Method for predicting propagation of user psychological benefit and adversarial generation network false and real guiding topic game

Abstract

The present application relates to a kind of based on user psychological benefit and the spread prediction method of false and real guiding topic game of adversarial generative network, belong to internet application technical field.The method includes: definition related parameter in the definition of network topic propagation process;According to the user basic information of user, user relationship network, user historical behavior and guiding topic data construct false and real guiding topic data space;Through Simcse and entropy weight method quantification user psychological benefit, and through adversarial generative network to user topic feature data is enhanced;Through the evolution game theory quantification false and real topic between the antagonistic game relationship, and the mutual influence of false and real topic is integrated into user relationship network adjacency matrix;Combining the enhanced user topic feature data and containing mutual influence information for adjacency matrix, adopt PI-GCN model to predict user behavior, and predict the propagation trend of future guiding topic statistics.

Description

Technical Field

[0001] This invention belongs to the field of Internet application technology and relates to a propagation prediction method based on user psychological benefits and adversarial generative networks for the game of false and real guiding topics. Background Technology

[0002] In recent years, many scholars both domestically and internationally have conducted in-depth research on the situational awareness of misleading topics and have achieved certain results. Their research mainly focuses on three aspects: the correlation between topics and users, the problem of data sparsity and imbalance, and the influence relationships between users. This section focuses on the three issues mentioned above, highlighting the achievements of scholars in recent years.

[0003] Firstly, regarding the factors linking topics and users, Jing et al. constructed a user propagation intention inference model based on propagation characteristics (behavioral and temporal features) and a bidirectional backpropagation (B-BP) deep neural network to discover the relationship between false leading topics and the intensity of users' intention to propagate the topic. L. Fang et al. proposed a novel tree variational autoencoder model and used a cross-alignment method to align multiple modalities to obtain the relationship between rumors and the inherent sentiment of the text. D. Varshney et al. proposed a Bayesian network-based method to predict the propagation trend of potential topic information driven by factors such as user interests and content similarity. While these studies explore the propagation factors of false leading topics by studying the relationship between the characteristics of the topic itself and users, few have considered the psychological benefits users gain during the propagation of false leading topics and their impact on topic dissemination.

[0004] Secondly, regarding the issue of data sparsity and imbalance in topic propagation research, some scholars have attempted to mitigate the problems caused by data sparsity using representation learning methods. Chen J, Kong et al. alleviated the data sparsity problem in recommender systems through network representation learning, while Fu et al. proposed a new representation learning framework, HIN2Vec, for heterogeneous information networks to alleviate the prediction inaccuracies caused by data sparsity in link prediction applications. Some literature employs an activation maximization method—PPGNs (Plug and Play Generative Networks)—to improve sample quality and sample diversity. Li Q et al. utilized the high accuracy of tensor completion in recovering lost data to perform homomorphic compensation on topic data to address the sparsity and imbalance of data encountered in rumor propagation. These studies have effectively alleviated the problems caused by data sparsity; however, in the context of a game-theoretic topic space structure involving false and true topics, no one has considered the impact of data sparsity and imbalance on topic propagation research.

[0005] Finally, research addresses the issue of influence relationships between users. Some scholars have started by modeling the structural characteristics of social networks based on user influence, measuring the impact of different user nodes on information dissemination. M. Cha et al. used data from Twitter to compare three main parameters influencing node influence: in-degree, number of retweets, and number of mentions, finding that nodes with higher influence play a significant role in multiple topics. Hara Fukunaga et al. proposed a user influence index model from the perspectives of both users and user microblogs. Literature also uses a large-scale field experiment to examine the role of social networks in online information dissemination, showing that weaker ties are the primary driver of information spread. Some scholars use game theory to study the interactions between users in social networks, thereby predicting individual behavior. Wang EK et al. proposed an incentive evolutionary game model for opportunity social networks, effectively combining credit-based incentive methods with evolutionary game models to eliminate abnormal nodes and maintain network node stability and reliability. These studies all use evolutionary game theory to study user relationships in social networks. However, in the context of a social network space structure of fake topics versus real topics, the game competition between fake topics and real topics and the potential relationship of influence among users are worthy of researchers' attention and study.

[0006] False and misleading topics refer to inaccurate and false information deliberately created by creators to mislead people for a specific purpose. These topics contain superficial or one-sided statements that fail to objectively reflect the essence of things and are typically purposeful, misleading, and verifiable. The rise of popular social media platforms such as Facebook, Twitter, and Weibo has provided a suitable environment for the spread of false and misleading topics, seriously negatively impacting daily life and national social stability. In academia, S / N has published numerous articles in recent years on the spread of false and misleading topics, describing the many harms caused by their dissemination on social networks. Therefore, effectively utilizing scientific and technological methods to help governments and relevant departments promptly detect and control the spread of false and misleading topics, create a correct and truthful online environment, and disseminate accurate and objective information about guiding topics is crucial.

[0007] Among numerous online topic dissemination events, we have observed a common phenomenon: whenever a false topic spreads widely on social networks, the users who spread it are largely those who previously forwarded or posted similar topics and received strong feedback from other netizens. For example, in this case study, a user previously forwarded the topic "Persimmons easily cause allergies and other adverse reactions," which garnered many likes, comments, and shares. Later, when they saw the unproven false topic "Eating persimmons with seafood will cause poisoning," they were clearly more inclined to accept this information and chose to forward it. Therefore, the false topic exploited the user's psychological benefit from participating in the topic, further guiding them to forward the false topic. Simultaneously, during the spread of false topics, some accompanying topics also emerge, especially genuine topics that contradict the false topic. These genuine topics are often accurate and truthful statements from official media or authoritative institutions regarding the false topic, such as the authoritative institution's statement that "There is no scientific basis for the claim that eating persimmons with seafood will cause poisoning." These two topics constitute a game of "falsehood and truth," a relationship of confrontation and mutual promotion. Therefore, considering the psychological benefits to users, and taking into account the game-theoretic relationship between false and true topics, can more comprehensively reveal the dissemination patterns of such misleading and misleading topics, thereby helping governments, official platforms, and other entities to quickly and accurately control the spread of false topics. In summary, although many readers both domestically and internationally have achieved considerable success in predicting the spread of such misleading and misleading topics, research on the dissemination trends of multi-information misleading topics from a topic-leading perspective still faces some challenges:

[0008] 1. The impact of guiding topics on users' psychological gains is difficult to quantify. Whether the topics are genuine or fabricated, they often possess a degree of psychological guidance, leveraging the psychological benefits users gain from participating in related content to further encourage them to spread the guiding topic. Effectively identifying potential users who may have been manipulated from historical user data is a challenge. Furthermore, the feedback users receive during the dissemination of fabricated guiding topics is multifaceted, making it difficult to effectively extract and calculate the psychological gains from this data.

[0009] 2. The actual effective data in the space of false and true guiding topics is sparse and unbalanced. The proportion of forwarding of false and true guiding topics in the entire topic dissemination space is actually very small. Furthermore, after users who have forwarded the topic realize that they have forwarded a false topic, they may delete the data, resulting in scarce and unbalanced data, which poses a challenge to the study of the dissemination patterns of false and true guiding topics.

[0010] 3. Measuring the adversarial game relationship between fake and real topics is also a challenge. Fake topics often encounter obstacles from real topics in guiding user participation in their dissemination. How to accurately quantify this adversarial game relationship and effectively measure the psychological benefits to users participating in these topics are problems that need to be seriously addressed. Furthermore, the irregular structure of such social network data places certain demands on the selection of predictive models. Summary of the Invention

[0011] In view of this, the purpose of this invention is to provide a propagation prediction method for false and true guiding topic games based on user psychological gains and adversarial generative networks. The entropy weight method is used to quantify the psychological gains of users participating in topics, and the Simcse text representation learning method is combined to mine the impact of users' associated historical topic data on their psychological gains. Simultaneously, adversarial generative networks are used to homomorphically enhance the sample data of false and true guiding topic spaces, and evolutionary game theory is introduced. Combined with the actual adversarial game scenario in the propagation process of false and true guiding topics, the invention conducts predictive research on user behavior in this topic space.

[0012] To achieve the above objectives, the present invention provides the following technical solution:

[0013] A propagation prediction method based on user psychological gains and adversarial generative networks in a game of deceptive versus real-driven topic engagement, comprising the following steps:

[0014] S1. Define the relevant parameters in the process of online topic propagation and formalize the propagation prediction problem;

[0015] S2. Construct a space for false and true guiding topic data based on users' basic information, user relationship networks, user historical behavior, and guiding topic data;

[0016] S3. Quantify users' psychological benefits using Simcse and entropy weighting, and enhance user topic feature data using adversarial generative networks;

[0017] S4. Quantify the adversarial game relationship between false and real topics through evolutionary game theory, and integrate the mutual influence of false and real topics into the adjacency matrix of user relationship network;

[0018] S5. Combining enhanced user topic feature data with an adjacency matrix containing mutual influence information, the PI-GCN model is used to predict user behavior and predict the future spread trend of guiding topics.

[0019] Furthermore, in step S1, the following parameters are defined:

[0020] Define users who participate in the topic With topic dissemination network ,in, This represents the set of users participating in the topic at time t. This represents the set of directed edges connecting users who participated in the topic. Indicates user Follow users ;

[0021] Define the user's own basic attributes :

[0022]

[0023] in Indicates user age, Indicates user gender, Indicates user The number of followers Indicates user The number of users who follow;

[0024] Define user activity :

[0025]

[0026] in, Indicates user The number of original topics published within a certain period of time. Indicates user The number of times someone else's topic is forwarded within a certain period of time. , This represents the weights of the two factors;

[0027] Define user psychological benefits :

[0028]

[0029] in, For users' historical topics Topics of Falsehood and Reality Similarity between them Indicate topic The overall popularity is calculated based on three user behavior dimensions: likes, comments, and shares; M represents the number of topics a user has discussed in their past posts.

[0030] Define user intimacy :

[0031]

[0032] in, This is an indicator function that indicates to potential users. Do you pay attention to neighboring users? K represents the user. The total number of topics of interaction between them; This represents the user interaction behavior coefficient, which is determined based on different interaction behaviors. Let t be the time decay function, representing the current time. Indicates user For users The time of the a-th action in the k-th topic;

[0033] Defining the popularity of a topic :

[0034]

[0035] in This indicates the number of times the current topic has been forwarded within the first preset dissemination time after its publication. This indicates the number of times the current topic has been forwarded within the second preset dissemination period after its publication. This indicates the total number of times the current topic has been forwarded;

[0036] Formalizing the propagation prediction problem:

[0037]

[0038] in, The current user propagation network for both real and false topics is a directed graph consisting of user nodes U and user relationship edges E. Represents the set of all users' own attributes, where Represents the user's basic attribute set; Indicates user exist Take action at all times ,in This represents the set of users' historical behaviors during the spread of the topic; This refers to the method for predicting user participation in forwarding topics proposed in this paper; This indicates the predicted engagement behavior of potential users in both deceptive and genuinely suggestive topics.

[0039] Furthermore, in step S2, the false and true guiding topic data space is based on user basic information, user relationship network, user historical behavior and guiding topic data, and calculates user basic characteristics, user relationship adjacency matrix, user self factors and topic driving factors.

[0040] Furthermore, step S3 specifically includes the following steps:

[0041] S31. Obtain word embedding representations based on contrastive learning through the Simcse model, obtain the corresponding topic representation vectors, and calculate the cosine similarity between the historical topic representation vectors and the current false and true topic representation vectors.

[0042] S32. Calculate the weights of the topic popularity index using the entropy weight method;

[0043] S33. Calculate the user's psychological benefit based on the weights of the calculated text similarity and topic popularity indicators.

[0044] Furthermore, in step S31, a positive and negative sample group (X, X+, X-) is first constructed, where X is the original sample, X+ is the positive sample generated using X, and X- is the negative sample that is not related to X. The contrast loss is optimized by reducing the distance between X and X+ and increasing the distance between X and X- in the feature space.

[0045] The InfoNCE loss function is used as the optimization loss function, and it is expressed as follows:

[0046]

[0047] in, , and These are the normalized representation vectors of the original sample, positive sample, and negative sample, respectively. This refers to temperature hyperparameters.

[0048] Historical topics derived from the Simcse model a The representation vector is A ( , ,..., ,..., ), Fake vs. Real Topics b for B ( , ,..., ,..., If the cosine similarity is calculated as follows:

[0049]

[0050] The range of cosine values ​​is [-1, 1], where -1 means completely dissimilar and 1 means completely similar. The closer to 1, the more similar the two values ​​are, and the closer to -1, the less similar the two values ​​are.

[0051] Furthermore, in step S32, the objective weights of comments, likes, and reposts for each topic are calculated using the entropy weight method, which includes:

[0052] S321. Standardize the original data using the min-max standardization method. The standardized data is represented as follows:

[0053]

[0054] In the formula, This represents the j-th sample for the i-th indicator. and Let represent the minimum and maximum values ​​of the i-th indicator in the sample, respectively;

[0055] S322. Calculate the proportion of likes, comments, and shares for each topic in all samples:

[0056]

[0057] Where m is the number of samples;

[0058] S323. Calculate the entropy value of each indicator:

[0059]

[0060] in, It is a constant used to normalize the entropy value;

[0061] S324. Calculate the weight of each indicator using entropy values:

[0062]

[0063] Where n is the number of indicators.

[0064] Furthermore, in step S33, combining the calculated text similarity and topic popularity index weights, the user's psychological benefit from historical topics on the current spread of false and true topics is calculated as follows:

[0065]

[0066]

[0067] in M For the number of user's historical topics, N The number of historical hot topics is an indicator. and These are the spurious leading topic vector and the real leading topic vector calculated by Simcse, respectively.

[0068] Furthermore, step S4 includes:

[0069] S41. By introducing adversarial generative networks, we can mine the potential relationships generated by the adversarial interaction between fake and real topics in the data, and calculate the topic influence of fake or real topics. Furthermore, it quantifies the mutual influence between false and true topics through evolutionary game theory:

[0070]

[0071]

[0072] in, The payoff functions for the two game strategies, "forwarding false information" and "forwarding true information," are given respectively. and These represent the effects of fake and real topics on users after a game of negotiation. The impact of communication activities;

[0073] S42. Based on the mutual influence of false and real topic games under the influence of user psychological benefits, construct... t The Interplay Matrix of Fake and Real Topics in the Moment :

[0074]

[0075] in, .

[0076] Furthermore, in step S41, the adversarial generative network includes a generative model G and a discriminative model D. The goal of the generative model G is to deceive the discriminative model D, and the goal of the discriminative model D is to distinguish between the data generated by the generative model G and the real data. The two constitute a dynamic game process, ultimately aiming to reach a Nash equilibrium.

[0077] The real topic feature dataset is represented as datas[x1,x2,...,xn], assuming that the topic feature sequence follows a certain distribution. The method for establishing the generative model G is described as finding the maximum likelihood of the generative model, i.e.:

[0078]

[0079] The iterative process of generating and discriminating topic feature sequences is described as follows: Let Gz represent the topic sample generation model, and z represent the data after random sampling of the original topic feature sequence. The model G generates topic feature data from the randomly sampled data z.

[0080] D is a topic feature sequence discrimination model. For any input feature sequence x, the discrimination model outputs a real number between 0 and 1, which represents the probability that the set of feature sequences comes from real collected sample data.

[0081] and Let represent the distributions of real topic data and generated topic data, respectively. Then, the objective function of the discriminant model D is:

[0082]

[0083] The optimization function of the entire model is expressed as:

[0084]

[0085] The entire optimization process can be represented as iterating over D and G until the entire process converges, which can be expressed as: wait infinitely close to ;

[0086] The influence of fake or real topics It consists of user-specific factors and topic-driven factors. User-specific factors include: the user's basic attributes and user activity level, namely:

[0087]

[0088] The driving factors for a topic include: user intimacy and basic topic attributes, namely:

[0089]

[0090] Taking into account both user-specific factors and topic-driven factors, the following functions are constructed using a multiple linear regression algorithm to determine the influence of false and true topics:

[0091]

[0092]

[0093] in, , , These are the partial regression coefficients obtained using a multiple linear regression algorithm. , This indicates the proportion of each factor in the topic's influence;

[0094] Based on game theory, two game strategies are defined: "forwarding false information" and "forwarding true information," with payoff functions for each strategy as follows:

[0095]

[0096]

[0097] in and For users The ratio of friends spreading fake and real topics; and The topic influence is determined by the psychological benefits users experience from participating in historical topics, as well as the impact of related fake and real topics.

[0098] Finally, using evolutionary game theory to calculate the formula, and taking into account the impact of users' psychological gains, the final interaction between false and real topics is constructed.

[0099] Furthermore, in step S5, PI-GCN is used as the prediction model, wherein the input to the prediction model is:

[0100] Feature matrix X=N×A ,in N This indicates the number of user nodes in the network spreading false information. A It is the topic feature vector of each user node, which includes the user's own attribute features and topic driving factor features;

[0101] Adjacency Matrix in Game Theory ,express t At any given moment, the relationship connection information between all users in the two topic spaces of fake topics and real topics, under the influence of users' psychological benefits;

[0102] First, a two-layer graph convolutional neural network with an intermediate Dropout layer is used as the model for predicting online rumors. First, the weights and biases are randomly initialized. Then, X is multiplied by W, the bias is added, and then multiplied by Aˆ. Next, the ReLU function is used as the activation function for this layer, and Dropout is performed during model training. Finally, the SoftMax activation function is used to represent the convolutional output as the probability values ​​of different node categories; expressed as:

[0103]

[0104] in, Let be the weight matrix corresponding to the i-th layer in the graph convolutional network. ;

[0105] Let the model output Based on the model output, perform a three-class classification prediction problem, including potential users. We will forward fake topics and potential users in the next time period. We will forward real topics and potential users in the next time period. We will not participate in the dissemination of false or true topics during the next time period.

[0106] The beneficial effects of this invention are as follows:

[0107] This invention addresses the complex influencing factors of the psychological benefits of user participation in topics by proposing a method for quantifying user psychological benefits based on Simcse and entropy weighting. The Simcse representation learning method is used to extract and mine historical data related to user participation in both false and true guiding topics. The relationship between false and true guiding topics and users is found based on topic content similarity. Finally, the entropy weighting method is used to comprehensively evaluate user feedback on topic participation using multiple indicators, calculating the psychological benefits of user participation in both false and true guiding topics.

[0108] This invention addresses the problem of sparse actual effective samples for both genuine and false guiding topics by proposing a data augmentation method for guiding topics based on adversarial generative networks. The method uses an unsupervised adversarial generative network model to perform homomorphic data augmentation on topic samples, thereby more realistically restoring the relationship between users' psychological benefits and the spread of genuine and false guiding topics.

[0109] This invention proposes a PI-GCN-based propagation model of false and real topic game theory, which effectively handles the special structure of data in social networks. It introduces evolutionary game theory and combines the psychological benefits of users participating in topics to truly quantify the driving influence of users under the influence of the two topic games, constructs a user psychological benefit fluctuation matrix, and finally uses a propagation prediction model to effectively predict user behavior in the entire network topic space.

[0110] This invention can not only effectively predict user behavior and popularity in the context of false-true guiding topics, but also more realistically reflect the spread of such adversarial guiding topics under the influence of users' psychological benefits and social relationships.

[0111] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0112] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0113] Figure 1 This is a schematic diagram of the overall process of the propagation prediction method of the present invention based on user psychological benefits and adversarial generative networks for false and real guided topic games.

[0114] Figure 2This is a schematic diagram illustrating the process of obtaining topic vectors using the Simcse method of the present invention. Detailed Implementation

[0115] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0116] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0117] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0118] Please see Figures 1-2 This is a propagation prediction method based on user psychological benefits and adversarial generative networks in the game of false and real guiding topics.

[0119] Example

[0120] This embodiment presents a propagation prediction method for a game of false and real guided topics based on user psychological gains and adversarial generative networks, such as... Figure 1 As shown, it includes the following steps:

[0121] S1. Define the relevant parameters in the process of online topic propagation and formalize the propagation prediction problem;

[0122] S2. Construct a space for false and true guiding topic data based on users' basic information, user relationship networks, user historical behavior, and guiding topic data;

[0123] S3. Quantify users' psychological benefits using Simcse and entropy weighting, and enhance user topic feature data using adversarial generative networks;

[0124] S4. Quantify the adversarial game relationship between false and real topics through evolutionary game theory, and integrate the mutual influence of false and real topics into the adjacency matrix of user relationship network;

[0125] S5. Combining enhanced user topic feature data with an adjacency matrix containing mutual influence information, the PI-GCN model is used to predict user behavior and predict the future spread trend of guiding topics.

[0126] In step S1 of this embodiment, the main purpose of this method is to predict the potential user's forwarding of false or true topics and the development trend of false and true guiding topics in the topic space by analyzing user basic information, user relationships, user psychological benefits, and the game relationship between false and true topics. Based on this, this embodiment defines the participating users, topic propagation network, user's own basic attributes, user activity, user psychological benefits, user intimacy, and topic propagation popularity, and formalizes the problem solved by this invention according to the defined content.

[0127] Specifically, defining users who participate in the topic With topic dissemination network ,in This represents the set of users participating in the topic at time t. Let represent the set of directed edges relating users who participated in a topic. Indicates user Followed users That is, users User The focus is on users. The set of users participating in the topic and the set of user relationships constitute the topic dissemination network. .

[0128] Define the user's own basic attributes The basic attributes of a user include, for example, the user's age, gender, number of followers, and number of users they follow. These attributes have a certain influence on whether a user will participate in the dissemination of a topic. Therefore, the basic attributes of a user are defined in the text as follows:

[0129]

[0130] in Indicates user age, Indicates user gender, Indicates user The number of followers Indicates user The number of users who follow.

[0131] Define user activity User activity level represents a user's enthusiasm for participating in the dissemination of topics. Generally, the more active a user is, the greater the likelihood that they will forward rumors or debunking messages. In this embodiment, user activity level is represented as:

[0132]

[0133] in, Indicates user The number of original topics published in the previous month. Indicates user The number of times someone else's topic was shared in the previous month. , This represents the weights of the two factors, with values ​​ranging from 0 to 1.

[0134] Define user psychological benefits User psychological benefit is represented by the similarity between a user's past topics and current false or genuine guiding topics, as well as the popularity of past topics (number of likes, comments, and shares), defined as follows:

[0135]

[0136] in, For users' historical topics Topics of Falsehood and Truth Similarity between them Indicate topic The overall popularity is calculated based on three user behavior dimensions: number of likes, number of comments, and number of shares. M represents the number of topics a user has discussed in their past posts.

[0137] Define user intimacy User intimacy is represented by the mutual forwarding, liking, and commenting behaviors between a user and their friends. Users who interact more frequently are more likely to forward each other's topics. Furthermore, because user intimacy is highly time-sensitive, this embodiment introduces a time decay function. To quantify the impact of user interaction, user intimacy is defined as:

[0138]

[0139] in, For indicator functions, that is: , Indicates potential users Follow neighboring users , Indicates potential users I haven't followed any neighboring users. K represents the user. The total number of topics of interaction between them; This represents the coefficient of user interaction behavior, specifically,

[0140]

[0141] Indicates user Forwarded Topics posted Indicates user Commented Topics posted Indicates user Liked Topics discussed.

[0142] Let t be the time decay function, representing the current time. Indicates user For users The time of the a-th action in the k-th topic.

[0143] Defining the popularity of a topic The number of times a topic's information is forwarded within different time periods reflects its popularity. Therefore, in this embodiment, the popularity of a topic is defined as:

[0144]

[0145] in This indicates the number of times the current topic has been forwarded one hour after it was published. This indicates the number of times the current topic has been forwarded one day after it was posted. This indicates the total number of times the current topic has been forwarded.

[0146] Then, based on the parameters defined above, the problem of this invention, as well as the input and output of the problem, are clarified. Specifically, the user propagation network of the current false and true topics is defined as a directed graph composed of user nodes U and user relationship edges E, i.e. Meanwhile, this embodiment uses Represents the set of all users' own attributes, where Represents the user's basic attribute set; Indicates user exist Take action at all times ,in This represents the set of users' historical behaviors during topic dissemination; then, based on users' historical participation in topics, the psychological benefits of participating in the current topic are mined; finally, by combining user attributes, user intimacy, and user relationship networks, the participation behavior of potential users in deceptive and genuinely guiding topics is predicted. A more specific definition of the problem is as follows:

[0147]

[0148] In step S2 of this embodiment, user basic information, user relationship network, user historical behavior, and guiding topic data are used as model inputs to construct a data space for false and true guiding topics. Based on this, important information such as user basic characteristics, user relationship adjacency matrix, user personal factors, and topic driving factors are calculated.

[0149] In step S3 of this embodiment, the Simcse method is used as the core text representation method for calculating the similarity between user history topics and fake and real topics. In social networks, the spread of topics is often based on users' historical behaviors. For example, if a user has posted or forwarded topic information similar to fake and real topics, that user is more likely to participate in the spread of the current topic. Therefore, effectively calculating the similarity between user history topics and fake and real topics is particularly important. Specifically, this embodiment uses the Simcse word embedding representation method based on contrastive learning, which can more effectively handle short text data such as user topics. Simcse can also effectively combine and utilize the BERT pre-trained model and effectively generate accurate topic text vectors in an unsupervised learning mode, achieving more accurate short text similarity calculation results. Therefore, this embodiment selects the Simcse method as the core text representation method for calculating the similarity between user history topics and fake and real topics.

[0150] Simcse, short for Simple Contrastive Learning of Sentence Embeddings, can generate high-quality word vectors through contrastive learning in an unsupervised environment. The core idea of ​​contrastive learning is to construct a set of positive and negative sample pairs (X, X+, X-), where X is the original sample, X+ is a positive sample generated using X, and X- is a negative sample unrelated to X. The contrastive loss is optimized by reducing the distance between X and X+ and increasing the distance between X and X- in the feature space, thereby improving the quality of the resulting feature vectors. The positive and negative sample pairs (X, X+, X-) are crucial in contrastive learning. In Simcse, a sentence is input into the neural network twice. Due to the dropout layer in the model, neurons are randomly deactivated, resulting in different feature vectors generated by the two inputs. These two vectors are X and X+, while other sentences in the same batch are X-.

[0151] The Simcse model employs the InfoNCE loss function, whose main goal is to optimize sentence embeddings through a contrastive learning mechanism, making semantically similar sentences closer together in the embedding space, while dissimilar sentences are further apart. The formula for the InfoNCE loss function is as follows:

[0152]

[0153] in, , and These are the normalized representation vectors of the original sample, positive sample, and negative sample, respectively. The temperature hyperparameter is used. Since the text representation has been normalized, the InfoNCE loss function is essentially the cross-entropy softmax loss commonly used in multi-class classification tasks. The numerator in the loss function is the similarity between positive sample pairs, and the denominator is the similarity of all samples in the same batch. The training objective of the model is to minimize the loss function, that is, the larger the numerator and the smaller the denominator, the closer positive sample pairs become and the further apart negative sample pairs become in the same dataset. This is the core idea of ​​contrastive learning. These are common hyperparameters in Softmax. The smaller the value, the closer Softmax is to the true max function. The larger the value, the closer it is to a uniform distribution. The overall schematic diagram of the Simcse model is shown below. Figure 2As shown, for the historical topic word vectors and fake and real topic word vectors calculated by the Simcse model, the cosine similarity method is used to calculate the similarity between fake topics, real topics, and historical topics. Cosine similarity, used to determine the degree of similarity between texts, refers to the cosine value of the angle between two N-dimensional vectors in N-dimensional space, and measures the difference between the two vectors. Assume that the historical topics obtained by the Simcse model... a The representation vector is A ( , ,..., ,..., ), Fake vs. Real Topics b for B ( , ,..., ,..., If the cosine similarity is calculated as follows:

[0154]

[0155] The range of cosine values ​​is [-1, 1], where -1 means completely dissimilar and 1 means completely similar. The closer to 1, the more similar the two values ​​are, and the closer to -1, the less similar the two values ​​are.

[0156] The popularity of a topic is often a core factor reflecting the psychological benefits users gain from participating in it. The popularity of a user's historical reposts and posts is reflected in many aspects, including the number of reposts, comments, and likes. Entropy weighting is a weighting method used for comprehensive evaluation of multiple indicators, widely applied in decision analysis, evaluation, and ranking. It does not rely on the decision-maker's subjective judgment, automatically calculating the weight of each indicator through statistical analysis of data. It is highly suitable for multi-indicator, multi-dimensional decision-making problems, comprehensively considering the weight of each indicator, and automatically adjusting strategies based on data changes, making it suitable for handling complex decision-making environments. Therefore, this embodiment starts with topic popularity indicators such as the number of reposts, comments, and likes, using entropy weighting as a comprehensive evaluation method to more objectively and effectively measure the historical popularity of related topics, thereby calculating the psychological benefits users may gain from participating in fake and real topics.

[0157] This embodiment uses the entropy weight method to objectively calculate the weights of comments, likes, and shares for each topic, thereby calculating the direct psychological benefit of each topic to the user. The specific calculation steps are as follows:

[0158] 1. Data Standardization: Standardize the raw data to eliminate the influence of units, using the min-max standardization method. The standardized data can be represented as:

[0159]

[0160] in, This represents the j-th sample for the i-th indicator. and Let represent the minimum and maximum values ​​of the i-th indicator in the sample, respectively.

[0161] 2. Calculate the proportion of each indicator: Calculate the proportion of likes, comments, shares, etc. for each topic in all samples:

[0162]

[0163] Where m is the number of samples.

[0164] 3. Calculate the entropy value: Calculate the entropy value for each indicator. The formula for calculating the entropy value is:

[0165]

[0166] in, It is a constant used to normalize the entropy value.

[0167] 4. Calculate the weights: Calculate the weights of each indicator using entropy values. The formula for calculating the weights is:

[0168]

[0169] Where n is the number of indicators.

[0170] Combining the text similarity derived from the Simcse model and the topic popularity index weights calculated using the entropy weight method, the psychological benefits to users from the spread of historical topics versus current false and true topics can be calculated as follows:

[0171]

[0172]

[0173] in M For the number of user's historical topics, N This refers to the number of metrics for trending historical topics (likes, shares, comments). and These are the spurious leading topic vector and the real leading topic vector calculated by Simcse, respectively.

[0174] In step S4 of this embodiment, a Generative Adversarial Network (GAN) is introduced to solve the problem of sparse data between fake and real topics. Its adversarial mechanism is used to deeply explore the potential relationship between fake and real topics in the data and generate high-quality sample data based on this relationship.

[0175] A Generative Adversarial Network (GAN) consists of a generative model G and a discriminative model D. In this embodiment, the goal of the fake and real topic data generation model is to generate as much real topic data as possible to deceive the discriminative model D. Meanwhile, the goal of the discriminative model D is to distinguish the data generated by the generative model D from the collected real data. Thus, G and D constitute a dynamic "game." The optimization goal is to reach Nash equilibrium, where the generative and discriminative models continuously improve their respective generation and discrimination capabilities during the learning process. This allows the models to generate data that is homomorphically and identically distributed with the collected topic samples, thereby generating good user behavior and topic sample data to alleviate the sparsity of the actual effective cross-data between the fake and real topic spaces.

[0176] The topic-related dataset used in this embodiment can be represented as datas[x1,x2,...,xn], assuming that the topic feature sequence follows a certain distribution. Then, the method for establishing the generative model G can be described as finding the maximum likelihood of this generative model, that is:

[0177]

[0178] The iterative process for generating and discriminating topic feature sequences is described as follows: Let Gz represent the topic sample generation model, and z represent the data after random sampling of the original topic feature sequence. Model G generates topic feature data from the randomly sampled data z. D is a topic feature sequence discrimination model. For any input feature sequence x, Dx will output a real number between 0 and 1, which represents the probability that the feature sequence comes from real collected sample data. and Let represent the distributions of real topic data and generated topic data, respectively. Then, the objective function of the discriminant model is:

[0179]

[0180] The optimization function of the entire model can be expressed as:

[0181]

[0182] The entire optimization process can be represented as iterating over D and G until the entire process converges. This process can be expressed as: wait infinitely close to .

[0183] The influence of fake or real topics It consists of user-specific factors and topic-driven factors. User-specific factors include: the user's basic attributes and user activity level, namely:

[0184]

[0185] The driving factors for a topic include: user intimacy and basic topic attributes, namely:

[0186]

[0187] Taking into account both user-specific factors and topic-driven factors, the following functions are constructed using a multiple linear regression algorithm to determine the influence of false and true topics:

[0188]

[0189]

[0190] in, , , These are the partial regression coefficients obtained using a multiple linear regression algorithm, while , This indicates the proportion of each factor in the topic's influence.

[0191] Due to the unique and complex nature of social networks, while a piece of false information is spreading, genuine information is also spreading in a game of strategy. This game relationship is a significant factor influencing users' forwarding behavior. Therefore, this embodiment uses evolutionary game theory to quantify the mutual influence of false and genuine topics. First, based on game theory, two game strategies are defined: "forwarding false information" and "forwarding genuine information." The payoff functions for the two strategies are as follows:

[0192]

[0193]

[0194] in and For users The ratio of spreading false and true topics among friends. and The topic influence is calculated by considering the psychological benefits users experience from related fake and real topics in their historical participation. Finally, using evolutionary game theory and taking into account the impact of user psychological benefits, the final mutual influence between fake and real topics is constructed as follows:

[0195]

[0196]

[0197] in, and These represent the effects of fake and real topics on users after a game of negotiation. The impact of communication activities.

[0198] Finally, based on the mutual influence of the game between false and real topics under the influence of users' psychological gains, a structure was constructed. t The Interplay Matrix of Fake and Real Topics in the Moment :

[0199]

[0200] in, .

[0201] In step S5 of this embodiment, PI-GCN is used as the prediction model of this invention, and GCN graph convolutional neural network is used to process social networks, which are typical non-Euclidean structure data. Considering the promotion and inhibition relationships between fake and real topics in the spread of information, a prediction model for the behavior of fake and real information dissemination groups based on PI-GCN is proposed. The goal of the prediction task in this embodiment is to predict the participation of potential user nodes in the fake topic. If they participate, it is determined whether they forward the fake topic or the real topic, thus transforming it into a three-class classification task. The model input for this embodiment is as follows:

[0202] 1. Feature matrix X=N×A ,in N This indicates the number of user nodes in the network spreading false information. A It is the topic feature vector of each user node, which includes the user's own attribute features and the features of topic driving factors.

[0203] 2. Adjacency Matrix in Game Theory ,express t At any given moment, this refers to the relationship connections between all users in both the fake and real topic spaces, influenced by the user's psychological benefit.

[0204] In this embodiment, a two-layer graph convolutional neural network with an intermediate Dropout layer is used as a model for predicting the forwarding of online rumors. First, the weights and biases are randomly initialized. Then, X is multiplied by W, the bias is added, and then multiplied by Aˆ. Next, the ReLU function is used as the activation function for this layer, and Dropout is performed during model training. Finally, this embodiment uses the SoftMax activation function to represent the convolutional output as the probability value of different node categories. The specific formula is expressed as follows:

[0205]

[0206] in, Let be the weight matrix corresponding to the i-th layer in the graph convolutional network. .

[0207] Since this system model solves a three-class classification prediction problem, the model output is set to... Make the model output Perform a three-class classification prediction based on the model output, including potential users. We will forward fake topics and potential users in the next time period. We will forward real topics and potential users in the next time period. We will not participate in the dissemination of false or true topics during the next time period.

[0208] In this embodiment, the model's input includes topic information data, User relationship network within the space of constant false and real topics User's own attribute set User historical behavior collection The model constructs a matrix of interactions between false and real influences using evolutionary game theory and user psychological gains, and then uses a GCN graph convolutional neural network to predict... At any given moment, this table shows the forwarding behavior of potential users towards both deceptive and genuinely guiding topics. The specific propagation model algorithm is shown in Table 1.

[0209] Table 1

[0210]

[0211] Furthermore, this embodiment analyzes the time complexity of this algorithm. The overall time complexity of the user psychological benefit calculation method composed of Simcse and entropy weight method is... , M The number of topics forwarded by each user throughout their history. The time complexity of the user feature data augmentation model based on generative adversarial networks is O(n). , K The network depth is generally much smaller than the number of samples. N。 The algorithm also calculates user-specific factors and topic-driven factors. The time complexity of the PI-GCN prediction model is The time complexity of the entire algorithm model is O(n). Since only a small portion of the data in the fake and real topics section is actually valid, the input data for the PI-GCN model is not large. Furthermore, the model employs two layers of graph convolution, ensuring that the amount of data used in training remains within a predictable range. Therefore, the final time complexity of the entire algorithm is O(n log n). And it is acceptable.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A propagation prediction method based on user psychological gains and adversarial generative networks for the game of deceptive versus real guiding topics, characterized by: The method includes the following steps: S1. Define relevant parameters in the process of online topic propagation and formalize the propagation prediction problem; formalize the propagation prediction problem as follows: in, The current user propagation network for both real and false topics is a directed graph consisting of user nodes U and user relationship edges E. Represents the set of all users' own attributes, where Represents the user's basic attribute set; Indicates user exist Take action at all times ,in This represents the set of users' historical behaviors during the spread of the topic; This indicates a method for predicting user participation in forwarding topics. This indicates the predicted engagement behavior of potential users in both deceptive and genuinely suggestive topics. S2. Construct a space for false and true guiding topic data based on users' basic information, user relationship networks, user historical behavior, and guiding topic data; S3. Quantify users' psychological benefits using Simcse and entropy weighting, and enhance user topic feature data using adversarial generative networks; S4. Quantify the adversarial game relationship between false and real topics through evolutionary game theory, and integrate the mutual influence of false and real topics into the adjacency matrix of user relationship network; S5. Combining the enhanced user topic feature data and the adjacency matrix containing mutual influence information, the PI-GCN model is used to predict user behavior and predict the future propagation trend of guiding topics. In step S5, PI-GCN is used as the prediction model, and the input to the prediction model is: Feature matrix X=N×A ,in N This indicates the number of user nodes in the network spreading false information. A It is the topic feature vector of each user node, which includes the user's own attribute features and topic driving factor features; Adjacency Matrix in Game Theory ,express t At any given moment, the relationship connection information between all users in the two topic spaces of fake topics and real topics, under the influence of users' psychological benefits; First, a two-layer graph convolutional neural network with an intermediate Dropout layer is used as the model for predicting online rumors. First, the weights and biases are randomly initialized. Then, X is multiplied by W, the bias is added, and then multiplied by Aˆ. Next, the ReLU function is used as the activation function for this layer, and Dropout is performed during model training. Finally, the SoftMax activation function is used to represent the convolutional output as the probability values ​​of different node categories; expressed as: in, Let be the weight matrix corresponding to the i-th layer in the graph convolutional network. ; Let the model output Perform a three-class classification prediction based on the model output, including potential users. We will forward fake topics and potential users in the next time period. We will forward real topics and potential users in the next time period. We will not participate in the dissemination of false or true topics during the next time period.

2. The propagation prediction method for a false and real guided topic game based on user psychological gains and adversarial generative networks as described in claim 1, characterized in that: In step S1, the following parameters are defined: Define users who participate in the topic With topic dissemination network ,in, This represents the set of users participating in the topic at time t. This represents the set of directed edges connecting users who participated in the topic. Indicates user Follow users ; Define the user's own basic attributes : in Indicates user age, Indicates user gender, Indicates user The number of followers Indicates user The number of users who follow; Define user activity : in, Indicates user The number of original topics published within a certain period of time. Indicates user The number of times someone else's topic is forwarded within a certain period of time. , This represents the weights of the two factors; Define user psychological benefits : in, For users' historical topics Topics of Falsehood and Reality Similarity between them Indicate topic The overall popularity is calculated based on three user behavior dimensions: likes, comments, and shares; M represents the number of topics a user has discussed in their past posts. Define user intimacy : in, This is an indicator function that indicates to potential users. Do you pay attention to neighboring users? K represents user The total number of topics of interaction between them; This represents the user interaction behavior coefficient, which is determined based on different interaction behaviors. Let t be the time decay function, representing the current time. Indicates user For users The time of the a-th action in the k-th topic; Defining the popularity of a topic : in This indicates the number of times the current topic has been forwarded within the first preset dissemination time after its publication. This indicates the number of times the current topic has been forwarded within the second preset dissemination period after its publication. This indicates the total number of times the current topic has been forwarded.

3. The propagation prediction method for a false and real guided topic game based on user psychological gains and adversarial generative networks as described in claim 2, characterized in that: In step S2, the false and true guiding topic data space is based on user basic information, user relationship network, user historical behavior and guiding topic data, and calculates user basic characteristics, user relationship adjacency matrix, user self factors and topic driving factors.

4. The propagation prediction method for a false and real guided topic game based on user psychological gains and adversarial generative networks as described in claim 3, characterized in that: Step S3 specifically includes the following steps: S31. Obtain word embedding representations based on contrastive learning through the Simcse model, obtain the corresponding topic representation vectors, and calculate the cosine similarity between the historical topic representation vectors and the current false and true topic representation vectors. S32. Calculate the weights of the topic popularity index using the entropy weight method; S33. Calculate the user's psychological benefit based on the weights of the calculated text similarity and topic popularity indicators.

5. The propagation prediction method for a false and real guided topic game based on user psychological gains and adversarial generative networks as described in claim 4, characterized in that: In step S31, positive and negative sample groups (X, X+, X-) are first constructed, where X is the original sample, X+ is the positive sample generated using X, and X- is the negative sample that is not related to X. The contrast loss is optimized by reducing the distance between X and X+ and increasing the distance between X and X- in the feature space. The InfoNCE loss function is used as the optimization loss function, and it is expressed as follows: in, , and These are the normalized representation vectors of the original sample, positive sample, and negative sample, respectively. This refers to temperature hyperparameters. Historical topics derived from the Simcse model a The representation vector is A ( , ,..., ,..., ), Fake vs. Real Topics b for B ( , ,..., ,..., If the cosine similarity is calculated as follows: The range of cosine values ​​is [-1, 1], where -1 means completely dissimilar and 1 means completely similar. The closer to 1, the more similar the two values ​​are, and the closer to -1, the less similar the two values ​​are.

6. The propagation prediction method for a false and real guided topic game based on user psychological gains and adversarial generative networks as described in claim 5, characterized in that: In step S32, the objective weights of comments, likes, and reposts for each topic are calculated using the entropy weight method, which includes: S321. Standardize the original data using the min-max standardization method. The standardized data is represented as follows: In the formula, This represents the j-th sample for the i-th indicator. and Let represent the minimum and maximum values ​​of the i-th indicator in the sample, respectively; S322. Calculate the proportion of likes, comments, and shares for each topic in all samples: Where m is the number of samples; S323. Calculate the entropy value of each indicator: in, It is a constant used to normalize the entropy value; S324. Calculate the weight of each indicator using entropy values: Where n is the number of indicators.

7. The propagation prediction method for a false and real guided topic game based on user psychological gains and adversarial generative networks as described in claim 6, characterized in that: In step S33, combining the calculated text similarity and topic popularity index weights, the user's psychological benefit from historical topics on the current spread of false and true topics is calculated as follows: in M For the number of user's historical topics, N The number of historical hot topics is an indicator. and These are the spurious leading topic vector and the real leading topic vector calculated by Simcse, respectively.

8. The propagation prediction method for a false and true guided topic game based on user psychological gains and adversarial generative networks as described in claim 7, characterized in that: Step S4 includes: S41. By introducing adversarial generative networks, we can mine the potential relationships generated by the adversarial interaction between fake and real topics in the data, and calculate the topic influence of fake or real topics. Furthermore, it quantifies the mutual influence between false and true topics through evolutionary game theory: in, The payoff functions for the two game strategies, "forwarding false information" and "forwarding true information," are given respectively. and These represent the effects of fake and real topics on users after a game of negotiation. The impact of communication activities; S42. Based on the mutual influence of false and real topic games under the influence of user psychological benefits, construct... t The Interplay Matrix of Fake and Real Topics in the Moment : in, .

9. The propagation prediction method for a false and real guided topic game based on user psychological gains and adversarial generative networks as described in claim 8, characterized in that: In step S41, the adversarial generative network includes a generative model G and a discriminative model D. The goal of the generative model G is to deceive the discriminative model D, and the goal of the discriminative model D is to distinguish between the data generated by the generative model G and the real data. The two constitute a dynamic game process, ultimately aiming to reach a Nash equilibrium. The real topic feature dataset is represented as datas[x1,x2,...,xn], assuming that the topic feature sequence follows a certain distribution. The method for establishing the generative model G is described as finding the maximum likelihood of the generative model, i.e.: The iterative process of generating and discriminating topic feature sequences is described as follows: Let Gz represent the topic sample generation model, and z represent the data after random sampling of the original topic feature sequence. The model G generates topic feature data from the randomly sampled data z. D is a topic feature sequence discrimination model. For any input feature sequence x, the discrimination model outputs a real number between 0 and 1, which represents the probability that the set of feature sequences comes from real collected sample data. and Let represent the distributions of real topic data and generated topic data, respectively. Then, the objective function of the discriminant model D is: The optimization function of the entire model is expressed as: The entire optimization process can be represented as iterating over D and G until the entire process converges, which can be expressed as: wait infinitely close to ; The influence of fake or real topics It consists of user-specific factors and topic-driven factors. User-specific factors include: the user's basic attributes and user activity level, namely: The driving factors for a topic include: user intimacy and basic topic attributes, namely: Taking into account both user-specific factors and topic-driven factors, the following functions are constructed using a multiple linear regression algorithm to determine the influence of false and true topics: in, , , These are the partial regression coefficients obtained using a multiple linear regression algorithm. , This indicates the proportion of each factor in the topic's influence; Based on game theory, two game strategies are defined: "forwarding false information" and "forwarding true information," with payoff functions for each strategy as follows: in and For users The ratio of friends spreading fake and real topics; and The topic influence is determined by the psychological benefits users experience from participating in historical topics, as well as the impact of related fake and real topics. Finally, using evolutionary game theory to calculate the formula, and taking into account the impact of users' psychological gains, the final interaction between false and real topics is constructed.

Images