A method for predicting propagation of derivative topics based on multi-topic sentiment measurement

Abstract

The application belongs to the field of network topic propagation analysis, and particularly relates to a derivative topic propagation prediction method based on multi-topic sentiment measurement; the method comprises the following steps: obtaining topic data; performing message factor feature extraction and user factor feature extraction on the topic data; calculating message influence according to topic heat, user driving force, user activity and topic personal sentiment matching degree; dividing user networks to obtain user classification results; calculating user independent forwarding probability according to the message influence; calculating conversion probability between different types of users according to topic correlation and the user independent forwarding probability; constructing a dynamic equation according to the user independent forwarding probability and the conversion probability between different types of users; solving the dynamic equation to obtain a propagation prediction result of the derivative topic by users; and the application can effectively predict the propagation trend of derivative topics in social networks and has high practicability.

Description

Technical Field

[0001] This invention belongs to the field of online topic propagation analysis, specifically involving a method for predicting the propagation of derivative topics based on multi-topic sentiment measurement. Background Technology

[0002] In recent years, with the widespread adoption of the internet and its ever-expanding scale, people can conveniently and quickly access information from various social media platforms, and netizens are increasingly inclined to express their thoughts on social networks. Furthermore, with the development and application of technologies and software such as smartphones, Weibo, and WeChat, a vast amount of information on various topics floods the internet, and the channels, speed, and scope of information dissemination are unprecedented. Whenever a breaking event occurs, netizens always engage in intense and diverse discussions. This discussion and forwarding promotes the fermentation of public opinion regarding the breaking event online, and also generates one or more derivative topics corresponding to the current event, expanding the scope and accelerating the speed of public opinion dissemination.

[0003] In real life, online public opinion has an increasing impact on public perception, and social media crises can erupt suddenly at any time. Therefore, predicting public opinion topics to a certain extent is crucial for overall control and real-time monitoring of the spread of public opinion and the derivative trends of public opinion topics, thereby guiding the public to make relatively accurate and objective judgments about emergencies. During the dissemination of topics, users' subjective perceptions and psychology greatly influence their forwarding behavior. As a difficult-to-quantify and uncertain factor, predicting user psychology becomes a major challenge. Therefore, accurately acquiring online public opinion information and studying topics and their derivative trends are of great significance for guiding and monitoring online public opinion.

[0004] In recent years, scholars have conducted extensive research on derivative topic propagation models, primarily based on the SIR infectious disease model, machine learning algorithms, and deep learning algorithms. The SIR infectious disease model-based topic propagation prediction model mainly categorizes users into three states: susceptible (S), infected (I), and immune (R). Susceptible users are those who haven't yet encountered the topic but may spread it; infected users are those who have already encountered the topic and will actively participate in its spread; and immune users are those who have encountered the topic but will not participate in its spread. Machine learning algorithm-based prediction models mainly extract features influencing user propagation, transforming the prediction problem into a classification or regression problem. Machine learning algorithms can process massive amounts of data and are suitable for handling complex problems in social networks. Since the process of topic propagation is similar to the spread of infectious diseases, the SIR infectious disease model can be used to predict the propagation trend of derivative topics.

[0005] An analysis of current research on topic propagation prediction reveals significant progress in the study of derivative topic propagation, but several challenges remain: 1. The complexity of prior emotions. Prior emotions are an indispensable factor in the propagation of derivative topics, but users' individual emotional preferences are diverse and uncertain, making the measurement of these preferences challenging. 2. The multi-layered nature of derivative topics. The trends and characteristics of message propagation differ across different levels of topics, yet connections and transitions between levels are necessary. Combining the differences between multiple levels of topics with the state transitions between levels presents a challenge. 3. The complexity of sustained attention psychology. In the process of multi-topic propagation, besides the factors inherent in the topics themselves, sustained attention psychology influences users' continued engagement with new topics derived from a given topic. Measuring the impact of this psychological factor on topic propagation remains a challenge. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention proposes a method for predicting the spread of derivative topics based on multi-topic sentiment measurement. This method includes:

[0007] S1: Obtain topic data, which includes user history behavior information and topic participation information;

[0008] S2: Extract message factor features and user factor features from topic data; message factors include topic popularity and topic relevance, while user factors include user engagement, user activity, and topic personal sentiment matching degree.

[0009] S3: Calculate message influence based on topic popularity, user engagement, user activity, and topic-based personal sentiment matching.

[0010] S4: Divide the user network to obtain the user classification results;

[0011] S5: Calculate the user's independent forwarding probability based on the message's impact;

[0012] S6: Calculate the conversion probability between different user groups based on topic relevance and the user's independent forwarding probability;

[0013] S7: Construct a dynamic equation based on the user's independent forwarding probability and the conversion probability between different user classes;

[0014] S8: Solve the dynamic equations to obtain the prediction results of user propagation of derivative topics.

[0015] Preferably, the process of calculating message influence includes: calculating message factor influence based on topic popularity and user engagement; calculating user factor influence based on user activity and topic personal sentiment matching degree; and calculating message influence based on message factor influence and user factor influence.

[0016] Furthermore, the formula for calculating the impact of a message is:

[0017] Eff(u i )=β0+β1*fac user (u i )+β2*fac topic (u i )

[0018] Among them, Eff(u i ) represents user u i The influence of the news, β0 represents the first partial regression coefficient, β1 represents the second partial regression coefficient, fac user (u i ) represents user u i User factors influence, β2 represents the third partial regression coefficient, fac topic (u i ) represents user u i The influence of news factors.

[0019] Preferably, the process of segmenting the user network includes: dividing users into those who have participated in the preceding topic and those who have not; further dividing those who have participated in the preceding topic and those who have not into susceptible users, infected users, and immune users, thus obtaining the user classification results.

[0020] Preferably, the formula for calculating the independent forwarding probability of a user is:

[0021]

[0022]

[0023] Where λ represents the user's independent forwarding probability, ψ(t) represents the intermediate parameter, and m represents the user u i The number of neighbors, where n represents the number of neighbors who forwarded the topic, Eff(u i ) represents user u i The influence of the news.

[0024] Preferably, the process of calculating the conversion probability between different user groups based on topic relevance and the user's independent forwarding probability includes:

[0025] The cross-model state transition probability is calculated based on the topic relevance and the user's independent forwarding probability. Immune users are transformed into susceptible users who participate in derivative topics based on the cross-model state transition probability.

[0026] The conversion probability of susceptible users who have participated in the preceding topic is calculated based on the topic relevance and the user's independent forwarding probability. Susceptible users are then converted into infected users who participate in the derivative topic based on this conversion probability.

[0027] The conversion probability of susceptible users who have not participated in the preceding topic is calculated based on the topic relevance and the user's independent forwarding probability. Susceptible users are then converted into infected users who participate in the derivative topic based on this conversion probability.

[0028] Infected users become immune users with a fixed probability.

[0029] Furthermore, the formula for calculating the cross-model state transition probability is as follows:

[0030]

[0031] Where, p ab Rlv(r) represents the probability of a user who is immune to topic a transition to a user who is susceptible to topic b, λ represents the probability of a user independently forwarding the message, and Rlv(r) represents the probability of a user independently forwarding the message. a ,r b ) represents the topic relevance between topic a and topic b, and β represents the topic relevance threshold.

[0032] Furthermore, the formula for calculating the switching probability of susceptible users who have participated in precursor topics is as follows:

[0033] λ p =λ+(1-λ)×Rlv(r) a ,r b )

[0034] Where, λ p Rlv(r) represents the switching probability of susceptible users who have participated in the preceding topic, λ represents the independent forwarding probability of a user, and Rlv(r) represents the switching probability of a susceptible user who has participated in the preceding topic. a ,r b The ) represents the degree of topic relevance between topic a and topic b.

[0035] Furthermore, the formula for calculating the switching probability of susceptible users who have not participated in the preceding topics is as follows:

[0036] λ q =(1-Rlv(r) a ,r b ))×λ

[0037] Where, λ q Rlv(r) represents the probability of a susceptible user switching from a topic that has not been involved in the preceding discussion, λ represents the user's independent forwarding probability, and Rlv(r) represents the probability of switching from a topic that has not been involved in the preceding discussion. a ,r b The ) represents the degree of topic relevance between topic a and topic b.

[0038] Preferably, the kinetic equation is expressed as:

[0039]

[0040] in, Rlv(r) represents the independent forwarding probability of a user; α represents the correlation between topic i and its predecessor topics. i-1 ,r i G0 represents a user network that has not participated in the dissemination of the precursor topic, and S... i (t) and I i (t) represents the proportions of susceptible and infected users participating in topic i at time t, respectively. i-1 (t) represents the proportion of immune users who participated in the propagation of the precursor topic i at time t, G1 represents the user network that participated in the propagation of the precursor topic, and p (i-1)i S represents the cross-model state transition probability between topic i and its predecessor topics, β represents the topic relevance threshold, and S i I i R i These represent the proportions of susceptible, infected, and immune users at any given time.

[0041] The beneficial effects of this invention are as follows: This invention constructs a multi-layer iterative SIR model to simulate the mutual influence process between the topic and derivative topics during the dissemination process, introduces the psychological factor of users' continuous attention to the topic, quantifies the influence of users' psychological state on the dissemination of the topic under different derivative topics, and constructs a dynamic equation for the dissemination of derivative topics by combining topic and user characteristics. Based on the equation results, the prediction results of users' dissemination of derivative topics are obtained. This invention can effectively predict the dissemination trend of derivative topics in social networks and has high practicality. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the structure of the derivative topic propagation prediction method based on multi-topic sentiment measurement in this invention;

[0043] Figure 2 This is a schematic diagram illustrating the process of quantifying the influence of a message in this invention;

[0044] Figure 3 This is a schematic diagram illustrating the user-to-user conversion process in this invention. Detailed Implementation

[0045] 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.

[0046] This invention proposes a method for predicting the spread of derivative topics based on multi-topic sentiment metrics, such as... Figure 1 As shown, the method includes the following:

[0047] S1: Obtain topic data, which includes user history behavior information and topic participation information.

[0048] Topic data can be obtained from publicly available data websites or using mature social network public APIs. What's needed here is the user history behavior information and topic participation information of all participants (users) throughout the lifecycle of the topic and its derivative topics. Topic participation information includes the time when the topic was forwarded and commented on, the basic information of participating users, and their friend relationship information (including following and being followed). User history behavior information includes the information on forwarded and commented content by the user in the past.

[0049] S2: Extract message factor features and user factor features from topic data; message factors include topic popularity and topic relevance, while user factors include user engagement, user activity, and topic personal sentiment matching degree.

[0050] Topic popularity P(t):

[0051] The topic popularity P(t) represents the volume of traffic to the topic, or its degree of popularity. Each relatively independent topic experiences a gradual increase in popularity followed by a gradual decline. The traffic of a topic slowly decreases over time, eventually reaching a low final value, indicating that the topic has faded from public view. This decline process is similar to the half-life of physical elements. This invention introduces a half-life function. To describe the decline in topic traffic, the topic popularity P(t) is defined as follows:

[0052]

[0053] Where t and t0 represent the current moment and the start moment of the topic's decline in popularity, respectively, and T is the average propagation period of the topic. Preferably, T can be set to 1000.

[0054] Topic relevance Rlv(r1,r2):

[0055] Topic relevance is defined as the degree of connection or similarity between the content of two topics. Keywords or tags are extracted from the topic content and normalized using the Jaccard similarity coefficient. A higher Jaccard coefficient indicates a greater relevance between the two topics, and vice versa. This invention defines topic relevance as follows:

[0056]

[0057] Here, Topic(r1) and Topic(r2) represent the keywords or tags of topic r1 and topic r2, respectively.

[0058] User-driven power I(u) i ):

[0059] After a user posts a topic, their followers view, comment on, and share it, causing the topic to gradually spread on social networks; They represent user u respectively i The average number of views, comments, and reposts of all Weibo posts published in the period preceding the topic's inception. Since views are usually significantly greater than the other two metrics and have little impact on topic spread, a weakening coefficient μ is added before the average views. The driving force of users in the information dissemination process can be defined as:

[0060]

[0061] User activity Act(u i ):

[0062] User activity represents the activity level of a user. i The activity level of users in the recent spread of the topic. We generally believe that users with higher activity levels are more likely to participate in the spread of the topic and have a greater impact on its dissemination. This invention defines user activity as follows:

[0063] Act(u i )=η×Num[orig(u i )]+Num[retw(u i )]

[0064] Among them, Num[orig(u i )] and Num[retw(u i )] respectively represent user u i The number of original and retweeted Weibo posts published in the period leading up to the spread of the topic. Generally, users create far fewer original Weibo posts than retweet them, so η∈[0,1] is defined as a weakening coefficient to reduce the impact of the number of original Weibo posts on the calculation of user activity.

[0065] Topic Personal Sentiment Match (r) i ,u i ):

[0066] Similar to the definition of relevance between topics, we define an individual's emotional affinity for a topic to describe their level of interest in it. The higher the affinity, the more likely a user is to share the topic. The definition of an individual's emotional affinity for a topic is as follows:

[0067]

[0068] Among them, Topic(r i ) indicates topic r i Keywords or tags, Inter(u i ) represents user u i Interest preference tags.

[0069] S3: Calculate message influence based on topic popularity, user engagement, user activity, and topic-personal sentiment matching.

[0070] like Figure 2 As shown, in social networks, the factors determining a message's influence include both the message itself and user factors. Message-related factors include topic popularity and relevance, while user factors include activity level, emotional connection, and the psychology of attention. Combining these two factors allows for the measurement of a message's influence; specifically:

[0071] The impact of a message is calculated based on its popularity and user engagement:

[0072] fac topic (u i )=I(u i )*P(t)

[0073] The influence of user factors is calculated based on user activity and the degree of personal sentiment relevance to the topic:

[0074] fac user (u i ) = Act(u i Match(r) i ,u i )

[0075] Message influence is calculated based on the influence of message factors and the influence of user factors:

[0076] Eff(u i )=β0+β1*fac user (u i )+β2*fac topic (u i )

[0077] Among them, Eff(u i ) represents user u iThe message's influence; β0, β1, and β2 represent the first, second, and third partial regression coefficients obtained by training and fitting a multiple linear regression model, respectively. β1 and β2 are used to represent the weights of user factors and message factors, describing the proportion of each factor in the overall influence; fac user (u i ) represents user u i User factors influence, fac topic (u i ) represents user u i The influence of news factors.

[0078] S4: Divide the user network to obtain the user classification results.

[0079] This invention takes a preceding topic as its research background and divides the entire topic network into two propagation networks: one that has participated in the propagation of the preceding topic and the other that has not. Let G0 = {V0, E0} and G1 = {V1, E1} represent the user networks that have not participated in the propagation of the preceding topic and those that have participated, respectively (where V0 represents the set of users who have not participated in the propagation of the preceding topic, and E0 represents the set of relationships between users who have not participated in the propagation of the preceding topic; V1 is similar to E1). Then G = (G0 ∪ G1) represents the social network under the background of the preceding topic.

[0080] Users who participated in the precursor topic and those who did not participate in the precursor topic were further divided into susceptible users (S), infected users (I), and immune users (R), resulting in user classification results.

[0081] S5: Calculate the probability of a user's independent forwarding based on the message's influence.

[0082] Considering the repetitive nature of the lifecycle during topic derivation, this invention couples the SIR model into multiple layers based on different topic layers, constructing an Iterative-SIR (multi-topic iterative SIR) derivation topic propagation model. Due to the unidirectional nature of information propagation at each derivation topic layer, the user's state transition within the same layer is also unidirectional. That is, the user's state can only progress from a susceptible state to an infected state and then to an immune state. Assume a user node u... i There are m neighbor nodes, and the probability that n of them forward a topic at this level follows a binomial distribution:

[0083]

[0084] Able to obtain any user u i The probability of forwarding this topic at time t is:

[0085]

[0086] Using mean-field theory, we obtain the infection rate at time t, which is the user's independent forwarding probability:

[0087] S6: Calculate the conversion probability between different types of users based on topic relevance and the user's independent forwarding probability.

[0088] Based on actual online dissemination patterns of topics, this invention defines the following rules for topic propagation in social networks:

[0089] Because topic dissemination is characterized by its short duration and explosive nature, this invention does not consider population dynamics factors such as birth, death, and migration. It assumes that the number of user nodes in a social network is equal at any given time within a fixed period. Therefore, the sum of the ratios S of users in different states within the same layer of the model at any given time is... i +I i +R i =1.

[0090] The spread of topics is similar to the spread of infectious diseases. When a new user comes into contact with a user who has already spread the topic, there will inevitably be a certain rate of infection.

[0091] When a derivative topic is created, it will directly inherit the proportions of susceptible, infected, and immune users from the preceding topic.

[0092] At the same topic level, when an infected node comes into contact with a susceptible node, without considering other influencing factors, the susceptible node will be converted into an infected node with probability λ. When psychological factors and prior emotions are considered, this probability becomes λ. p or λ q .

[0093] As the topic's popularity diminishes and time progresses, infection nodes will reappear with a fixed probability. It can be transformed into an immune node. However, for a given node, an immune node can still be transformed into a susceptible node of another topic layer with probability p, and become interested in that other topic layer again.

[0094] like Figure 3As shown, this invention uses quantitative and continuous attention to psychological factors to better predict users' forwarding process. If there exist topic 0 (precursor topic), topic 1 (derived topic 1), and topic 2 (derived topic 2); where topic 1 is derived from topic 0, topic 2 is derived from topic 1, and topic 1 has a high correlation with topic 0, while topic 2 has a low correlation with topic 1. When topic 1 is generated, the user network is distinguished by whether or not topic 0 has been spread. Therefore, when a user first encounters a derived topic, there are two states: state X1∈{G0S1,G1S1}, where G0S1 represents not having forwarded the predecessor topic 0 and being in a susceptible state for derived topic 1, and G1S1 represents having forwarded the predecessor topic 0 and being in a susceptible state for derived topic 1. Similarly, when topic 2 is generated, the user network is distinguished by whether or not topic 1 has been spread. Then, there is state X2∈{G0S2,G1S2}, with the same meaning.

[0095] Assume that for an independent topic, the infection rate is λ, which is the probability of transitioning from state S to state I. When topic 2 is spreading, since it has little correlation with topic 1, the probability of a user forwarding topic 2 is unaffected regardless of whether the user has forwarded topic 0. That is, the probability of transitioning from state G0S2 to state G1S2 to state I2 remains λ.

[0096] The transition probability between different user classes is defined as follows:

[0097] The cross-model state transition probability is calculated based on topic relevance and the user's independent forwarding probability. Immune users are transformed into susceptible users who participate in derivative topics based on the cross-model state transition probability. The formula for calculating the cross-model state transition probability is:

[0098]

[0099] Where, p ab Rlv(r) represents the cross-model state transition probability, i.e., the transition probability from an immune user who participated in topic a to a susceptible user who participated in topic b. λ represents the user's independent forwarding probability. a ,r b ) represents the topic relevance between topic a and topic b, and β represents the topic relevance threshold.

[0100] Since topic 1 is highly related to topic 0, if a user has already forwarded topic 0, the probability of them forwarding topic 1 will increase due to the psychological effect of continued attention, reaching λ. p The conversion probability of susceptible users who participated in the preceding topic is calculated based on the topic relevance and the user's independent forwarding probability. Susceptible users are then converted into infected users who participate in the derivative topic based on this conversion probability. The calculation formula is as follows:

[0101] λ p =λ+(1-λ)×Rlv(r) a ,rb )

[0102] Where, λ p Rlv(r) represents the switching probability of susceptible users who have participated in the preceding topic, λ represents the independent forwarding probability of a user, and Rlv(r) represents the switching probability of a susceptible user who has participated in the preceding topic. a ,r b Rlv(r) represents the topic relevance between topic a and topic b. a ,r b = Rlv(r0,r1).

[0103] If a user has not forwarded topic 0, and topic 1 is similar to topic 0, the probability of the user forwarding topic 1 will decrease to λ. q Based on topic relevance and users' independent forwarding probability, the conversion probability of susceptible users who have not participated in the preceding topic is calculated. Susceptible users are then converted into infected users who participate in the derivative topic based on this conversion probability. This conversion probability is the probability of G0S1 converting to I1.

[0104] λ q =(1-Rlv(r) a ,r b ))×λ

[0105] Where, λ q Rlv(r) represents the probability of a susceptible user switching from a topic that has not been involved in the preceding discussion, λ represents the user's independent forwarding probability, and Rlv(r) represents the probability of switching from a topic that has not been involved in the preceding discussion. a ,r b Rlv(r) represents the topic relevance between topic a and topic b. a ,r b = Rlv(r0,r1).

[0106] Users are infected with a fixed probability. Convert to immune users, preferred.

[0107] S7: Construct a dynamic equation based on the user's independent forwarding probability and the conversion probability between different types of users.

[0108] Based on the Iterative-SIR model and the above propagation rules, a dynamic equation for k derivative topic layers is constructed according to the independent forwarding probability of users and the conversion probability between different types of users. Its expression is as follows:

[0109]

[0110] Furthermore, the dynamic equation can be written as:

[0111]

[0112] in, Rlv(r) represents the independent forwarding probability of a user; α represents the correlation between topic i and its predecessor topics. i-1 ,r i G0 represents a user network that has not participated in the dissemination of the precursor topic, and S... i (t) and I i (t) represents the proportions of susceptible and infected users participating in topic i at time t, respectively. i-1 (t) represents the proportion of immune users who participated in the propagation of the precursor topic i at time t, G1 represents the user network that participated in the propagation of the precursor topic, and p (i-1)i S represents the cross-model state transition probability between topic i and its predecessor topics, β represents the topic relevance threshold, and S i I i R i These represent the proportions of susceptible, infected, and immune users at any given time.

[0113] S8: Solve the dynamic equations to obtain the prediction results of user propagation of derivative topics.

[0114] Solving the dynamic equations yields the user state set {S} at different times and different topic layers. i (t)}{I i (t)}{R i (t)} represents the predicted spread of derivative topics by users.

[0115] Based on the output of the derivative topic propagation model of this invention, the propagation trend of derivative topics can be simulated and predicted, obtaining the proportion of user states in the network at each time point and the propagation trend of the entire topic. Public opinion monitoring departments can use these output results to monitor the topic propagation trend in real time and take corresponding control measures according to the actual situation and changes in topic propagation, thereby effectively preventing and resolving public opinion crises.

[0116] 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 method for predicting the spread of derivative topics based on multi-topic sentiment measurement, characterized in that, include: S1: Obtain topic data, which includes user history behavior information and topic participation information; S2: Extract message factor features and user factor features from topic data; message factors include topic popularity and topic relevance, while user factors include user engagement, user activity, and topic personal sentiment matching degree. S3: Calculate message influence based on topic popularity, user engagement, user activity, and topic-based personal sentiment matching. S4: Divide the user network to obtain the user classification results; S5: Calculate the user's independent forwarding probability based on message influence; the formula for calculating the user's independent forwarding probability is: ； ； in, Indicates the probability of a user independently forwarding the message. Indicates intermediate parameters. Indicates user The number of neighbors, This indicates the number of neighbors who forwarded the topic. Indicates user The influence of the news; S6: Calculate the conversion probability between different user groups based on topic relevance and users' independent forwarding probability; the process of calculating the conversion probability between different user groups based on topic relevance includes: The cross-model state transition probability is calculated based on topic relevance and the user's independent forwarding probability. Immune users are transformed into susceptible users to participate in derivative topics based on the cross-model state transition probability. The formula for calculating the cross-model state transition probability is: ； in, This represents the probability of a user who was immune to topic a transition to a user who was susceptible to topic b. This indicates the degree of topic relevance between topic a and topic b. Indicates the threshold for topic relevance; The conversion probability of susceptible users who have participated in the preceding topic is calculated based on the topic relevance and the user's independent forwarding probability. Susceptible users are then converted into infected users who participate in the derivative topic based on this conversion probability. The conversion probability of susceptible users who have not participated in the preceding topic is calculated based on the topic relevance and the user's independent forwarding probability. Susceptible users are then converted into infected users who participate in the derivative topic based on this conversion probability. Infected users become immune users with a fixed probability; S7: Construct a dynamic equation based on the independent forwarding probability of users and the conversion probability between different user classes; the dynamic equation is expressed as: ； in, , representing the user's independent forwarding probability; Indicate topic Its relevance to the preceding topics , This indicates a user network that has not participated in the dissemination of the pioneering topic. and These represent participation in the topic at time t. The ratio of susceptible users to infected users, This indicates participation in the topic at time t. The percentage of immune users to the preceding topics This indicates a user network that participated in spreading the pioneering topic. Indicate topic The cross-model state transition probability of its predecessor topic, , , These represent the proportions of susceptible, infected, and immune users at any given time. S8: Solve the dynamic equations to obtain the prediction results of user propagation of derivative topics.

2. The derivative topic propagation prediction method based on multi-topic sentiment measurement according to claim 1, characterized in that, The process of calculating message influence includes: calculating message factor influence based on topic popularity and user engagement; calculating user factor influence based on user activity and topic-based personal sentiment matching; and calculating message influence based on both message factor influence and user factor influence.

3. The derivative topic propagation prediction method based on multi-topic sentiment measurement according to claim 2, characterized in that, The formula for calculating the impact of a message is: ； in, Indicates user The influence of the news This represents the first partial regression coefficient. This represents the second partial regression coefficient. Indicates user User factors have an influence. This represents the third partial regression coefficient. Indicates user The influence of news factors.

4. The derivative topic propagation prediction method based on multi-topic sentiment measurement according to claim 1, characterized in that, The process of segmenting the user network includes: dividing users into those who have participated in the preceding topics and those who have not; further dividing users who have participated in the preceding topics and those who have not participated in the preceding topics into susceptible users, infected users, and immune users, thus obtaining the user classification results.

5. The derivative topic propagation prediction method based on multi-topic sentiment measurement according to claim 1, characterized in that, The formula for calculating the switching probability of susceptible users who have participated in precursor topics is: ； in, This indicates the switching probability of susceptible users who have participated in precursor topics. This represents the user's independent forwarding probability. This indicates the degree of topic relevance between topic a and topic b.

6. The derivative topic propagation prediction method based on multi-topic sentiment measurement according to claim 1, characterized in that, The formula for calculating the switching probability of susceptible users who have not participated in the preceding topics is: ； in, This indicates the probability of switching among susceptible users who have not participated in the preceding discussion. This represents the user's independent forwarding probability. This indicates the degree of topic relevance between topic a and topic b.

Images