such as public opinion guidance, brand promotion, and social dynamic management, and has important social and economic value; at the same time, it helps governments and regulatory agencies to better manage and guide the spread of intentional topics and maintain social order and public interests.

[0143] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium, which can include: ROM, RAM, disk or CD, etc.

[0144] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the intentional topic propagation trend, characterized in that: include: Topic data is obtained from social networks in real time online, and after preprocessing, basic user information, user historical behavior and basic topic information are obtained; Extract factors that affect the spread of intended topics from basic user information, user historical behavior, and basic topic information, and construct the message influence of intended topics based on the factors; The message influence is combined with a Bayesian linear regression model to dynamically quantify the user's emotional state towards the topic, and obtain the proportion of users with different emotional states towards the topic; The message influence is combined with the Ising model to predict the user's emotional state transitions to the topic, and obtain the probability of the user's emotional state transitions to the topic; Based on the proportion of users who present different emotional states to the topic, the probability of users switching to different emotional states to the topic and the SVICR model, a dynamic equation is constructed; The dynamic equation is solved to obtain the set of users' emotional states towards the topic at different times during the process of the intended topic propagation, and the propagation trend result of the intended topic is obtained based on the set of emotional states.

2. The method for predicting the intentional topic propagation trend according to claim 1, characterized in that: The factors that affect the spread of intentional topics include user factors and topic factors. The user factors include user emotion matching degree Fit (U i ), User Positivity (U i )、User Influence(U i ) and user credibility (U i ), the topic factors include topic popularity (t) and topic origin (t).

3. The method for predicting the intentional topic propagation trend according to claim 2, characterized in that: The message influence of the intended topic is: fac topic =Popularity(t)*Origin(t) fac user =Fit(U i )*Positivity(U i )*Influence(U i )*Credibility(U i ) Among them, fac topic represents the topic factor in the message influence, fac user Represents the user factor in the message influence.

4. The method for predicting the intentional topic propagation trend according to claim 1, characterized in that: The dynamic quantification of the user's emotional state on the topic is calculated as follows: ERM(U i )=ω1*fac topic +ω2*fac user +e Among them, ERM(Ui) represents the emotional response effect of the ith user on the intended topic, ω1 represents the first weight coefficient obtained by training and fitting the Bayesian linear regression model, and fac topic Indicates topic factors, fac user represents the user factor, ω2 represents the first weight coefficient obtained by training and fitting the Bayesian linear regression model, and ε is the error term.

5. The method for predicting the intentional topic propagation trend according to claim 1, characterized in that: The message influence is combined with the Ising model to predict the user's emotional state transitions on the topic. The prediction process specifically includes: The user's emotional state on the topic is defined as a state space, which includes susceptible state (S), verification state (Q), infected state (I), suppressed state (C) and immune state (R); The energy function is defined as: E(state)=w1*ERM S +w2*ERM V +w3*ERM I +w4*ERM C +w5*ERM R Among them, E(state) represents the emotional state energy, ERM S The energy value indicating the user's susceptibility to topic information, ERM V Indicates the energy value of the user's verification status of the topic information, ERM I Indicates the energy value of the user's infectious state to the topic information, ERM C The energy value indicating the user's suppression of topic information, ERM R Indicates the energy value of the user's immune state to the topic information, w1 represents the weight corresponding to the susceptible state energy, w2 represents the weight corresponding to the verification state energy, w3 represents the weight corresponding to the infection state energy, w4 represents the weight corresponding to the inhibition state energy, and w5 represents the weight corresponding to the immune state energy; Based on the energy function, the transition probability of the user from the current state to other states is defined as: Among them, P represents the transition probability of the user's attitude towards topic information from the current state to the new state, state_current represents the user's current state towards topic information, state_new represents the user's new state towards topic information, E(state_new) represents the energy corresponding to the user's new state towards topic information, E(state_current) represents the energy corresponding to the user's current state towards topic information, and T represents the temperature parameter, which is used to control the randomness of the transition; Based on the defined transition probability, a random sampling method is used to perform state transition and obtain the probability of users transitioning to different emotional states on the topic.

6. The method for predicting the intentional topic propagation trend according to claim 1, characterized in that: The kinetic equation is specifically: Among them, I(t) represents the proportion of users in the infected state I at any time t during the topic propagation process, S(t) represents the proportion of users in the susceptible state S at any time t during the topic propagation process, V(t) represents the proportion of users in the verification state V at any time t during the topic propagation process, C(t) represents the proportion of users in the inhibited state C at any time t during the topic propagation process, R(t) represents the proportion of users in the immune state R at any time t during the topic propagation process, α represents the probability of users transforming from the susceptible state S to the infected state I during the topic propagation process, and β represents the probability of users transforming from the susceptible state S to the infected state I during the topic propagation process. The probability that a user transforms from a susceptible state S to a verification state V, γ represents the probability that a user transforms from a susceptible state S to a suppressed state C during the topic propagation process, δ represents the probability that a user transforms from a verification state V to an infected state I during the topic propagation process, η represents the probability that a user transforms from an infected state I to an immune state R during the topic propagation process, φ represents the probability that a user transforms from a verification state V to an suppressed state C during the topic propagation process, ε represents the probability that a user transforms from a verification state V to an immune state R during the topic propagation process, and θ represents the probability that a user transforms from an immune state R to an suppressed state C during the topic propagation process.

7. The method for predicting the intentional topic propagation situation according to claim 6, characterized in that: Based on the annealing strategy, the optimized kinetic equations can be obtained by introducing control variables into the kinetic equations, which are specifically: Among them, V1 * (t) represents the optimal value of the cost lost by users in the infected state becoming immune users or inhibited users to a certain extent after the annealing strategy is implemented, V2 * (t) represents the optimal value of the cost lost by a user in a susceptible state to become an immune user to a certain extent after the annealing strategy is implemented. represents the probability of the user becoming infected at time t, represents the probability that the user switches to the verification state at time t, represents the probability of the user converting to the suppressed state at time t, m represents the cost of the infected user becoming an immune user, and n represents the cost of the infected user becoming a suppressed user.

Images