A topic propagation control method based on changes in user opinions

Abstract

The application belongs to the field of social networks, and particularly relates to a topic propagation control method based on user viewpoint transition, comprising: obtaining user information; constructing a topic propagation model; calculating the influence of each factor in the user information; processing the user information according to the influence of each factor using a multi-type viewpoint game driving mechanism to obtain the driving force of the user's negative, positive and negative-promoting viewpoints; processing the driving force of the negative, positive and negative-promoting viewpoints using a user viewpoint transition mechanism to obtain a user viewpoint transition dynamics equation; constructing a target function according to the user viewpoint transition dynamics equation, calculating the optimal solution of the target function; bringing the optimal solution into the user viewpoint transition dynamics equation to obtain the viewpoint state set of the user at different times; and controlling the propagation of the topic according to the viewpoint state set of the user at different times; the application realizes loss function minimization by introducing a Hamilton function to obtain an optimal control scheme.

Description

Technical Field

[0001] This invention belongs to the field of social networks, specifically relating to a method for controlling the spread of topics based on changes in user opinions. Background Technology

[0002] In ancient times, the mode of information dissemination was drastically different from today. Information dissemination relied primarily on oral traditions, writing, and handwritten artifacts. Important information and news in society were spread through word of mouth or traditional media such as letters. This method emphasized information sharing within communities, with topics spreading within limited spheres and at a relatively slow pace. This traditional method of information dissemination focused on interpersonal relationships and direct social connections, forming relatively closed information networks. However, with the arrival of the 21st century, the rapid development of mobile internet technology has completely transformed the landscape of information dissemination. The rise of social media has provided new mediums for topic dissemination. Social networking platforms have become the primary venues for people to share opinions and information, not only accelerating the speed of topic dissemination but also broadening the scope of their influence. The dissemination of topics in the digital age has begun to exhibit new characteristics such as rapid information spread and the network effects of social networks.

[0003] With the iterative development of network technology, social networks have become not only channels for information dissemination but also breeding grounds for negative and misleading topics. The widespread use of social networks allows negative and misleading topics to spread instantly, exerting a profound influence on public opinion and society. In communication studies, misleading topics refer to topics that intentionally guide or shape people's attention and discussion. The spread of negative topics can lead to misunderstandings and distrust among the public regarding specific issues, resulting in negative social impacts. Due to the sheer size of social networks, the spread of negative and misleading topics has become a major challenge. Therefore, studying the spread of misleading topics not only helps in understanding the mechanisms of information dissemination on social networks but also provides more effective means to prevent the large-scale spread of negative and misleading topics.

[0004] In recent years, scholars both domestically and internationally have studied the topic propagation process and control methods in social networks from multiple perspectives, achieving fruitful research results. Current research on propagation process models mainly focuses on two directions: one is to improve the SIR model by introducing new states to characterize the information propagation process in social networks; the other is to analyze the actual propagation status of topics in the network based on neural networks. Regarding the control problem in topic propagation, current research mainly focuses on control algorithms to control topic propagation in social networks from different aspects.

[0005] In summary, based on the current research of scholars both domestically and internationally, it is found that although many scholars have achieved some results in topic propagation models and control methods, some problems still exist:

[0006] 1. The Complexity of Quantifying Topic Influence Under Multiple Factors. In the process of guiding the dissemination of topics, topic influence is typically affected by both user factors and factors inherent to the topic itself. Therefore, quantifying the influence of guiding topics requires considering the complexity of the combined effects of multiple factors. Existing methods only address a single influencing factor, while how to quantify topic influence from multiple perspectives is a problem that needs to be considered.

[0007] 2. The Issue of User Opinion Divergence. The influence of guiding topics on user opinions is affected by multiple factors, but essentially, the divergence of user opinions is determined by the interaction of information. Previous research has focused more on explaining opinion formation through interactions between users, but the more important factor causing opinion formation and divergence is the interaction of information itself. How to establish a suitable model for the evolution of user opinions is a major current challenge.

[0008] 3. The issue of blocking the spread of negatively guiding topics. Current research has played a role in controlling such topics, but it's important to note that simply blocking negatively guiding topics is insufficient. How to more effectively prevent the spread of negatively guiding topics through other means is also a crucial question. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention proposes a topic propagation control method based on user opinion shifts. This method includes: acquiring user information, including user social relationship networks, basic user information, and historical user behavior data; constructing a topic propagation model; calculating the influence of each factor in the user information; processing the user information using a multi-type opinion game-driven mechanism based on the influence of each factor to obtain the driving forces of negative, positive, and pro-negative opinions; processing the driving forces of negative, positive, and pro-negative opinions using a user opinion shift mechanism to obtain a user opinion shift dynamic equation; constructing an objective function based on the user opinion shift dynamic equation and calculating the optimal solution of the objective function; substituting the optimal solution into the user opinion shift dynamic equation for solving to obtain the user's opinion state set at different times; and controlling the propagation of the topic based on the user's opinion state set at different times.

[0010] The beneficial effects of this invention are:

[0011] This invention proposes an influence quantification method combining a half-life function and multiple linear regression. This method, by introducing a half-life function, can more accurately reflect the decay of topic popularity over time. Simultaneously, multiple linear regression quantifies the relationship between user factors and topic influence. This method is better suited to the complexity of topic influence quantification under multiple factors. This invention establishes a user opinion change model combining evolutionary game theory. This model first introduces evolutionary game theory to establish a user opinion formation mechanism, providing a theoretical basis for user opinion formation. Then, based on cognitive difference theory, it introduces positive and negative opinion states, proposing a positive-negative-promoting-negative state opinion change model (Susceptible-Infected Positive Boost Recovered Model, SIPBR) to better characterize the evolution of user cognition about topics. This invention proposes a control strategy based on negative opinion management and positive opinion guidance. This strategy, based on optimal control theory, achieves control over the spread of negatively guiding topics. Simultaneously, a loss function is established, and a Hamiltonian function is introduced to minimize the loss function, obtaining the optimal control scheme. Attached Figure Description

[0012] Figure 1 This is an overall flowchart of the present invention;

[0013] Figure 2 This is a schematic diagram of the system input and output of the present invention;

[0014] Figure 3 This is a schematic diagram illustrating the quantification of the guiding topic influence of the present invention;

[0015] Figure 4 This is a schematic diagram illustrating the interplay between different viewpoints in this invention.

[0016] Figure 5 This is a schematic diagram of the controlled SIPBR model of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] A method for controlling topic propagation based on changes in user opinions, such as Figure 1 and Figure 2As shown, the method includes: acquiring user information, which includes the user's social relationship network, basic user information, and historical user behavior data; constructing a topic propagation model; calculating the influence of each factor in the user information; processing the user information using a multi-type opinion game-driven mechanism based on the influence of each factor to obtain the driving forces of negative, positive, and pro-negative opinions; processing the driving forces of negative, positive, and pro-negative opinions using a user opinion change mechanism to obtain a user opinion change dynamic equation; constructing an objective function based on the user opinion change dynamic equation and calculating the optimal solution of the objective function; substituting the optimal solution into the user opinion change dynamic equation for solving to obtain the user's opinion state set at different times; and controlling the propagation of the topic based on the user's opinion state set at different times.

[0019] A guided topic propagation control model oriented towards changes in user opinions includes: defining the basic concepts related to the model, namely defining the guided topic propagation network G = (G I ∪G P ∪G B It consists of a negative topic propagation network, a network that promotes the spread of negative topics, and a network that promotes the spread of positive topics. The negative topic propagation network G... I =(V I E I ), where V I E represents the set of negative users who participate in the spread of negative topics. I This represents the edge set of negative users who participate in the spread of negative topics. Similarly, other spread networks are represented as above.

[0020] Define internal factors: For users themselves, their sensitivity to information determines whether they can access various guiding topics, while their activity level on social networks and their interest in topics affect the likelihood of them forwarding them.

[0021] In social networks, most of the push notifications users receive come from their followers, and the number of followers a user has is directly proportional to the probability of encountering guiding topics. Therefore, the expression for a user's information perception ability is:

[0022]

[0023] Where, foc(u i ), foc ave (u net ) represent user u i The number of followers and the average number of followers per user on social media platforms.

[0024] Pos(u i ) is used to describe user u iUser engagement in the topic. Generally speaking, the likelihood of a user participating in a guided topic is directly proportional to their level of engagement. This article analyzes user engagement Pos(u i The definition of ) is as follows:

[0025]

[0026] pos(u i )=ζ*N[c(u i )]+N[r(u i )]

[0027] Where pos(u i ) represents user u i Positive index, pos ave (u net ) represents the average engagement level of users on the social media platform, N[c(u i )],N[r(u i )] respectively represent user u i The number of posts and retweets before the topic appears. Since users who frequently retweet are more likely to retweet leading topics, a weakening factor ζ is defined to reduce the influence of user tweets on this feature, where ζ∈[0,1].

[0028] The alignment between a topic and user interests is also a crucial factor in determining whether users participate in topic dissemination. The more interested users are in a topic, the greater the likelihood of them participating in its discussion and spread. By extracting high-frequency words from users' historical data and keywords from initiating topics, the Jaccard coefficient is used to measure the alignment between user interests and topic tags. The definition of user interest alignment is as follows:

[0029]

[0030] Among them, Label(u i () represents the user's historical high-frequency words, and Words(msg) represents the keywords for guiding topics.

[0031] In this embodiment, external factors are defined as follows: for user external factors, the popularity of guiding topics determines the user's attention and participation, while the neighbor's driving force influences the user's social behavior when participating in the spread of topics.

[0032] Any topic on social networks gradually loses its popularity over time. Therefore, we borrow the half-life function from physics to describe the change in topic popularity, and assign new meanings to its parameters. The expression for topic popularity is:

[0033]

[0034] Where t and t' represent the current time and the message publication time, respectively, and w is the regularization factor.

[0035] Preferably, the regularization factor is set to 100.

[0036] Neighbor driving force describes the information-driving power of neighboring users on a user. Neighbors play an indispensable role in guiding user participation in topic dissemination. Clearly, user participation in topic dissemination is positively correlated with the activity and sensitivity of neighboring users. Therefore, the information-driving force of neighboring users is defined as follows:

[0037]

[0038] Among them, Act(u n ),Sen(u n ) represent the activity level and sensitivity level of neighboring users, respectively, and N is the number of neighboring users.

[0039] Definition A={(a,u i ,t)} represents the user's historical behavior in participating in the spread of the topic, (a,u i ,t) represents u i The behavior of 'a' at time t. The model is formalized as follows:

[0040]

[0041] Where G and A are the inputs required to build the model, and based on the above inputs, the model output is the user set {S}. t},{I t},{P t},{B t},{R t} represents the user's state at different times.

[0042] In this embodiment, as Figure 3 As shown, the calculation of the influence of various factors in user information includes: internal factors and external factors; internal factors include user information perception, user activity, and user interest alignment; external factors include topic popularity and the information transmission power of neighboring users; defining the influence function of internal and external factors, and using a multiple linear regression model to fit and train the internal and external factors based on the influence function to obtain the partial regression coefficients of internal and external factors; and calculating the influence of each factor based on the partial regression coefficients.

[0043] Specifically, this includes defining the influence function of internal and external factors: the influence of a guiding topic describes its likelihood of changing users' opinions, and its influence can be quantified from two aspects. First, there are the user's own factors (eff). user (ui This consists of three aspects: user information perception, active participation in the topic, and level of interest in the topic. Secondly, there are external influences. out (u i It is composed of the driving force of users' neighbors and the popularity of topics on social networks. Both are defined as follows:

[0044] eff user (u i =User(u i )×Pos(u i )×Int(u i )

[0045] eff out (u i ) = Dri(u i )×Pop(t)

[0046] Among them, eff user (u i User(u) represents the influence of internal user factors. i To measure information perception by analyzing the number of followers in a user's basic information, Pos(u i To measure user engagement by combining the number of posts and reposts from their historical behavior, Int(u i This involves assessing the relevance of user interests by comparing high-frequency words from historical user behavior with keywords from the current topic; out (u i Dri(u) represents the influence of external factors on the user. i Pop(t) is used to calculate the information dissemination power of neighboring users by processing user social network relationship data, where Pop(t) represents the topic popularity.

[0047] By using a multiple linear regression model and training it, partial regression coefficients for user-specific and external factors are obtained, representing their respective weights and describing the proportion of influence each factor contributes to. Therefore, various guiding topics have a significant impact on user behavior. i The influence of an opinion can be expressed as:

[0048] Eff negative =ρ1+ρ2×eff user (u i )+ρ3×eff N out (u i )

[0049] Eff positive =ρ1+ρ2×eff user (u i )+ρ3×eff Pout (u i )

[0050] Eff boost_negative =ρ1+ρ2×eff user (u i )+ρ3×eff B out (u i )

[0051] Where ρ1, ρ2, and ρ3 are partial regression coefficients, and Eff negative Eff positive Eff boost_negative These represent the influence of negative topics, positive topics, and topics that promote negativity on user opinions, respectively.

[0052] In this embodiment, as Figure 4 As shown, the multi-type viewpoint game-driven mechanism for processing user information includes: defining three user viewpoint states and using P... N ,P P ,1-P N -P P The proportions of "negative views," "positive views," and "pro-negative views" among the user's neighbor nodes are used; a gain coefficient δ is used to measure the promoting effect of neighbors holding pro-negative views on the formation of negative views and the inhibiting effect on the formation of positive views; the payoff functions of the three game strategies are defined, and the user information is processed according to the payoff functions of the three game strategies to obtain the driving forces of negative, positive, and pro-negative views.

[0053] Specifically, this paper constructs a multi-type viewpoint game-driven mechanism. Considering that users, when simultaneously exposed to multiple guiding topics, will form different viewpoints due to their own cognitive differences, they will only forward specific topics that align with their own opinions. This paper defines three user viewpoint states based on different types of topics. Furthermore, it uses P... N ,P P ,1-P N -P P The percentages of "negative," "positive," and "pro-negative" viewpoints among the user's neighbor nodes represent this. Considering that neighbors with pro-negative views promote the formation of negative opinions, thus increasing the probability of the user forming a negative viewpoint, a gain coefficient δ is introduced to measure the promoting effect of neighbors with pro-negative views on the formation of negative opinions and the inhibiting effect on the formation of positive opinions, where δ∈[0,1]. The payoff functions for the three game strategies are defined as follows:

[0054] Pro N (u i )=(1+δ)×P N ×Eff negative

[0055] Pro P (u i )=(1-δ)×P p ×Eff positive

[0056] Pro B (u i )=(1-P N -P P )×Eff negative

[0057] In the formation of user opinions in social networks, three neighbor node states—negative, positive, and pro-negative—lead to a three-way game. This paper considers the user's final opinion as the optimal strategy after the game. Given that evolutionary game theory allows participants to choose the strategy that maximizes their benefit, this paper utilizes evolutionary game theory and combines it with various types of guiding topic influence mechanisms to formalize the driving forces behind users' formation of negative, positive, and pro-negative opinions as follows:

[0058]

[0059]

[0060]

[0061] Among them, Drf N (u i ), Drf P (u i ), Drf B (u i ) represent the driving forces that drive users to form negative, positive, and negative-promoting opinions, respectively. w1 and w2 are two adjustable parameters used to represent the different degrees of influence of negative and positive opinions on negative-promoting opinions, where w1 and w2 ∈ [0,1].

[0062] In this embodiment, as Figure 5 As shown, the process of establishing a user opinion shift mechanism includes: considering that when users are exposed to multiple guiding topics simultaneously, they form different opinions due to their own cognitive differences, and therefore will only forward specific topics that conform to their own opinions. Therefore, this paper considers improving the traditional SIR model by introducing user states of holding positive and negative opinions to establish a guiding topic propagation model SIPBR based on user opinion shifts. Here, S represents users with no opinion (easily changeable), I represents users with negative opinions, P represents users with positive opinions, B represents users with negative opinions, and R represents users with no opinion on the topic (immune to the topic).

[0063] The following rules apply to users participating in the dissemination of various types of guiding topics: 1. Assuming that the total number of users participating in the dissemination of guiding topics in an event remains constant, the sum of the proportions of each type of user is always 1; 2. Users of type S who do not hold an opinion have a certain rate of infection and form corresponding opinions when they come into contact with users of types I, P, and B; 3. The spread of guiding topics in social networks gradually disappears over time, and users who participated in them will become immune to the topics in the R state.

[0064] The following guidelines for user opinion shifts, based on user regulations governing participation in various types of guided topic dissemination, are as follows:

[0065] 1. Users with no opinion have an α probability of forming a negative opinion, a β probability of forming a positive opinion, and a γ probability of forming a pro-negative opinion. Users with a pro-negative opinion also have a λ or... Probability shapes negative or positive viewpoints Since there are always users immune to the topic throughout the entire process of guiding the spread of the topic, α+β+γ<1.

[0066] 2. Over time, user opinion states will all become immune to the topic. The probabilities of immunity for users with negative, positive, and pro-negative opinions are μ, η, and ε, respectively.

[0067] Based on the introduced new state and the established propagation rules, the following improved SIPBR dynamic equation can be derived to describe the user's viewpoint change process:

[0068]

[0069] Where S(t) represents users with no opinion, α is the probability that users with no opinion will form a negative opinion, I(t) represents users with a negative opinion, β is the probability that they will form a positive opinion, P(t) represents users with a positive opinion, γ is the probability that they will form a negative opinion, B(t) represents users with a negative opinion, λ is the probability that users with a negative opinion will form a negative opinion, and μ, η, and ε are all the immunity probabilities of users with negative opinions. To encourage users with negative views to form positive views, R(t) represents users who do not hold a view on the topic.

[0070] In this embodiment, the unknown coefficients of the above equations are solved using mean-field theory, representing the probability of user opinion state transitions. Since both user opinion transitions and topic propagation are unidirectional—meaning user opinions can only progress from neutral to negative, positive, and then to a state of promoting negativity—and ultimately eventually become immune to these states, users who initially promoted negative opinions can later shift to either negative or positive opinions. If user u... i There are m neighbor nodes, among which the probability of n users changing their viewpoint state follows a binomial distribution:

[0071]

[0072] Where X represents users whose opinions have changed. The coefficients are binomial coefficients, Drf(u i ( ) is the driving force behind the formation of viewpoints.

[0073] From the above formula, we can see that the probability of a user changing to a negative view at any given time is:

[0074]

[0075] The probability of shifting to a positive view is:

[0076]

[0077] The probability of shifting to a negative view is:

[0078]

[0079] Combining mean-field theory with the probability of viewpoint transition, we can obtain the following set of dynamic equations:

[0080]

[0081] in, The probability of shifting to a negative view. The probability of turning into a positive view. The probability of shifting to a negative viewpoint.

[0082] In this embodiment, control variables are introduced to reduce users with negative viewpoints. Specifically, this includes the fact that the spread of negatively influencing topics may lead to social instability and a series of problems. Control methods are highly effective in blocking the spread of negative topics; therefore, this paper proposes the following two control methods to block the spread of negatively influencing topics:

[0083] Intervention: Relevant social institutions take measures on social networks to guide susceptible users (S) who have not been exposed to negatively influencing topics to skip the topic's spread and directly transform into R-type users who are immune to it. In this article, u1 will represent institutional intervention.

[0084] Isolation: Implemented by the administrators of relevant online social platforms, this involves banning or directly deleting users who hold negative views and spread negative, misleading content. After a period of isolation or positive guidance, these users will gradually transform into P-type users who hold positive views. This article will use u2 to represent platform isolation.

[0085] Defining the objective function. To achieve the maximum control effect with minimum cost, this paper sets the system control objective function as a quadratic form. The reason for using a quadratic function approximation is that, firstly, there is no linear function that can accurately simulate the real situation; secondly, when the objective function is a quadratic function, the system is usually more likely to find the global minimum during the optimization process and avoid getting trapped in local minima, thereby improving the stability of the system. The objective function defined in this paper is as follows:

[0086]

[0087] The function consists of two parts: AI(t) represents the loss of social network users due to isolating users with negative views. This represents the cost of adopting the two control methods mentioned above. A, c1, and c2 are the corresponding weighting coefficients.

[0088] Solving for the optimal solution to the objective function. Based on the loss function described above, we establish an optimal control problem. The core of the problem is to find the optimal solution that minimizes the objective function, i.e.:

[0089]

[0090] Introducing the above control variables into the original SIPBR dynamic equation, the new equation is as follows:

[0091]

[0092] The constraint condition is Γ={u i |0<u i ≤1,i=1,2}, and simultaneously due to the boundedness of the right-hand side of the above dynamic equation and the objective function with respect to u i The convexity of (t)(i=1,2) yields the existence of the optimal solution.

[0093] To obtain the optimal solution, we first construct the Hamiltonian function H(t):

[0094]

[0095] Where λ S (t),λ I (t),λ P (t),λ B (t),λ R (t) are all state connection variables. Substituting the dynamic equation into the above equation, we get:

[0096]

[0097] By Pontryagin's theorem, the Hamiltonian function H(t) and its adjoint function satisfy the following equation:

[0098]

[0099] The goal of this optimization control problem is to reduce or even eliminate users with negative viewpoints in the system; therefore, at the final time, we have:

[0100] λ S (T)=λ I (T)=λ P (T)=λ B (T)=λ R (T)=0

[0101] Similarly, according to Pontryagin's theorem:

[0102]

[0103] The optimal control solution obtained is:

[0104]

[0105] Based on the optimal parameters, the final propagation dynamics equations are as follows:

[0106]

[0107] This invention considers the phenomenon that, during the propagation of guiding topics, users with varying levels of cognitive ability will develop different viewpoints when exposed to multiple types of topics, leading them to only spread the topics corresponding to their cognitive abilities. Therefore, it describes the propagation trend of guiding topics by modeling the process of user viewpoint changes. First, the influence of various topics is quantified, and evolutionary game theory is used to quantify the driving force of the interaction between various neighboring viewpoints on changes in user viewpoint states. Furthermore, to reduce the propagation of negative topics, a control method based on prevention and isolation is proposed, and an optimal control problem is established to obtain the optimal control solution. Finally, mean-field theory is used to model the process of user viewpoint changes during the propagation of guiding topics, deriving dynamic equations to describe the propagation trend of guiding topics.

[0108] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments 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 within the protection scope of the present invention.

Claims

1. A topic propagation control method based on a user opinion transition, characterized by, The method comprises the following steps: Obtaining user information, wherein the user information comprises a user social relationship network, user basic information and user historical behavior data; and constructing a topic propagation model; Calculating the influence of each factor in the user information according to the topic propagation model; Processing the user information according to the influence of each factor by using a multi-type opinion game driving mechanism to obtain driving forces of negative, positive and negative-promoting opinions of the user; processing the driving forces of the negative, positive and negative-promoting opinions by using a user opinion transition mechanism to obtain a user opinion transition dynamics equation; constructing a target function according to the user opinion transition dynamics equation, and calculating an optimal solution of the target function; Bringing the optimal solution into the user opinion transition dynamics equation to obtain a user opinion state set at different time points; Controlling the propagation of a topic according to the user opinion state set at different time points; The multi-type view game driving mechanism is used to process user information, including: defining three user view states, and using , , to represent the proportion of neighbor nodes holding "negative view", "positive view" and "promote negative view" for the user; using a gain coefficient to measure the promoting effect of the neighbor holding the promote negative view on forming the negative view and the inhibiting effect on forming the positive view; defining three game strategy payoff functions; according to the three game strategy payoff functions, the user information is processed again to obtain driving forces of the negative view, the positive view and the promote negative view; the three game strategy payoff functions are: wherein a benefit for forming a negative view, a benefit for forming a positive view, a benefit for forming a pro-negative view; The expression of the user opinion transition dynamics equation is: wherein, P is the probability of a user not holding an opinion, P is the probability of a user not holding an opinion forming a negative opinion, P is the probability of a user holding a negative opinion, P is the probability of a user forming a positive opinion, P is the probability of a user holding a positive opinion, P is the probability of a user forming a pro-negative opinion, P is the probability of a user holding a pro-negative opinion, P is the probability of a pro-negative opinion user forming a negative opinion, , P is the probability of a pro-negative opinion user forming a negative opinion, P is the probability of a pro-negative opinion user forming a negative opinion, P is the probability of a pro-negative opinion user forming a positive opinion, P is the probability of a user not holding an opinion on a topic.

2. The topic propagation control method based on user viewpoint transition according to claim 1, characterized in that, The topic propagation model comprises a guided topic propagation network G, which is composed of a negative topic propagation network , a positive topic propagation network , and an aggressive topic propagation network ; wherein the negative topic propagation network is , wherein, represents a set of negative state users participating in the negative topic propagation, represents an edge set of negative state users participating in the negative topic propagation; the structure of the positive topic propagation network and the aggressive topic propagation network is the same as that of the negative topic propagation network.

3. The topic propagation control method based on user viewpoint transition according to claim 1, characterized in that, The calculation of the influence of each factor in the user information comprises: the influence of each factor comprises internal factors and external factors; wherein the internal factors comprise user information perception, user positivity and user interest coincidence degree; the external factors comprise topic heat and information transmission force of neighbor users; an internal-external factor influence function is defined, the internal factors and the external factors are fitted and trained by using a multivariate linear regression model according to the internal-external factor influence function, and partial regression coefficients of the internal factors and the external factors are obtained; and the influence of each factor is calculated according to the partial regression coefficients.

4. The topic propagation control method based on user viewpoint transition according to claim 3, characterized in that, The internal-external factor influence function is: wherein, is the user internal factor influence, is the information perception measured by analyzing the number of attentions in the user basic information, is the activity measured by combining the number of posts and the number of retweets in the user historical behavior, is the interest coincidence degree measured by comparing the high-frequency words in the user historical behavior and the keywords of the current topic; is the user external factor influence, is the information propagation of the neighbor users calculated by processing the user social network relationship data, is the topic heat.

5. The topic propagation control method based on user viewpoint transition according to claim 4, characterized in that, The expression of the influence of each factor is: wherein, , , is a partial regression coefficient, , , respectively represent the influence of negative topics, positive topics, and pro-negative topics on the user's opinion, , , respectively represent external factors that influence the user to generate negative opinions, positive opinions, and pro-negative opinions.

6. The topic propagation control method based on user viewpoint transition according to claim 1, wherein, The driving forces of the negative, positive and negative-promoting opinions are: wherein, , , respectively represent the driving forces of the user to form negative, positive and pro-negative views, , are two adjustable parameters.

7. The topic propagation control method based on user viewpoint transition according to claim 1, wherein, The expression of the target function is: wherein, represents the social network user loss from isolating users holding negative opinions, is the cost, , , is the corresponding weight factor, is the prevention strategy cost, is the isolation strategy cost.

8. The topic propagation control method based on user viewpoint transition according to claim 1, wherein, Solving the user opinion transition dynamics equation includes: bringing the optimal solution calculated by the objective function into the user opinion transition dynamics equation and setting a constraint condition; constructing a Hamilton function ; bringing the user opinion transition dynamics equation with the set constraint condition into the Hamilton function, solving the optimized Hamilton function through the Pontryagin theorem to obtain optimal parameters, and bringing the optimal parameters into the user opinion transition dynamics equation to obtain the user opinion state set at different times.

Images