A method for maximizing influence based on multi-layer networks

The joint degree reduction method in multi-layer networks selects seed nodes to maximize influence spread, addressing overlapping propagation and ensuring efficient information dissemination across diverse platforms.

CN116094937BActive Publication Date: 2024-12-03ZHEJIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211730580.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-30
Publication Date
2024-12-03
Estimated Expiration
2042-12-30

AI Technical Summary

Technical Problem

Existing algorithms for influence maximization in multi-layer networks face challenges in achieving efficient propagation with low time complexity and addressing overlapping propagation issues, limiting their effectiveness in real-world applications such as marketing and rumor containment.

Method used

A heuristic joint degree reduction method is applied to select seed nodes in multi-layer networks, utilizing the union of neighbor node sets as joint degree values and reducing overlap through iterative selection and degree reduction, followed by simulation using the IC propagation model to evaluate influence.

Benefits of technology

The method maximizes influence spread while maintaining low time complexity, effectively addressing overlapping propagation and enhancing information dissemination across multiple platforms, optimizing advertising and information credibility.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116094937B_ABST
    Figure CN116094937B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for maximizing influence based on a multi-layer network. The present invention combines the topological properties of the multi-layer network to mine a set of nodes with high influence, including three modules: model establishment, seed node selection, and seed node identification effect evaluation. Model establishment is a task of establishing a multi-layer network model; the seed node selection process includes a neighbor union preprocessing method and a joint degree reduction algorithm, which are respectively used to preliminarily obtain the importance ranking of seed nodes, and through the heuristic joint degree reduction algorithm, the negative impact of propagation overlap is weakened, and the operation is repeated until the final seed node set is selected; finally, the actual effect of the seed node set is evaluated through a simulated propagation experiment on a multi-layer network. The present invention comprehensively considers the differences between nodes in different layers and the problem of propagation overlap, and proposes new ideas and methods to solve the maximization of the influence of multi-layer networks, which has great applications in both economic utility and social services.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, in particular to the field of influence technology of multi-layer networks, and relates to a method for maximizing influence based on a multi-layer network, which identifies an optimal seed node set according to the topological properties of the multi-layer network to achieve the maximum propagation scale. Background Art

[0002] Influence maximization is to find a group of nodes under a specific network propagation model so that the final influence (propagation range) of this group of nodes is maximized. The research on influence maximization has very important practical significance and is widely used in product marketing, disease control, and personalized recommendations. Since Kempe et al. proved that the influence maximization problem is an NP-hard problem and proposed two classic propagation models (independent cascade and linear threshold), the research on influence maximization has been based on single-layer networks for a long time, and the research topic revolves around the algorithm for finding seed nodes.

[0003] In today's era, people often have multiple social software and spread information to different platforms. The multi-layer network composed of different social platforms provides a new idea for the problem of maximizing influence. Multi-layer networks focus on describing the directionality, diversity and difference of the edges between nodes in the network. They are also a powerful tool for analyzing the complexity of network systems and the heterogeneous interaction characteristics between network layers. Domenico et al. have proved that information is better spread on multi-layer networks. How to maximize the spread of information on multi-layer social networks has important practical economic significance. This study hopes to find an algorithm with better running time and better spread effect by studying the problem of maximizing influence on multi-layer networks, and apply it to multi-layer networks to provide support for real-world applications such as marketing, advertising, and rumor containment.

[0004] The propagation dynamics on multi-layer networks is a topic worth exploring. The interactive coupling information between layers in multi-layer networks makes its dynamic process diverse and complex. In recent years, researchers have begun to pay attention to the problem of maximizing influence on multi-layer networks. Most of the algorithms are algorithms that were previously applied to single-layer networks and extended to multi-layer networks, including greedy, degree, CIM, CBIM, KSN, etc. However, how to design an influence maximization algorithm based on the network topology properties of multi-layer networks needs further research. Summary of the invention

[0005] The purpose of the present invention is to provide a method for maximizing influence based on a multi-layer network. The method utilizes the node properties of the multi-layer network and uses a heuristic joint degree reduction method to reduce the influence of neighbor overlap, and finally selects a set of seed nodes that can maximize the influence scale. At the same time, the present invention also has a low time complexity while ensuring the effect.

[0006] The method of the present invention is specifically as follows:

[0007] Step (1) Construct a multi-layer network model G M ={G1,…,G k ,…,G K}, where G k = {V k ,E k} represents the k-th layer network, which consists of the point set V in the layer k and the layer internal edge set E k , k = 1, 2, ..., K, K represents the total number of layers, the number of nodes in each layer is the same, and the nodes at the same position in each layer correspond one to one; the inter-layer edges only exist between the same nodes in different layers, and the edge relationship is represented by a fourth-order tensor It is represented as follows: the i-node of the α-th layer is connected to the i-node of the β-th layer; the point set within each layer of the multi-layer network model is taken as the union to obtain the node set V of the entire multi-layer network; V = |V1∪V2…∪V k …∪V K |, V k Represents the point set within the layer of the k-th layer network.

[0008] Step (2) Select seed nodes by joint degree reduction:

[0009] (2-1) Count the neighbor node sets of each node in each layer of the network, take the union of the neighbor node sets of the corresponding nodes in the multi-layer network, and use the length of the union as the influence value score D of the node m , that is, the joint degree value of the node; D m =|N1(v m )∪N2(v m )…∪N k (v m )…∪N K (v m )|,N k (v m ) represents node v m The set of neighbor nodes in the k-th layer network, m = 1, 2, ..., M, M is the total number of nodes in the layer network;

[0010] (2-2) The joint degree values ​​D1,…,D m ,…,D MArrange in descending order of value, select the node v′1 with the largest joint degree value as the seed node, and add it to the seed node set S. The subscript 1 indicates the first round of selection;

[0011] (2-3) Reduce the joint degree values ​​of nodes in the node set V except those that have been selected as seed nodes: First, traverse the node set V and count the number of seed nodes n in the first-order neighbor range of other nodes except those that have been selected as seed nodes; then, m The joint degree value of the updated joint degree value is D′ m =D m -e n ;

[0012] (2-4) For the updated joint degree values ​​D′1, D′2, …, D′ M-1 Arrange in descending order of value, select the node v′2 with the largest updated joint degree value as the seed node selected in this round, and add it to the seed node set S. The subscript 2 indicates the second round of selection;

[0013] (2-5) Repeat steps (2-3) and (2-4) to select the node with the maximum degree in each round and add it to the seed node set; until the seed nodes of the preset R rounds are selected as the seed node set with the greatest influence S = {v′1,v′2,…,v′ R}.

[0014] Step (3) maximizes the seed node recognition effect evaluation:

[0015] The seed node set selected as the most influential is simulated for propagation on the multi-layer network, and the propagation influence range of the seed node set is calculated; the IC propagation model is used to simulate the propagation dynamics of nodes on the multi-layer network, and the state of each node is described as two possible states: inactive and active; the active node infects its inactive neighbor nodes with probability γ, and each activated node has only one chance to activate the uninfected neighbor node. If it is not activated, it will not try again; the IC model is extended to the multi-layer network, while retaining the above properties, the multi-layer network propagation mechanism is designed as follows:

[0016] ①The propagation order starts from the top layer by default;

[0017] ②When a node is activated, it is assumed that the corresponding nodes in other layers also have the ability to propagate.

[0018] Take S as the initial seed node to conduct a simulated propagation experiment on the above IC propagation model. When the propagation is terminated, the number of nodes in the network that are in an activated state is counted. The number of active nodes in the network is used as the propagation influence range to quantify the propagation influence range of the selected seed node.

[0019] Beneficial effects of the present invention: The present invention designs a heuristic joint degree reduction algorithm, which is designed in combination with the topological properties of the multi-layer network and applied to the multi-layer network to maximize the influence. The method fully considers the phenomenon of overlapping propagation that may occur during the propagation process, that is, more than one seed node may be selected in the same area, resulting in a limited propagation range and some small edge clusters cannot be propagated. To solve this problem, the "joint degree reduction" method is adopted to punish the nodes within the local range of the seed node to a certain extent, and the intensity of the punishment increases with the increase of the seed nodes in the local range. While ensuring the importance of the node, it will not be too small because the punishment intensity is too small and the effect is minimal. When the way of information dissemination is no longer limited to a single social tool, the importance of multi-layer network research is self-evident, and it has great applications in both economic applications and social services. The method is in line with the multi-level and diversified themes of the current era. From the perspective of scientific research, it provides new ideas and methods for the study of maximizing the influence of multi-layer networks; from a practical perspective, in the context of the widespread use of multiple social platforms, advertising companies achieve the greatest publicity effect at the lowest cost; the government publishes information on different platforms to improve the credibility of the published information. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The figure is a flow chart of the overall framework of the method of the present invention. DETAILED DESCRIPTION

[0021] The present invention will be further described below by means of specific examples.

[0022] A method for maximizing influence based on a multi-layer network takes into account the topological properties of the multi-layer network, adopts the union size of the multi-layer neighbor sets as the "joint degree value", and after selecting the seed node, performs a reduction algorithm on the neighbor nodes of the node in each layer to reduce the negative impact of "propagation overlap". This is the key point of the technical solution and enriches the method of selecting seed nodes with maximum influence on a multi-layer network.

[0023] like Figure 1 As shown, the method is specifically:

[0024] Step (1) Construct a multi-layer network model G M ={G1,…,G k ,…,G K}, where G k = {V k ,E k} represents the k-th layer network, which consists of the point set V in the layer k and the layer internal edge set E kThe structure is k = 1, 2, ..., K, where K represents the total number of layers. The number of nodes in each layer is the same, and the nodes at the same position in each layer correspond to each other. The inter-layer edges only exist between the same nodes in different layers, and the edge relationship is represented by the fourth-order tensor It is represented as follows: the i-node of the αth layer is connected to the i-node of the βth layer. The node set V of the entire multi-layer network is obtained by taking the union of the points within each layer of the multi-layer network model; V = |V1∪V2…∪V k …∪V K |, V k Represents the point set within the layer of the k-th layer network.

[0025] Step (2) Selecting seed nodes by joint degree reduction;

[0026] (2-1) Count the neighbor node sets of each node in each layer of the network, take the union of the neighbor node sets of the corresponding nodes in the multi-layer network, and use the length of the union as the influence value score D of the node m , which is the joint degree value of the node;

[0027] D m =|N1(v m )∪N2(v m )…∪N k (v m )…∪N K (v m )|,N k (v m ) represents node v m The set of neighbor nodes in the k-th layer network, m = 1, 2, ..., M, M is the total number of nodes in the layer network.

[0028] (2-2) The joint degree values ​​D1,…,D m ,…,D M Arrange in descending order of value, select the node v′1 with the largest joint degree value as the seed node, and add it to the seed node set S. The subscript 1 indicates the first round of selection;

[0029] (2-3) Reduce the joint degree values ​​of nodes in the node set V except those that have been selected as seed nodes: First, traverse the node set V and count the number of seed nodes n in the first-order neighbor range of other nodes except those that have been selected as seed nodes; then, m The joint degree value of the updated joint degree value is D′ m =D m -e n ;

[0030] (2-4) For the updated joint degree values ​​D′1, D′2, …, D′ M-1Arrange in descending order of value, select the node v′2 with the largest updated joint degree value as the seed node selected in this round, and add it to the seed node set S. The subscript 2 indicates the second round of selection;

[0031] (2-5) Repeat steps (2-3) and (2-4) to select the node with the maximum degree in each round and add it to the seed node set; until the seed nodes of the preset R rounds are selected as the seed node set with the greatest influence S = {v′1,v′2,…,v′ R}.

[0032] Step (3) Maximize the seed node recognition effect evaluation:

[0033] The effectiveness is verified by simulating the propagation of the seed node set selected as the most influential on the multi-layer network and calculating the propagation influence range of the seed node set. The IC (Independent Cascade) propagation model is used to simulate the propagation dynamics of nodes on the multi-layer network. In the traditional IC propagation model, the state of each node is described as two possible states: inactive and active. The active node infects its inactive neighbor nodes with a probability of γ. Each activated node has only one chance to activate the uninfected neighbor node. If it is not activated, it will not be tried again. The IC model is extended to the multi-layer network. While retaining the above properties, the multi-layer network propagation mechanism is designed as follows:

[0034] ①The propagation order starts from the top layer by default;

[0035] ②When a node is activated, it is assumed that the corresponding nodes in other layers also have the ability to propagate.

[0036] Take S as the initial seed node to conduct a simulated propagation experiment on the above IC propagation model. When the propagation is terminated, the number of nodes in the network that are in an activated state is counted. The number of active nodes in the network is used as the propagation influence range to quantify the propagation influence range of the selected seed node.

Claims

1. A method for maximizing influence based on a multi-layer network, characterized by: Step (1) Construct a multi-layer network model G M ={G1,…,G k ,…,G K }, where G k = {V k ,E k } represents the k-th layer network, which consists of the point set V in the layer k and the layer internal edge set E k , k = 1, 2, ..., K, K represents the total number of layers, the number of nodes in each layer is the same, and the nodes at the same position in each layer correspond one to one; the inter-layer edges only exist between the same nodes in different layers, and the edge relationship is represented by a fourth-order tensor It is represented as follows: the i-node of the α-th layer is connected to the i-node of the β-th layer; the point set within each layer of the multi-layer network model is taken as the union to obtain the node set V of the entire multi-layer network; V = |V1∪V2…∪V k …∪V K |, V k Represents the point set within the layer of the k-th layer network; Step (2) Selecting seed nodes by joint degree reduction; (2-1) Count the neighbor node sets of each node in each layer of the network, take the union of the neighbor node sets of the corresponding nodes in the multi-layer network, and use the length of the union as the influence value score D of the node m , which is the joint degree value of the node; D m =|N1(v m )∪N2(v m )…∪N k (v m )…∪N K (v m )|,N k (v m ) represents node v m The set of neighbor nodes in the k-th layer network, m = 1, 2, ..., M, M is the total number of nodes in the layer network; (2-2) The joint degree values ​​D1,…,D m ,…,D M Arrange in descending order of value, select the node v′1 with the largest joint degree value as the seed node, and add it to the seed node set S. The subscript 1 indicates the first round of selection; (2-3) Reduce the joint degree values ​​of nodes in the node set V except those that have been selected as seed nodes: First, traverse the node set V and count the number of seed nodes n in the first-order neighbor range of other nodes except those that have been selected as seed nodes; then, m The joint degree value of the updated joint degree value is D′ m =D m -e n ; (2-4) For the updated joint degree values ​​D′1, D′2, …, D′ M-1 Arrange in descending order of value, select the node v′2 with the largest updated joint degree value as the seed node selected in this round, and add it to the seed node set S. The subscript 2 indicates the second round of selection; (2-5) Repeat steps (2-3) and (2-4) to select the node with the maximum degree in each round and add it to the seed node set; until the seed nodes of the preset R rounds are selected as the seed node set with the greatest influence S = {v′1,v′2,…,v′ R }; Step (3) maximize the seed node recognition effect evaluation; The seed node set selected as the most influential is simulated and propagated on the multi-layer network to calculate the propagation influence range of the seed node set; The IC propagation model is used to simulate the propagation dynamics of nodes in multi-layer networks. The state of each node is described as two possible states: inactive and active. An active node infects its inactive neighbor nodes with a probability of γ. Each activated node has only one chance to activate an uninfected neighbor node. If it is not activated, no attempt will be made. The IC model is extended to multi-layer networks. While retaining the above properties, the multi-layer network propagation mechanism is designed as follows: ①The propagation order starts from the top layer by default; ②When a node is activated, it is assumed that the corresponding nodes in other layers also have the ability to spread; Take S as the initial seed node to conduct a simulated propagation experiment on the above IC propagation model. When the propagation is terminated, the number of nodes in the network that are in an activated state is counted, and the number of active nodes in the network is used as the range of influence of the propagation.