A key-influence user discovery method for time-varying network

Abstract

The application belongs to the field of social network derived topic analysis, and particularly relates to a key influential user discovery method for time-varying network, which comprises the following steps: preprocessing original topic network data to obtain topic network data; slicing the topic network data along the time dimension to obtain topic network data at different time points; calculating the dynamic importance of user nodes at different time points, and obtaining the transfer sequence of each user node according to the dynamic importance; constructing the feature vector of each user node by using an attention mechanism; obtaining the cumulative interaction sequence and the recent interaction sequence of each user node, and obtaining the interest vector of each user node through a double-Transformer structure; splicing the feature vector and the interest vector of each user node to obtain comprehensive features; inputting all the comprehensive features into a full connection layer and a softmax function layer to obtain the key influential user identification result; and the application can effectively discover key influential elements in the topic space.

Description

Technical Field

[0001] This invention belongs to the field of social network derivative topic analysis, specifically involving a method for discovering key influential users in time-varying networks. Background Technology

[0002] With the rapid development of social networks, people widely participate in discussions and dissemination of various topics, leading to the rapid spread of derivative topics and the formation of influential public opinion hotspots. During this process, some key users leverage platform mechanisms and their own influence to drive the spread of these topics, attempting to guide public opinion, influence public thinking, and even achieve specific objectives. Therefore, timely identification of these key influential users in the early stages of a topic's emergence is crucial for understanding the mechanisms of topic dissemination, guiding public opinion, and effectively implementing public opinion management. In recent years, scholars have found that identifying and tracking core users with guiding capabilities has become an important issue in areas such as influence assessment, public opinion governance, and target group positioning. Identifying potential core users in the network of guiding topic dissemination can not only help relevant institutions more accurately grasp public opinion dynamics but also effectively prevent and respond to the potential spread of negative information. Therefore, scholars believe that identifying and managing core influential users in guiding topics has profound significance for promoting a healthy online ecosystem.

[0003] Currently, research on identifying key influential users in the early stages of social network-related topic outbreaks mainly focuses on three aspects: social network topic content analysis, network structure analysis, and node influence quantification. Research on topic content analysis primarily includes the mining of sentiment and content characteristics. Research on network structure analysis mainly focuses on the group structure and interaction patterns within social networks, revealing the propagation paths of information and the core dissemination groups. Regarding node influence quantification, researchers assess node influence and predict their key roles in dissemination by quantifying indicators such as node centrality.

[0004] In summary, although the academic community has produced a wealth of research findings on trending topics, the identification of key influential users in the early stages of derivative topics still faces the following challenges:

[0005] 1. The dynamic complexity of topic space network structure. The interaction relationships and topic structure among users in social networks change constantly over time. How to effectively mine and quantify these multi-dimensional features and represent them as node vectors is a challenge.

[0006] 2. The variability of topic content. Topic content changes constantly over time, making direct analysis and modeling difficult. Effectively capturing the complex relationships and similarities between topics, and combining this with in-depth analysis of user behavior over time, is a challenge. Summary of the Invention

[0007] To address the above problems, this invention provides a method for discovering key influential users in time-varying networks, comprising the following steps:

[0008] S1. Obtain raw topic network data from social platforms and preprocess it to obtain topic network data;

[0009] S2. Slice the topic network data along the time dimension to obtain topic network data at different time points;

[0010] S3. For the topic network data at each time point, calculate the dynamic importance of user nodes, calculate the transition probability of random walk based on the dynamic importance, and obtain the transition sequence of each user node.

[0011] S4. Based on the transition sequence at each time point, an attention mechanism is used to construct the feature vector of each user node;

[0012] S5. Obtain the cumulative interaction sequence and the most recent interaction sequence for each user node based on the topic network data;

[0013] S6. Based on the cumulative interaction sequence and the most recent interaction sequence, obtain the interest vector of each user node through a dual Transformer structure;

[0014] S7. For each user node, the feature vector and interest vector are concatenated to obtain a comprehensive feature; the comprehensive features of all user nodes are input into a fully connected layer and a softmax function layer to obtain the key influence user identification results.

[0015] The beneficial effects of this invention are:

[0016] This invention starts with the dynamic characteristics of topic networks, analyzes user behavior over time, combines hidden user features at different time scales, and integrates them with comprehensive user interest features to obtain the final judgment result of influential users in the topic network. Compared with existing technologies, this invention considers time sensitivity and historical influence factors, and combines them with comprehensive user interests to consider the comprehensive impact of user behavior from a time-dimensional perspective, thus improving the accuracy of the prediction results. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention;

[0018] Figure 2 This is a schematic diagram illustrating the acquisition of user node feature vectors according to the present invention. Detailed Implementation

[0019] 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.

[0020] This invention provides a method for discovering key influential users in time-varying networks, such as... Figure 1 , Figure 2 As shown, it includes the following steps:

[0021] S1. Obtain raw topic network data from social platforms and preprocess it to obtain topic network data.

[0022] Specifically, relevant raw topic network data is obtained through social media APIs and public dataset platforms, including basic user information data, historical user behavior data, and relevant data throughout the topic's lifecycle. Basic user information data includes personal information such as user ID and gender, while historical user behavior data includes interaction data between users and other users or topics, including actions such as liking and sharing.

[0023] The raw topic network data is preprocessed, including removing some invalid and duplicate data, as well as simple cleaning, to transform the unstructured raw data into structured data.

[0024] S2. Slice the topic network data along the time dimension to obtain topic network data at different time points.

[0025] S3. For the topic network data at each time point, calculate the dynamic importance of user nodes, calculate the transition probability of random walk based on the dynamic importance, and obtain the transition sequence of each user node.

[0026] Specifically, this invention uses a user as a user node. The user relationship network topology constructed based on topic network data at different time points differs, and the importance of user nodes changes over time. This invention defines user node i at time point n t. n The dynamic importance of W i (t n ), represented as

[0027] W i (t n )=βS i (T)+(1-β)H i (t n )

[0028] Where β∈[0,1] represents the balance coefficient, used to balance the weights of short-term time sensitivity and cumulative historical influence in the dynamic importance calculation. T represents a fixed-length time period used to statistically analyze the interaction behavior of user nodes within this time range; specifically, it calculates the interaction behavior of user node i at time point n t. n The dynamic importance of W i (t n When ), statistics are performed within the time period [t]. n ,t n The interaction behavior of user node i within [+T]. i (T) represents the time sensitivity, indicating the relative connection frequency of user node i within the time window T. i (t n The cumulative historical influence represents the influence of user node i at the current time t. n The long-term influence is determined by combining interactive behaviors over the past m time points; among which

[0029]

[0030] I i (T) represents the frequency of interaction between user node i and other user nodes within the time window T. This represents the total frequency of all user interactions within the time window T. α is a decay factor used to control the impact of historical behavior on the current dynamic importance; the larger the time interval, the smaller the weight. i (t k ) represents user node i at time point t k Frequency of interaction with other user nodes.

[0031] By adjusting β and α, the importance of user nodes in dynamic networks can be better reflected.

[0032] Specifically, the transition probability of a random walk, calculated based on dynamic importance, is expressed as follows:

[0033]

[0034] Where P(j,i) represents the transition probability from user node i to user node j, ω(i,j) represents the edge weight between user node i and user node j, indicating the degree of correlation between the two nodes, and E represents the set of relationship edges among users participating in the topic. W j This represents the dynamic importance of user node j. Z is the normalization coefficient. p,q (i,j) represents the walk bias term, which is based on the shortest distance d from user node i to user node j. i,j The possible values ​​are [0, 1, 2], representing returning to the previous node, the direct neighbor, and the furthest node, respectively. That is, p and q control the walk bias, including:

[0035]

[0036] S4. Based on the transition sequence at each time point, an attention mechanism is used to construct the feature vector of each user node.

[0037] Specifically, step S4 constructs a vector representation of each user node based on the transition sequence at each time point using an attention mechanism, such as... Figure 2 As shown, it includes:

[0038] S41. Embed the transition sequence of each user node at each time point to obtain the corresponding structural features;

[0039] S42. Perform position encoding on each time point to obtain a time-position embedding vector;

[0040] S43. Based on the structural features and temporal embedding vectors, obtain the input vectors of each user node at different time points, represented as follows:

[0041] x i (t n =concat(h) i (t n ),PE(t n ))

[0042] Where concat(·) represents the vector concatenation operation, x i (t n ) represents user node i at time point t n The input vector, h i (t n ) represents user node i at time point t n Structural characteristics, PE(t) n ) represents time point t n The time position embedding vector;

[0043] S44. Calculate the weighted vector of each user node at different time points based on the input vector, denoted as:

[0044]

[0045]

[0046] in, This indicates that user node i is at time point t. n The input vector for user node i at time t k Attention weights of the input vector, Q i (t n ) indicates that based on user node i at time point tn The query vector is calculated from the input vector, K. i (t k ) indicates that based on user node i at time point t k The key vector calculated from the input vector, d k V represents the dimension of the key vector; i (t k ) indicates that based on user node i at time point t k The input vector is used to calculate the value vector, where N represents the total number of time points, and z i (t n ) represents user node i at time point t n The weighted vector;

[0047] S45. Input the weighted vector of each user node at all time points into the feedforward neural network to obtain the feature vector of the user node.

[0048] S5. Obtain the cumulative interaction sequence and the most recent interaction sequence for each user node based on the topic network data.

[0049] Specifically, the cumulative interaction sequence and the most recent interaction sequence are the interaction behavior data of user nodes with other user nodes in different time periods, such as likes, reposts, comments, etc.

[0050] S6. Based on the cumulative interaction sequence and the most recent interaction sequence, obtain the interest vector of each user node through a dual Transformer structure.

[0051] Specifically, the dual Transformer structure includes a shallow Transformer and a deep Transformer.

[0052] Specifically, the shallow Transformer is used to capture the main points of interest in the user's recent interactions. Shallow networks are computationally efficient and can quickly extract the main features of short-term behavior. A 2- to 4-layer encoder can better balance model complexity and the need for capturing short-term interests; in this embodiment, the shallow Transformer uses a 2-layer encoder. The deep Transformer is used for in-depth feature extraction from the user's historical cumulative interactions; more layers enable the model to capture complex long-term behavioral patterns and global semantic features; therefore, in this embodiment, the deep Transformer uses a 6-layer encoder. Furthermore, to enable the shallow and deep Transformers to work collaboratively, they share some parameters.

[0053] Step S6, which involves obtaining the interest vector of any user node using a dual Transformer structure, includes:

[0054] S61. Input the cumulative interaction sequence into a deep Transformer and output the user's long-term preferences, which can be represented as follows:

[0055]

[0056] in, This represents the long-term user preferences of user node i. Θ represents the cumulative interaction sequence of user node i. deep T represents the parameters of the deep Transformer. deep This indicates a deep Transformer.

[0057] S62. Input the most recent interaction sequence into a shallow Transformer, and output the user's short-term preferences, which can be represented as follows:

[0058]

[0059] in, This represents the short-term preferences of user node i. Θ represents the most recent interaction sequence of user node i. shallow T represents the parameters of the shallow Transformer. shallow This indicates a shallow Transformer.

[0060] S63. The interest vector is obtained by fusing long-term and short-term user preferences, which can be represented as follows:

[0061]

[0062] Here, pi represents the interest vector of user node i, and γ represents a weighting parameter used to balance long-term and short-term interests. This combination balances the weights of short-term interests and long-term preferences, ensuring that the user's interest vector can dynamically reflect recent changes while also demonstrating long-term stability.

[0063] S7. For each user node, the feature vector and interest vector are concatenated to obtain a comprehensive feature; the comprehensive features of all user nodes are input into a fully connected layer and a softmax function layer to obtain the key influence user identification results.

[0064] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "setting," "connection," "fixing," "rotation," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal connection of two components or the interaction between two components. Unless otherwise explicitly limited, those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for discovering key influential users in time-varying networks, characterized in that, Includes the following steps: S1. Obtain raw topic network data from social platforms and preprocess it to obtain topic network data; S2. Slice the topic network data along the time dimension to obtain topic network data at different time points; S3. For the topic network data at each time point, calculate the dynamic importance of user nodes, calculate the transition probability of random walk based on the dynamic importance, and obtain the transition sequence of each user node. S4. Based on the transition sequence at each time point, an attention mechanism is used to construct the feature vector for each user node, including: S41. Embed the transition sequence of each user node at each time point to obtain the corresponding structural features; S42. Perform position encoding on each time point to obtain a time-position embedding vector; S43. Based on the structural features and temporal embedding vectors, obtain the input vectors of each user node at different time points, represented as follows: x i (t n )=concat(h i (t n ),PE(t n )) Where concat(·) represents the vector concatenation operation, x i (t n ) represents user node i at time point t n The input vector, h i (t n ) represents user node i at time point t n Structural characteristics, PE(t) n ) represents time point t n The time position embedding vector; S44. Calculate the weighted vector of each user node at different time points based on the input vector; S45. Input the weighted vector of each user node at all time points into the feedforward neural network to obtain the feature vector of the user node; Step S44 calculates the weighted vector of each user node at different time points based on the input vector, denoted as: in, This indicates that user node i is at time point t. n The input vector for user node i at time t k Attention weights of the input vector, Q i (t n ) indicates that based on user node i at time point t n The query vector is calculated from the input vector, K. i (t k ) indicates that based on user node i at time point t k The key vector calculated from the input vector, d k V represents the dimension of the key vector; i (t k ) indicates that based on user node i at time point t k The input vector is used to calculate the value vector, where N represents the total number of time points, and z i (t n ) represents user node i at time point t n The weighted vector; S5. Obtain the cumulative interaction sequence and the most recent interaction sequence for each user node based on the topic network data; S6. Based on the cumulative interaction sequence and the most recent interaction sequence, obtain the interest vector of each user node through a dual Transformer structure; S7. For each user node, the feature vector and interest vector are concatenated to obtain a comprehensive feature; the comprehensive features of all user nodes are input into a fully connected layer and a softmax function layer to obtain the key influence user identification results.

2. The method for discovering key influential users in time-varying networks according to claim 1, characterized in that, User node i at time point n t n The dynamic importance of W i (t n ), represented as W i (t n )=βS i (T)+(1-β)H i (t n ) Where β∈[0,1] represents the balance coefficient; T represents a fixed-length time period; S i (T) represents the relative connection frequency of user node i within the time window T; H i (t n ) indicates that user node i is at the current time point t n Its long-term influence; among which I i (T) represents the frequency of interaction between user node i and other user nodes within the time window T, where α is the decay factor; i (t k ) represents user node i at time point t k Frequency of interaction with other user nodes.

3. The method for discovering key influential users in time-varying networks according to claim 1, characterized in that, The transition probability of a random walk, calculated based on dynamic importance, is expressed as follows: Where P(j,i) represents the transition probability from user node i to user node j, ω(i,j) represents the edge weight between user node i and user node j, and E represents the set of relationship edges among users participating in the topic; W j Z represents the dynamic importance of user node j, and Z is the normalization coefficient; b p,q (i,j) represents the wandering bias term, expressed as: Where, d i,j represents the shortest distance from user node i to user node j, and p and q are the control walk bias factors.

4. The method for discovering key influential users in time-varying networks according to claim 1, characterized in that, The dual Transformer structure includes a shallow Transformer and a deep Transformer. The shallow Transformer uses a 2-layer encoder, and the deep Transformer uses a 6-layer encoder.

5. The method for discovering key influential users in time-varying networks according to claim 4, characterized in that, Step S6, which involves obtaining the interest vector of any user node using a dual Transformer structure, includes: S61. Input the cumulative interaction sequence into a deep Transformer and output the user's long-term preferences; S62. Input the recent interaction sequence into a shallow Transformer and output the user's short-term preferences; S63. Combine long-term user preferences and short-term user preferences to obtain an interest vector.

Images