A method for predicting the spread of derivative topics based on emotion transfer and evolutionary game theory

Abstract

This invention belongs to the field of Internet application technology, specifically relating to a method for predicting the spread of derivative topics based on sentiment transfer and evolutionary game theory. The method includes: acquiring user topic spread data; preprocessing the topic spread data; extracting relevant attributes from the preprocessed topic spread data; dividing the extracted relevant attributes into training and testing sets; performing sentiment analysis and topic spread analysis on the data in the training set to obtain a topic-user sentiment matrix; inputting the topic-user sentiment matrix into a derivative topic spread prediction model for training; testing the trained derivative topic spread prediction model using a testing set to obtain the optimal derivative topic spread prediction model; and processing the data to be spread using the optimal derivative topic spread prediction model to obtain the derivative topic spread prediction result. This invention optimizes the performance and accuracy of the model.

Description

Technical Field

[0001] This invention belongs to the field of Internet application technology, specifically involving a method for predicting the spread of derivative topics based on emotional transfer and evolutionary game theory. Background Technology

[0002] In today's digital age, social networks have become crucial channels for information dissemination and the formation of public opinion. User interactions on these platforms, such as posting, commenting, liking, and reposting, not only reflect personal emotional inclinations but also drive the spread and evolution of topics. For example, trending topics on Weibo often spark heated discussions and become focal points of public attention. With the popularization of social networks and the increase in the number of users, massive amounts of data are generated and stored on these platforms. This data contains valuable information that can provide insights into social trends and predict the spread of topics.

[0003] In recent years, researchers have conducted extensive studies on early-stage topic propagation prediction, primarily employing collaborative filtering algorithms, traditional machine learning algorithms, and deep learning algorithms. These methods aim to understand user interests and behavioral patterns, topic characteristics, and propagation trends. With the continuous expansion of e-commerce platform data and the gradual maturation of deep learning technology, nonlinear models such as neural networks and deep learning have gradually become the mainstream choice for marketing data analysis and prediction models. Alireza et al. proposed an ensemble transfer learning multimodal sentiment analysis model based on weighted convolutional neural networks, utilizing the fusion of text and image classifier outputs, and ultimately using extended Dempster–Shafer theory to determine sentiment polarity at the decision level.

[0004] Currently, there are still some problems in predicting the spread of early-stage derivative topics: 1. Sparsity of early-stage topics. Early-stage topics have fewer participants, and user behavior data such as forwarding and commenting on the same topic is insufficient to support the prediction of derivative topic spread, which will inevitably affect the prediction accuracy of the entire spread model. 2. Complexity of user sentiment towards topics. Different users have different sentiment inclinations towards different topics. The complexity of topic sentiment can lead to the curse of dimensionality during model training and inference. How to effectively vectorize complex features in a low-rank manner will be a challenge. 3. The adversarial and cooperative relationship between primary and derivative emotions in the spread of derivative topics. When primary and derivative emotions are in conflict, contradictions and conflicts will arise, which may adversely affect the spread of derivative topics. When primary and derivative emotions cooperate, they may promote the spread of derivative topics. The adversarial and cooperative relationship between primary and derivative emotions will have different effects on the spread of derivative topics. Determining which strategies to promote or inhibit the spread of derivative topics is an urgent problem to be solved. Summary of the Invention

[0005] To address the problems existing in the above-mentioned technologies, this invention proposes a method for predicting the spread of derivative topics based on sentiment transfer and evolutionary game theory. This method includes:

[0006] Acquire user topic dissemination data and preprocess the topic dissemination data; the topic dissemination data includes original topic data, derivative topic data, and user data;

[0007] The method of transfer learning is used to fuse the preprocessed original topic data and derived topic data, and the generative adversarial network is used to process the fused topic data to generate topic features of specific types; a user topic relationship matrix is ​​constructed based on the topic features and user data.

[0008] The ME2vce algorithm is used to perform sentiment analysis on the user-topic relationship matrix to obtain a user-topic sentiment knowledge graph;

[0009] Obtain relevant attributes from user data, and integrate these attributes with the user-topic sentiment knowledge graph to obtain a user feature matrix;

[0010] The user feature matrix is ​​input into a pre-trained derivative topic propagation prediction model to obtain the derivative topic propagation prediction results.

[0011] The beneficial effects of this invention are:

[0012] This invention employs the ME2vec algorithm to construct the relationship between user sentiment and topics, utilizes semantic relationships to infer the similarity and relevance of sentiments among users, and introduces sparse representation from the perspectives of time, space, and sentiment to vectorize the feature space of derived topics into a low-rank vector. Finally, it uses an evolutionary game theory and adaptive mechanism-based native-derived topic propagation model to predict early-stage derived topic propagation, simulating the strategy selection of nodes in topic propagation, thereby optimizing the performance and accuracy of the model. Attached Figure Description

[0013] Figure 1 This is an overall flowchart of the present invention;

[0014] Figure 2 A schematic diagram illustrating the enrichment of the native topic domain by the DATR-GAN algorithm of this invention;

[0015] Figure 3 This is a schematic diagram of the ME2vec algorithm of the present invention;

[0016] Figure 4 This is a schematic diagram of the AEGT-GCN model of the present invention;

[0017] Figure 5 This is a flowchart of the derivative topic propagation prediction method based on emotion transfer and evolutionary game theory of the present invention. Detailed Implementation

[0018] 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.

[0019] A method for predicting the spread of derivative topics based on emotional transfer and evolutionary game theory, such as Figure 5 As shown, the method includes:

[0020] Acquire user topic dissemination data and preprocess the topic dissemination data; the topic dissemination data includes original topic data, derivative topic data, and user data;

[0021] The method of transfer learning is used to fuse the preprocessed original topic data and derived topic data, and the generative adversarial network is used to process the fused topic data to generate topic features of specific types; a user topic relationship matrix is ​​constructed based on the topic features and user data.

[0022] The ME2vce algorithm is used to perform sentiment analysis on the user-topic relationship matrix to obtain a user-topic sentiment knowledge graph;

[0023] Obtain relevant attributes from user data, and integrate these attributes with the user-topic sentiment knowledge graph to obtain a user feature matrix;

[0024] The user feature matrix is ​​input into a pre-trained derivative topic propagation prediction model to obtain the derivative topic propagation prediction results.

[0025] like Figures 1-4 As shown, the input of this invention is the user's basic information, original topic data, and user relationship information on the social platform. The output after passing through the prediction model is the prediction result of whether the user will participate in the topic.

[0026] This invention includes the following three steps:

[0027] S1: Online Data Acquisition. The method for acquiring data is to directly query the topic dissemination data in the database provided by the company. What is needed here is the interaction data of the topic dissemination during its lifecycle (such as forwarding, commenting, sharing, etc.), and the historical behavior data of users who interacted with the topic dissemination. For topic dissemination participation, we need to obtain the time when users participated in the topic interaction, the basic information of the interacting users, and the basic information of the topic.

[0028] S2: Extract relevant attributes. Based on the basic information and historical behavior of participating users during the lifecycle of a specific topic activity, relevant attributes are extracted from aspects such as the topic content and interaction information.

[0029] S3: Model Building. First, user relationship networks can be obtained through data such as original social networks and interaction frequencies; then, a user-topic sentiment matrix is ​​constructed to map users' sentiment tendencies towards different topics; finally, it is integrated with the original user basic data, and a derivative topic propagation prediction model AEGT-GCN is constructed using a graph convolutional neural network to study the development trend of user groups towards a certain derivative topic.

[0030] In this embodiment, obtaining the data source includes: obtaining raw data. Raw data can be obtained through public APIs of social networks or by directly downloading existing data sources. Simple data cleaning. Raw data is usually unstructured and cannot be directly used for data analysis. Simple data cleaning can structure most of the unstructured data. For example, deleting duplicate data and cleaning up invalid information.

[0031] In this embodiment, the extracted attributes related to topic dissemination data include: user's own attributes, user relationship network, user's historical behavior, topic information, social intimacy, user's social influence, and message dissemination popularity.

[0032] S21: User's own attributes User(u i );

[0033] A user's personal attributes include their age, number of followers, and number of accounts they follow, which are somewhat correlated with the development of topics. Therefore, user personal attributes are defined as follows:

[0034] S22: User Relationship Network R u ;

[0035] A user relationship network includes all participating users and the relationships between them; therefore, a user relationship network is defined as follows:

[0036] R u ={U,E u}

[0037] S23: User Historical Behavior Action(u i );

[0038] User historical behavior includes the content and number of posts, the content and number of reposts, and the content and number of comments. These attributes reflect user behavioral preferences. Therefore, user historical behavior is represented as:

[0039] Action(u i)=[public{n,c},forword{n,c},comment{n,c}]

[0040] Where public{n,c} represents the content c and the number n of posts made by users on the topic, forword{n,c} represents the content and number of posts forwarded by users on the topic, and comment{n,c} represents the content and number of comments made by users on the topic.

[0041] S24: Topic information (w);

[0042] The topic information includes information about the products mentioned in the topic, the start time, and the content. Therefore, the topic information is:

[0043] topic(w)=[product{w,p},time(w),content(w)]

[0044] Where product{w,p} represents information about product p in topic w, time(w) represents the start time of the topic, and content(w) represents the content of the topic.

[0045] S25: Social Intimacy {u i ,u j};

[0046] Social intimacy refers to user interactions such as forwarding, liking, and commenting with other users. Users with higher intimacy levels are more likely to participate in each other's conversations. Therefore, social intimacy is defined as:

[0047] intimacy{u i ,u j}=[retran{u i ,u j},like{u i ,u j},review{u i ,u j}]

[0048] Where, retran{u i ,u j} represents the number of times a user forwards another user's topic, like{u i ,u j} represents the number of times a user liked another user's topic, review{u i ,u j} indicates the number of times a user comments on another user's topic.

[0049] S26: User's social influence Inf(u i );

[0050] In social networks, a user's social influence is a key channel for spreading topics and often determines the course of events. Users with greater influence have a wider reach. Therefore, user social influence is defined as follows:

[0051]

[0052] Among them, fol(U) i ), fol avg (net) represents user U respectively i The number of fans and the average number of followers a user follows on the internet.

[0053] S27: Message dissemination popularity MsgHeat(m);

[0054] MsgHeat(m) represents the current popularity of message m. Generally, the higher the popularity of a message, the more likely it is to attract other users to take action. The definition of message popularity is as follows:

[0055]

[0056] Where Num[transpond(m)] represents the number of times message m is forwarded, and Num[fans(ui)] represents the original creator u of message m. i The number of followers is used, and α and β are adjustable parameters, where α, β∈(0,1), to neutralize the difference in influence between the number of message forwards and the number of followers. Furthermore, since the spread of a message gradually decreases over time, this paper introduces a time decay function. This simulates how message popularity changes over time, where λ is the popularity decay factor, which varies for different topics and is derived from fitting real data. t is the current time, and t0 is the original creation time of the message.

[0057] In this embodiment, establishing the Mo model includes three main stages: topic relationship construction, sentiment measurement between users and topics, and model algorithm design, such as... Figure 1 As shown. In the first stage, attribute data that may influence user behavior is extracted from the original topic domain and derived topic domain. Then, the IT2vec representation learning algorithm is used to perform multi-domain iterative representation of this information to uncover richer features hidden in the topic space.

[0058] Simultaneously, generative adversarial networks are introduced to enhance implicit relationships between topics. In the second stage, the ME2vec algorithm is proposed to construct a derived sentiment knowledge graph based on the SentiWordNet dictionary, using semantic relationships to infer users' sentiment tendencies towards different topics. In the third stage, an AEGT-GCN native-derived topic propagation model based on evolutionary game theory and adaptive mechanisms is established to predict early-stage derived topic propagation and simulate the strategy choices of nodes in topic propagation, thereby optimizing the model's performance and accuracy.

[0059] The process of using a generative adversarial network (GAN) to process the fused topic data includes: randomly sampling the fused topic data and setting conditional information; inputting the randomly sampled data and conditional information into a topic data generator to generate specific topic data; using a topic discriminator to verify the probability that the generated topic data is real data; calculating the loss function of the GAN based on the calculated probability, and optimizing and updating the parameters in the topic discriminator based on the loss function; updating the parameters in the topic data generator based on the updated parameters of the topic data discriminator; and iterating repeatedly on the input data by the topic data generator and the topic discriminator according to the rules of local optimization and global optimization until the topic data generated by the topic data generator is similar to the real topic data, thus obtaining the enhanced topic features.

[0060] The rules for local and global optimization include: calculating the parameter gradients of the topic data generator and the topic discriminator based on the loss function; updating the parameters of the topic data generator based on the parameter gradients of the topic data generator to minimize the loss function value; and updating the parameters of the topic discriminator based on the parameter gradients of the topic discriminator to maximize the loss function value.

[0061] Sentiment analysis of the user topic relationship matrix using the ME2vce algorithm includes: randomly generating a topic distribution for each document using the Dirichlet assignment model; modeling the topic distribution using the Dirichlet distribution as the conjugate prior distribution of the multinomial to obtain the topic themes corresponding to users; and constructing a user-topic sentiment knowledge graph based on the topic themes.

[0062] Specifically, it includes:

[0063] S31: Early sparse topic compensation;

[0064] First, obtain a rich dataset of pre-labeled original topic domain features, D. S And the relatively sparse, unlabeled derivative topic domain dataset D t The original topic domain data is D. s ={(x s1 ,y s1 ),...,(x sn,y sn )}, where x si ∈X is the input, y si ∈X represents the corresponding output. Similarly, the derivative topic domain data is D. t ={x t1 ,...,x tn}, where the input is y ti ∈Y. Use P(X) respectively. S ) and Q(X t P(X) represents the data distribution of the original topic domain and the data of the derived topic domain. S ) and Q(X t Based on this, and combined with the domain adaptation method of TCA, a P(X) is found. S ) and Q(X) t A common latent representation H exists between the two domains, which retains common configuration information after transformation. That is, there exists a nonlinear mapping function φ:X→H, such that X′ S ={x′ si}={φ(x si )},X′ t ={x′ ti}={φ(x ti The expected final value is P(X). s ′)=Q(X t Therefore:

[0065]

[0066] We only need to minimize the distance between the two distributions to obtain an optimal nonlinear mapping function φ, such that in P(X) s )≠Q(X t In the case of ), P(Y) is obtained through the mapping function φ. s |X s ′)=P(Y t |X t ′). Subsequently, the optimal solution of the nonlinear mapping function φ is learned using the TCA-based eigenvalue decomposition algorithm.

[0067] S32: Sentiment measurement between users and topics.

[0068] Considering that users have different emotional inclinations towards different topics, and these inclinations may directly influence whether users participate in a topic, effectively quantifying users' emotional inclinations towards different topics has a positive impact on predicting the spread of derivative topics. To delve deeper into this phenomenon, we first treat users' historical participation in topics as documents d, and use Latent Dirichlet Allocation (LDA) to randomly generate a topic distribution θ for each document d. d For each word w in topic d, according to θd Randomly select a topic z, and then analyze the word distribution φ of topic z. z This generates the word w. Simultaneously, it uses the Dirichlet distribution as the conjugate prior distribution of the multinomial distribution, the prior distribution of the document's topic, and the prior distribution of the word within the topic:

[0069]

[0070] Here, B(α,β) is the beta function, used to normalize the probability density so that the sum is 1. α and β are parameters of the distribution, random variables taking values ​​in the range [0,1].

[0071] The probability mass function of the multinomial distribution is:

[0072]

[0073] Where n is the total number of trials, x i p is the number of times the i-th result appears. i Let i be the probability of the i-th result occurring. Maximize the lower bound of the evidence, i.e.:

[0074] L(q)=Ε q [logp(x,z)]-E q [logq(z)]

[0075] Where q(z) is the variational distribution, p(x,z) is the joint distribution, and E represents the expectation.

[0076] The sentiment analysis tool TextBlob was used to assess users' sentiment towards the topic and used as the weights of the user-topic sentiment knowledge graph edges. User features such as age and gender were also integrated into the feature matrix.

[0077] S33: Design of propagation prediction model.

[0078] Current research on topics derived from social networks rarely considers both user-to-user relationships and user-topic sentiment bias simultaneously. Building upon the user relationship networks and user-topic sentiment graphs discussed in the previous two sections, this paper proposes a prediction model called AEGT-GCN by designing and fusing a time decay function. This model utilizes a graph convolutional neural network as a deep learning model, enabling it to capture the relationships and structural features between nodes in graph data.

[0079] In this paper, the prediction of user group behavior is defined as a binary classification problem, namely, using the feature vectors of trending topics across the entire network during time period t, predicting whether potential users in time period t+1 will participate in the dissemination of trending topics. The model input for this paper is as follows:

[0080] Feature matrix M t = N×F, where N represents the number of user nodes in the hot topic propagation network, and F represents the input feature dimension of each node, including user self-attribute features and external factor features.

[0081] The adjacency matrix G of the user at time t t ={U t E t} represents the user relationship network under hot topics within time period t, where U t E represents the set of user nodes in the propagation space at time t. t This represents the set of edges between nodes in the propagation space.

[0082] Training the derivative topic propagation prediction model includes: constructing a training set; extracting features from the data in the training set to obtain a feature matrix; inputting the feature matrix into a graph convolutional neural network; adjusting the edge weights using a time decay function and an activity decay function to obtain a weight matrix; multiplying the feature matrix and the weight matrix and adding a bias; and then comparing this result with the average matrix of the user-topic relationship network. Multiplication; ReLU function is used to process the multiplication result and Dropout operation is performed; Sigmoid activation function is used to represent the convolution output as probability values ​​of different categories; loss function of the model is calculated based on probability values, the model parameters are adjusted, and model training is completed when the loss function value converges.

[0083] This paper uses a two-layer graph convolutional neural network with Dropout as a hot topic behavior prediction model. First, the weights W and biases are randomly initialized. In the first layer of the GCN, the feature matrix M is first... t The edge weights are adjusted using time decay functions and activity decay functions.

[0084]

[0085] Among them, w i,j Indicates user u i ,u j The weighting coefficients between them Let be the time decay function. Let σ be the activity decay function, τ be the time decay factor, and t be the activity decay factor. i , t j This is the timestamp for the input time and the output time.

[0086] User feature matrix M t Multiply with the weight matrix W and add the bias, and then... Multiplication is performed. Then, the ReLU function is used as the activation function for this layer, and Dropout is applied during model training. Finally, the Sigmoid activation function is used to represent the convolution output as probability values ​​for different categories. The final formula is expressed as:

[0087]

[0088] Among them, W l Let be the weight matrix corresponding to the l-th layer in the graph convolutional network. Add a self-loop I to the adjacency matrix G.

[0089]

[0090] Since the Sigmod function can compress the output value to the range [0,1], an output value greater than 0.5 indicates that the user is participating in the current topic dissemination, while an output value less than 0.5 indicates that the user is not participating.

[0091] The loss function is defined as:

[0092]

[0093] Where N is the size of the experimental dataset, Y i Provide the real sample label for the i-th user in the dataset. This is the output of the user behavior prediction model.

[0094] 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 emotional transfer and evolutionary game theory, characterized in that, include: Acquire user topic dissemination data and preprocess the topic dissemination data; The topic dissemination data includes original topic data, derivative topic data, and user data. The transfer learning method is used to fuse the preprocessed original topic data and derived topic data, and the generative adversarial network is used to process the fused topic data to generate topic features of specific types. Construct a user topic relationship matrix based on topic characteristics and user data; The process of using a generative adversarial network (GAN) to process the fused topic data includes: randomly sampling the fused topic data and setting conditional information; inputting the randomly sampled data and conditional information into a topic data generator to generate specific topic data; using a topic discriminator to verify the probability that the generated topic data is real data; calculating the loss function of the GAN based on the calculated probability, and optimizing and updating the parameters in the topic discriminator based on the loss function; updating the parameters in the topic data generator based on the updated parameters of the topic data discriminator; and iterating repeatedly on the input data by the topic data generator and the topic discriminator according to the rules of local and global optimization until the topic data generated by the topic data generator is similar to the real topic data, thus obtaining the enhanced topic features. The ME2vce algorithm is used to perform sentiment analysis on the user-topic relationship matrix to obtain a user-topic sentiment knowledge graph. Specifically, this includes: using the Dirichlet distribution model to randomly generate a topic distribution for each document; using the Dirichlet distribution as the conjugate prior distribution of the multinomial to model the topic distribution to obtain the topic themes corresponding to users; and constructing a user-topic sentiment knowledge graph based on the topic themes. Obtain relevant attributes from user data, and integrate these attributes with the user-topic sentiment knowledge graph to obtain a user feature matrix; The user feature matrix is ​​input into a pre-trained derivative topic propagation prediction model to obtain the derivative topic propagation prediction results. Training the derivative topic propagation prediction model includes: constructing a training set; extracting features from the data in the training set to obtain a feature matrix; inputting the feature matrix into a graph convolutional neural network; adjusting the edge weights using a time decay function and an activity decay function to obtain a weight matrix; multiplying the feature matrix and the weight matrix and adding a bias; and then comparing this result with the average matrix of the user-topic relationship network. Multiplication; ReLU function is used to process the multiplication result and Dropout operation is performed; Sigmoid activation function is used to represent the convolution output as probability values ​​of different categories; loss function of the model is calculated based on probability values, the model parameters are adjusted, and model training is completed when the loss function value converges.

2. The derivative topic propagation prediction method based on emotion transfer and evolutionary game theory according to claim 1, characterized in that, The fusion of original topic data and derived topic data includes: using the TCA domain adaptation method to map topic data in the derived topic domain to topic data in the original topic domain, thus obtaining fused topic data.

3. The derivative topic propagation prediction method based on emotion transfer and evolutionary game theory according to claim 1, characterized in that, The rules for local and global optimization include: calculating the parameter gradients of the topic data generator and the topic discriminator based on the loss function; updating the parameters of the topic data generator based on the parameter gradients of the topic data generator to minimize the loss function value; and updating the parameters of the topic discriminator based on the parameter gradients of the topic discriminator to maximize the loss function value.

4. The method for predicting the spread of derivative topics based on emotion transfer and evolutionary game theory according to claim 1, characterized in that, The user feature matrix is ​​obtained by: analyzing the sentiment information contained in the document through TextBlob analysis and using the sentiment information as the weight of the edge of the user-sentiment knowledge graph; extracting relevant attributes of user data and integrating the relevant attribute features of users into the user-sentiment knowledge graph according to the weights to obtain the user feature matrix; among which, the relevant attributes include the user's age, gender, browsing preferences and comments data.

5. The derivative topic propagation prediction method based on emotion transfer and evolutionary game theory according to claim 1, characterized in that, The derivative topic propagation prediction model uses a two-layer graph convolutional neural network.

6. The derivative topic propagation prediction method based on emotion transfer and evolutionary game theory according to claim 1, characterized in that, matrix The expression is: in, It is a diagonal matrix. For the adjacency matrix of the user-topic relationship network.

7. The derivative topic propagation prediction method based on emotion transfer and evolutionary game theory according to claim 1, characterized in that, The loss function expression for the model is: Where N is the size of the experimental dataset, Y i Provide the real sample label for the i-th user in the dataset. This is the output of the user behavior prediction model.

Images