A token-based influence user selection method under a dynamic social network

Abstract

The application provides a token-based influence user selection method under a dynamic social network, which comprises the following steps: 1, obtaining the information of users in the social network and the relationship information among the users, generating the closeness among the users as the weight of edges in the network, and building the whole social network model; 2, each user in the social network generates a certain number of tokens according to the original influence; 3, through the iteration mechanism, the tokens are distributed to the neighbor users according to the network node relationship and the self-influence rule after being continuously generated in each round; 4, the number of tokens absorbed by the social network user represents the influence of the user; 5, the suitable z seed set is selected through the final convergence judgment; and 6, the effect of the result seed set is verified through the IC diffusion model. The application uses the token as the medium, solves the influence maximization problem of the social network from the distributed perspective, proposes a user selection model, and is suitable for large-scale and dynamic network scenes, and has a small cost.

Description

Technical Field

[0001] This invention belongs to the field of social networks and influence maximization, specifically relating to a token-based influence user selection method in dynamic social networks. Background Technology

[0002] Over the past two decades, with the development of the "word-of-mouth" effect, maximizing influence has become one of the most popular research topics in the field of social network research. This problem involves selecting a group of influential users, or "seeds," who then spread their influence to other users within the social network. With the support of these influential users, businesses can be developed to attract more customers and rapidly gain high visibility and significant profits within the social network.

[0003] To address the problem of maximizing influence, numerous user selection models have been proposed. Current user selection models can be categorized into centralized and distributed types, with centralized models being the mainstream. In centralized models, a central controller provides a global overview of the social network, collecting user status and relationship information, and selecting influential users based on this information. Some researchers have used Monte Carlo methods to select nodes with the highest marginal returns as influential nodes; others have employed degree-discount heuristics to select influential users, iterating through the process of selecting influential users based on the decreasing neighboring user relationships of a selected user; still others have proposed using ensemble PageRank algorithms to solve the influence maximization problem. Over the past two decades, with the development of the internet and social media, the user base of social networks has grown exponentially, and the status and relationships of users within these networks are dynamically changing. In this context, centralized user selection models face two main limitations. On the one hand, as the user base grows, the network communication overhead of the central controller in collecting user information in social networks is increasing significantly; on the other hand, many centralized user selection models are based on computationally expensive mechanisms, and the dynamic changes of users in social networks will lead to constant changes in their social influence, which requires the central controller to reselect influential users, thus significantly increasing the computational overhead of the central controller.

[0004] To overcome the limitations of centralized user selection models, distributed user selection models have been proposed in recent years. For example, a distributed influence maximization model, SIMiner, has been proposed, which uses a mobile agent to simulate ants based on the ant colony algorithm. In SIMiner, ants walk through users in the social network and select influential users by distributing pheromones. While SIMiner overcomes the high communication and computational overhead of centralized user selection, it also has two limitations. First, SIMiner introduces external mobile agents, i.e., "ants," to select influential users in the social network. Therefore, SIMiner has high model overhead. Second, the "ants" in SIMiner are based on a random walk strategy, so many users cannot be acquired and updated in a timely manner within a social network. Therefore, SIMiner has a relatively low convergence speed. Summary of the Invention

[0005] To address the aforementioned unresolved key issues, this invention proposes a token-based method for selecting influential users in dynamic social networks. By using tokens as a medium for user influence in social networks, the aim is to solve the problem of maximizing influence in complex and dynamic social networks from a distributed perspective, without relying on external proxies, while simultaneously improving the model's convergence speed and the coverage of user influence diffusion.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A token-based influence user selection method in a dynamic social network includes the following steps:

[0008] Step 1: Propose a token-based dynamic social network user selection model, TBIUS, which uses tokens as a medium and obtains a seed set through an iterative mechanism.

[0009] Step 2: Construct a social network graph G from the existing real dataset. This graph describes in detail the users' personal information, unique identifiers, and the strength and intimacy of the relationships between users, providing a foundation for subsequent analysis.

[0010] Step 3: Determine the generation mechanism of user tokens and mark their source to make each token identifiable. This process ensures that the token accurately reflects its source and propagation path during distribution.

[0011] Step 4: Develop an iterative token distribution strategy. Combining the user's own information, neighbor information, and the current set of tokens held, in each iteration, the user distributes their own tokens to their neighboring users according to the strategy, simulating the information flow process.

[0012] Step 5: After each iteration, filter the tokens held by the user, remove redundant information, form an intermediate result view, and gradually approach the target seed set.

[0013] Step 6, use J accar The d and Dice similarity coefficients are used to perform convergence analysis on the seed set to ensure that the final seed set has propagation potential when it meets the preset convergence criteria.

[0014] Step 7: Conduct an influence diffusion test on the final seed set using the IC independent cascade diffusion model. By activating other nodes in the network, compare the diffusion coverage at the same time scale to evaluate and optimize the propagation efficiency.

[0015] As a preferred embodiment, the steps for constructing the social network G in step 2 are as follows:

[0016] Step 2-1: A social network consists of many users with different relationships. A social network is represented using a graph structure as described below.

[0017] G = {V, E}.

[0018] Where V = {v1, v2, ... v} n} is a user set, v i For the i-th user in the social network, E = {e 12 , ..., e (n-1)n} is the edge set of users in the network, and edge e ij Corresponding to v i and v j The relationship between them is initially defined by the number of connections between them, representing the strength and intimacy of the relationship between the users, and e i，j =e j，i That is, an undirected graph.

[0019] Step 2-2: For the edge set E = {e 12 , ..., e (n-1)n Perform a normalization operation on e. ij Values ​​are mapped to (0, 1], where 0 and 1 represent the furthest and closest relationships between two users, respectively. The normalization formula is as follows:

[0020]

[0021] Steps 2-3: The social network G is defined as follows:

[0022] v i ={id, inf i LE i Lv i T i}

[0023] Where id is user v i The identifier, inf i It is v i Self-influence It is v i The adjacent edge set, It is v i A set of directly connected neighboring users, v j It is v i A user within a neighboring user set. T i It is v i A token pool containing v i All tokens owned, T i The defined formula is: T i ={t1, t2, ... t n}, t i Indicates v i A token owned is represented as: t i ={s id ,l},s id The node that generated the token is represented by , and l represents the path length of the token.

[0024] Preferably, the token generation mechanism in step 3 includes:

[0025] Step 3-1: In each iteration, all users generate a certain number of tokens in the social network G. This user v i The number of tokens generated in the iteration depends primarily on v. i Self-influence, i.e., INF i Since the token is the smallest indivisible unit representing the following effects in the model, the token generation coefficient tgc is introduced into the token generation calculation during subsequent token distribution, where tgc is an integer greater than 1. Finally, the number of tokens GT generated in the iteration is... i The calculation is as follows:

[0026] ΔT i =inf i ×tgc

[0027] Where tgc is the token generation coefficient, ranging from 1 to 30, with experiments conducted in steps of 1. A suitable value is fitted based on the convergence speed of the seed set obtained from the results; faster convergence speed results in better tgc. i It is v i The self-influence of the token means that after the token is generated, the newly generated token will be added to the user's token pool, i.e., T. i =T i +ΔT i .

[0028] As a preferred approach, the token iterative distribution strategy in step 4 is as follows:

[0029] Step 4-1: v i calculate The influence ratio of each user node, as one of the standards for token distribution, is calculated as follows:

[0030]

[0031] Among them, inf j User v i Neighbor v j The influence User v i The sum of the influence of neighbors.

[0032] Step 4-2: Through node v i Token pool T i The composition of the data is used to calculate the neighbor nodes from user v in one iteration. i The number of tokens obtained, since tokens are differentiated, v j From v i The obtained token set is calculated as follows:

[0033]

[0034] Where, weight ji Represents node v j Occupy node v i The proportion of the sum of the influence of neighboring nodes, where n is the number of nodes in the social network G, and S is the weight of the sum of the influence of neighboring nodes. k Indicates v i The token pool from v k The set of tokens, e ij Indicates v j With v i The degree of intimacy between them, that is, the probability of transmission between them.

[0035] Step 4-3: Token Normalization. Due to the atomic nature of tokens, the number of tokens must be an integer during the transfer process. Therefore, the calculation result needs to be rounded down to maintain the integrity of the transfer process and the consistency of the model. Indicates will Round down to the nearest integer.

[0036] Preferably, step 5 involves filtering the tokens held by the user after each iteration to remove redundant information and form an intermediate result view, including:

[0037] Step 5-1: Calculate the evaluation function for node scores, as shown in the following formula:

[0038]

[0039] Where d(v) represents the types of tokens in the node's token pool, l k This indicates the v that generated the token. k Node and v i Node distance, t k This indicates that the token pool contains tokens from v. k The number of tokens generated by the user, α and β are weight parameters, obtained by fitting.

[0040] Step 5-2: Filter out tokens with fewer than 50 types in the token pool, calculate the user node evaluation function, and obtain the intermediate result S of the iteration.

[0041] As a preferred embodiment, the seed set convergence calculation steps in step 6 are as follows:

[0042] Step 6-1: Perform convergence analysis on the seed set using Jaccard and Dice similarity coefficients to ensure that the final seed set has propagation potential when it meets the preset convergence criteria. The formula is as follows:

[0043]

[0044] Where S represents the intermediate result of the seed set obtained in this iteration, and S′ represents the intermediate result of the seed set obtained in the previous iteration.

[0045] When the similarity coefficient between Jaccard and Dice is consistently below 0.05, the model is considered to have reached the convergence condition.

[0046] Preferably, step 7 involves using the IC independent cascade diffusion model to test the influence diffusion of the resulting seed set S obtained through iteration. It is assumed that at time t, node v... i There are x neighbor nodes a1, a2, ..., a3 that are activated at time t-1. n The probabilities of these nodes activating node v are respectively Then node v i The probability of being activated is:

[0047]

[0048] If node v i If it is activated, it indicates that v i When covered, the IC diffusion range increases.

[0049] Among the seed sets selected by different models, the IC diffusion model has the highest coverage, indicating that the effect of maximizing the influence selection is the best. The closer the seed set selected by this model is to the optimal seed set, the better the model's performance.

[0050] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0051] To address the problems of user selection models, such as centralized models (high network communication overhead for the central controller collecting user information in social networks as the user base grows), and distributed models (resulting in increased computational overhead due to the dynamic changes in users' social influence and the central controller reselecting influential users), this invention proposes a token-based decentralized user selection model, TBIUS. TBIUS uses tokens as the medium for user influence in social networks. Furthermore, the TBIUS model proposes an iterative mechanism where token generation and distribution are iteratively performed by users in the social network. After a certain number of iterations, the user with the highest score is selected as the influential user. The contributions of the TBIUS model are summarized as follows: 1. The proposed model is a distributed user selection model that allows users to select influential users in the social network using only their local view; 2. The proposed iterative mechanism enables users to generate and distribute tokens without involving external agents; 3. Experiments on real social networks—Facebook, Wikipedia voting, and email—demonstrate that the TBIUS model performs excellently in terms of convergence speed and the spread of selected influence. Attached Figure Description

[0052] Figure 1 This is a flowchart of the token-based influence user selection method in a dynamic social network according to the present invention;

[0053] Figure 2 This is a partial schematic diagram of social network modeling in this invention;

[0054] Figure 3 This is a partial schematic diagram of social network token generation in this invention;

[0055] Figure 4 This is a partial schematic diagram of social network token distribution in this invention;

[0056] Figure 5 This is a schematic diagram illustrating the convergence of the TBIUS model during iteration in this invention;

[0057] Figure 6 This is a map showing the extent of the TBIUS model's diffusion across the Facebook dataset in this invention.

[0058] Figure 7 This is a schematic diagram illustrating the TBIUS model in a dynamic network in this invention;

[0059] Figure 8This is a schematic diagram illustrating the effect of the TBIUS model in a dynamic network in this invention;

[0060] Figure 9 This is a comparison chart of the TBIUS model with other models in this invention; Detailed Implementation

[0061] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0062] like Figure 1 As shown, the technical route of this invention mainly includes six steps: constructing a social network model from user information and relationships between users in the social network; generating the corresponding initial token number for user nodes in the social network according to the generation formula; determining the token distribution mechanism; iterating multiple rounds of token generation and distribution until convergence is achieved; obtaining the result seed set; and verifying the IC diffusion model.

[0063] This example is based on the Email-Enron, Facebook, and Wiki Vote datasets. The Facebook dataset was collected from survey participants using the Facebook application. It contains 4039 nodes, 88234 edges, and 0 connected components. The Wiki Vote dataset is historical data extracted from all Wikipedia administrator elections and votes since January 3, 2008. It contains 7066 network nodes and 24 connected components. The Email-Enron dataset network nodes are email addresses, and the undirected edges represent emails sent between nodes. It contains 36692 nodes, 183831 edges, and 1065 connected components. This invention employs a token-based influence user selection method in dynamic social networks to solve the problem of maximizing social network influence. The following sections will introduce this example from four aspects: social network model establishment, iterative mechanism for token generation and distribution, diffusion testing of the IC independent cascade model, and dynamic network adaptability.

[0064] 1) Establishing a social network model

[0065] Step 1-1: A social network consists of many users with different relationships. A social network is represented using a graph structure as described below.

[0066] G = {V, E}.

[0067] Where V = {v1, v2, ... v} n} is a user set, v i For the i-th user in the social network, E = {e 12 , ..., e (n-1)n} is the edge set of users in the network, and edge e ij Corresponding to v i and v j The relationship between them is initially defined by the number of connections between them, representing the strength and intimacy of the relationship between the users, and e i，j =e j，i That is, an undirected graph.

[0068] Step 1-2: For the edge set E = {e 12 , ..., e (n-1)n Perform a normalization operation on e. ij Values ​​are mapped to (0, 1], where 0 and 1 represent the furthest and closest relationships between two users, respectively. The normalization formula is as follows:

[0069]

[0070] Steps 1-3: The definitions of social network G are as follows:

[0071] v i ={id, inf i LE i LV i T i}

[0072] Where id is user v i The identifier, inf i It is v i Self-influence It is v i The adjacent edge set, It is v i A set of directly connected neighboring users, v j It is v i A user within a neighboring user set. T i It is v i A token pool containing v i All tokens owned, T i The defined formula is: T i ={t1, t2, ... t n}, t i Indicates v i A token owned is represented as: t i ={s id ,l},s id The node that generated the token is represented by 'l', and 'l' represents the path length of the token.

[0073] In the embodiment, a partial schematic diagram of the initial modeling is shown below. Figure 2As shown, the edges between nodes have weights, representing the degree of closeness between them. In the model, these weights are related to the probability of successful information transfer between nodes and are mapped to (0, 1) through a normalization operation. Nodes themselves possess token and influence attributes, with the token count initialized to 0. The relationships between nodes are represented by undirected graphs. Instances Email-Enron, Facebook, and Wiki Vote are modeled as three undirected social network graphs.

[0074] 2) Iterative mechanism for token generation and distribution

[0075] Step 2-1: In each iteration, all users generate a certain number of tokens in the social network G. This user v i The number of tokens generated in the iteration depends primarily on v. i Self-influence, i.e., INF i Since the token is the smallest indivisible unit representing the following effects in the model, the token generation coefficient tgc is introduced into the token generation calculation during subsequent token distribution, where tgc is an integer greater than 1. Finally, the number of tokens GT generated in the iteration is... i The calculation is as follows:

[0076] ΔT i =inf i ×tgc

[0077] Where tgc is the token generation coefficient, ranging from 1 to 30, with experiments conducted in steps of 1. A suitable value is fitted based on the convergence speed of the seed set obtained from the results; faster convergence speed results in better tgc. i It is v i The self-influence of the token means that after the token is generated, the newly generated token will be added to the user's token pool, i.e., T. i =T i +ΔT i .

[0078] Table 1. Experimental parameter settings

[0079]

[0080]

[0081] Figure 3 The example demonstrates the generation of tokens using a partial implementation. Here, the TGC value is set to 10. A large number of comparative experiments will be conducted to determine the optimal TGC parameter value.

[0082] Step 2-2: v i calculate The influence ratio of each user node, as one of the standards for token distribution, is calculated as follows:

[0083]

[0084] Among them, inf j User v i Neighbor v j The influence User v i The sum of the influence of neighbors.

[0085] Steps 2-3: Through node v i Token pool T i The composition of the data is used to calculate the neighbor nodes from user v in one iteration. i The number of tokens obtained, since tokens are differentiated, v j From v i The obtained token set is calculated as follows:

[0086]

[0087] Where, weight ji Represents node v j Occupy node v i The proportion of the sum of the influence of neighboring nodes, where n is the number of nodes in the social network G, and S is the weight of the influence of neighboring nodes. k Indicates v i The token pool from v k The set of tokens, e ij Indicates v j With v i The degree of intimacy between them, that is, the probability of transmission between them.

[0088] Steps 2-4: Token Normalization. Due to the atomic nature of tokens, the number of tokens must be an integer during the transfer process. Therefore, the calculation results need to be rounded down to maintain the integrity of the transfer process and the consistency of the model. Indicates will Round down to the nearest integer.

[0089] Figure 4 This describes the token distribution after the first distribution in a partial embodiment, where v1 contains tokens from v1 and v3.

[0090] Step 2-5 calculates the evaluation function for node scores, using the following formula:

[0091]

[0092] Where d(v) represents the types of tokens in the node's token pool, l k This indicates the v that generated the token. k Node and v i Node distance, tk This indicates that the token pool contains tokens from v. k The number of tokens generated by the user, α and β are weight parameters, obtained by fitting.

[0093] Steps 2-6: Filter out tokens with fewer than 50 types in the token pool, calculate the user node evaluation function, and obtain the intermediate result S of the iteration.

[0094] Steps 2-7: Perform convergence analysis on the seed set using Jaccard and Dice similarity coefficients to ensure that the final seed set has propagation potential when it meets the preset convergence criteria. The formula is as follows:

[0095]

[0096] Where S represents the intermediate result of the seed set obtained in this iteration, and S′ represents the intermediate result of the seed set obtained in the previous iteration.

[0097] When the similarity coefficient between Jaccard and Dice is consistently below 0.05, the model is considered to have reached the convergence condition.

[0098] like Figure 5 As shown, when tgc is 10, the entire model converges in 15 rounds, and subsequent fluctuations are due to individual nodes with similar influence, which can be ignored.

[0099] 3) Diffusion test of IC independent cascade model

[0100] Step 3-1 uses the IC independent cascade diffusion model to test the influence diffusion of the final seed set. Assume that at time t, node v... i There are x neighbor nodes a1, a2, ..., a3 that are activated at time t-1. n These nodes activate node v i The probabilities are respectively Then node v i The probability of being activated is:

[0101]

[0102] If node v i If it is activated, it indicates that v i When covered, the IC diffusion range increases.

[0103] Among the seed sets selected by different models, the IC diffusion model has the highest coverage, indicating that the effect of maximizing the influence selection is the best. The closer the seed set selected by this model is to the optimal seed set, the better the model's performance.

[0104] like Figure 6As shown in the diagram, the TBIUS model spreads across the Facebook dataset. It reaches 4,000 user nodes in about 11 rounds, which is a very objective spreading effect, indicating that the seed set of influence selected by the TBIUS model has a large influence.

[0105] 4) Dynamic network adaptability

[0106] In real-world social networks, relationships between users are highly dynamic. In this embodiment, firstly, the social structure network is as follows: Figure 2 As shown. After 50 iterations, users v1 and v2 and their edges are deleted, as follows. Figure 7 As shown. At this point, the TBIUS model processes and analyzes the network, as follows: Figure 8 As shown, after 10 iterations, the relationship between the number of node tokens tends to stabilize, meaning that the changes in influence between nodes are complete and can be quickly captured by the tokens. This indicates that the TBIUS model can respond quickly to dynamic social networks.

[0107] The TBIUS model was compared with other relevant models to comprehensively evaluate the effectiveness of the model of this invention. The number of seeds was set to 10, 20, 30, 40, and 50. Based on the IC model, the number of other users in the social network influenced by the selected user was calculated as an indicator of influence diffusion coverage. Influence diffusion coverage experiments were all based on multiple Monte Carlo simulations. Figure 9 As shown, the X-axis represents the number of selected users, while the Y-axis represents the influence of the selected users on other users in the social network based on the IC model. The TBIUS model of this invention consistently outperforms other centralized and distributed model results.

[0108] Based on the above analysis, after a certain number of iterations, experiments on real social network datasets show that the proposed TBIUS model has good performance in terms of influence diffusion coverage and convergence, speed and model overhead, and is suitable for large-scale and dynamic social networks.

[0109] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art should understand that certain modifications and variations can be made to the present invention without departing from the spirit and substance of the present invention, but these modifications and variations should be covered within the scope of protection of the present invention.

Claims

1. A token-based influence user selection method in a dynamic social network, characterized in that, Includes the following steps: Step 1: Propose a token-based dynamic social network user selection model, TBIUS, which uses tokens as a medium and obtains a seed set through an iterative mechanism. Step 2: Construct a social network graph G from the existing real dataset. This graph describes in detail the users' personal information, unique identifiers, and the strength and intimacy of the relationships between users, providing a foundation for subsequent analysis. Step 3: Determine the generation mechanism of user tokens and mark their source to make each token identifiable; This process ensures that the token accurately reflects its origin and transmission path during its dissemination; The token generation mechanism includes: Step 3-1: In each iteration, all users generate a certain number of tokens in the social network G; this user v i The number of tokens generated in the iteration depends primarily on v. i Self-influence, i.e., INF i In the subsequent token distribution process, the token generation coefficient tgc is incorporated into the token generation calculation, where tgc is an integer greater than 1. Finally, the number of tokens generated in the iteration is calculated as follows: ΔT i =inf i ×tgc Where tgc is the token generation coefficient, ranging from 1 to 30, with experiments conducted in steps of 1. A suitable value is fitted based on the convergence speed of the seed set obtained from the results; faster convergence speed results in better tgc. i It is v i The user's self-influence means that after the token is generated, the newly generated token will be added to the user's token pool; Step 4: Develop an iterative token distribution strategy; combining the user's own information, neighbor information, and the current set of tokens held, in each iteration, the user passes their own token to their neighboring users according to the strategy to simulate the information flow process; the token iterative distribution strategy method is as follows: Step 4-1: Calculation The influence ratio of each user node, as one of the standards for token distribution, is calculated as follows: Among them, inf j User v i Neighbor v j The influence User v i The sum of the influence of neighbors; Step 4-2: Through node v i Token pool T i The composition of the data is used to calculate the neighbor nodes from user v in one iteration. i The number of tokens obtained, since tokens are differentiated, v j From v i The obtained token set is calculated as follows: Where, weight ji Represents node v j Occupy node v i The proportion of the sum of the influence of neighboring nodes, where n is the number of nodes in the social network G, and S is the weight of the sum of the influence of neighboring nodes. k Indicates v i The token pool from v k The set of tokens, e ij Indicates v j With v i The degree of intimacy between them, i.e., the probability of transmission between them; Step 4-3: Token Normalization. Due to the atomic nature of tokens, the number of tokens must be an integer during the transfer process. Therefore, the calculation result needs to be rounded down to maintain the integrity of the transfer process and the consistency of the model. Indicates will Round down; Step 5: After each iteration, filter the tokens held by the user, remove redundant information, form an intermediate result view, and gradually approach the target seed set. Step 6: Perform convergence analysis on the seed set using Jaccard and Dice similarity coefficients to ensure that the final seed set has propagation potential when it meets the preset convergence criteria.

2. The method as described in claim 1, characterized in that, The steps to construct a social network graph G are as follows: Step 2-1: A social network consists of many users with different relationships; a social network is represented by a graph structure as described below; G = {V, E} Where V = {v1, v2, ... v} n } is a user set, v i For the i-th user in the social network, E = {e 12 ,…,e (n-1)n } is the edge set of users in the network, and edge e ij Corresponding to v i and v j The relationship between them is initially defined by the number of connections between them, representing the strength and intimacy of the relationship between the users, and e i,j =e j,i That is, an undirected graph form; Step 2-2: For the edge set E = {e 12 ,…,e (n-1)n Perform a normalization operation on e. ij Values ​​are mapped to (0,1], where 0 and 1 represent the furthest and closest relationships between two users, respectively. The normalization formula is as follows: Steps 2-3: The definition of the social network graph G is as follows: v i ={id,inf i ,LE i ,LV i ,T i } Where id is user v i The identifier, inf i It is v i Self-influence It is v i The adjacent edge set, It is v i A set of directly connected neighboring users, v j It is v i A user in a neighboring user set; T i It is v i A token pool containing v i All tokens owned, T i The defined formula is: T i ={t1, t2, ... t n }, t i Indicates v i A token owned is represented as: t i ={s id ,l},s id The node that generated the token is represented by 'l', and 'l' represents the path length of the token.

3. The method as described in claim 1, characterized in that, The intermediate result view is generated as follows: after each iteration, the tokens held by the user are filtered to remove redundant information, forming an intermediate result view, including: Step 5-1: Calculate the evaluation function for node scores, using the following formula: Where d(v) represents the types of tokens in the node's token pool, l k This indicates the v that generated the token. k Node and v i Node distance, t k This indicates that the token pool contains tokens from v. k The number of tokens generated by the user, α and β are weight parameters, obtained by fitting; Step 5-2: Filter out tokens with fewer than 50 types in the token pool, calculate the user node evaluation function, and obtain the intermediate result S of the iteration.

4. The method as described in claim 1, characterized in that, The steps for calculating seed set convergence are as follows: Step 6-1: Perform convergence analysis on the seed set using Jaccard and Dice similarity coefficients to ensure that the final seed set has propagation potential when it meets the preset convergence criteria. The formula is as follows: Where S represents the intermediate result of the seed set obtained in this iteration, and S′ represents the intermediate result of the seed set obtained in the previous iteration; When the similarity coefficient between Jaccard and Dice is consistently below 0.05, the model is considered to have reached the convergence condition.

5. The method as described in claim 1, characterized in that, An IC model diffusion step can also be added. The IC model diffusion test method is as follows: The seed set S obtained through iteration is then used to conduct an influence diffusion test on the final seed set using the IC independent cascade diffusion model. It is assumed that at time t, node v... i There are x neighbor nodes a1, a2, ..., a3 that are activated at time t-1. n The probabilities of these nodes activating node v are respectively Then node v i The probability of being activated is: If node v i If it is activated, it indicates that v i When covered, the IC diffusion range increases; Among the seed sets selected by different models, the IC diffusion model has the highest coverage, indicating that the effect of maximizing the influence selection is the best. The closer the seed set selected by this model is to the optimal seed set, the better the model's performance.

Images