Method and system for establishing dynamic random graph model

Abstract

The present invention belongs to the technical field of random systems, and specifically discloses a method and system for establishing a dynamic random graph model. The method for establishing a dynamic random graph model assumes that the number of nodes in the graph network is constant and the positions are unchanged. The distance between nodes is calculated by the node positions, and the standardized distance between each node is obtained. An influence factor is assigned to each node in the initial graph structure, and the power-law distribution of the actual network is simulated by the subsequent edge generation method. Each node is traversed in a loop, and the edges between nodes are generated to form an initial graph structure. The neighbor node set obtained by the initial graph structure is further traversed for each node. The neighbor node set of its neighbor nodes is further added to the graph through the influence factor. If the characteristics of the graph structure reach a preset range, the graph structure is reconstructed. The technical solution is adopted to simulate the power-law distribution characteristics of the real-world relationship network, and the dynamic changes of the real-world relationship network are simulated through the dynamic random graph model.

Description

Technical Field

[0001] The invention belongs to the technical field of random systems and relates to a method and a system for establishing a dynamic random graph model. Background Art

[0002] Relational networks in real life are often complex and difficult to analyze. However, due to the common characteristics and similar patterns such as power-law distribution and clustering coefficient in multiple networks in real life, it is possible to design graph models that simulate real social networks.

[0003] Social network structure is crucial for simulating information diffusion, community cohesion, knowledge aggregation, social interaction phenomena, etc. But at the same time, social behavior is also complex, and random graph models can capture the laws of the process of generating network connections and recognize the variability in them. If the designed graph structure can be made similar to the social relationship network in real life in some aspects, it will be possible to understand and analyze the real-world network at a very low cost.

[0004] A graph structure is a computer data structure consisting of a collection of vertices and edges between vertices, usually expressed as: G(V,E), where V is the collection of vertices in the graph structure G, and E is the collection of edges in the graph structure G. The random graph model introduces randomness in the process of generating edges between vertices, so that its generation process can simulate the real social network as realistically as possible. However, in the existing technology, most models treat each node the same, so they cannot simulate the power-law distribution of the "edges" of real-world relationships; and the existing models are basically static models, that is, the graph network remains unchanged after the generation is completed, while the real-world social network is often in dynamic change. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for establishing a dynamic random graph model to simulate the power-law distribution characteristics of real-world relationship networks.

[0006] In order to achieve the above object, the basic scheme of the present invention is: a method for establishing a dynamic random graph model, comprising the following steps:

[0007] S1, assuming that the number of nodes in the graph network is fixed and their positions remain unchanged;

[0008] S2, calculate the distance between nodes through node positions and obtain the standardized distance between each node;

[0009] Assign an influence factor to each node in the initial graph structure. The influence factor set conforms to the right half of the Poisson distribution or the normal distribution, or the power-law distribution, and simulates the power-law distribution of the actual network through the subsequent edge generation method;

[0010] Loop through each node, generate edges between nodes and form the initial graph structure;

[0011] S3, the neighbor node set obtained through the initial graph structure, further traverses the neighbor node set of its neighbor nodes for each node, and further adds edges to the graph through the influence factor;

[0012] S4: If the characteristics of the graph structure reach a preset range, the graph structure is rebuilt.

[0013] The working principle and beneficial effects of this basic solution are as follows: through the multi-stage generation and deletion of edges, the relationship network in the real world is dynamically simulated, and the establishment of the relationship network is divided into multiple stages. The dynamic simulation of multiple stages provides a way to analyze and explore the evolution of the real relationship network. Nodes are distinguished by influence factors, and the power-law distribution of the actual network is simulated by combining multi-stage processing of graph structure edges. For the generation of graph structure edges, the distance factor between nodes is introduced to simulate the actual network, which is simple to operate and easy to use.

[0014] Further, after the graph structure is reconstructed, steps S2-S3 are repeated on the reconstructed graph structure until the model reaches an expected similarity with the real network in terms of expected network features.

[0015] Optimized model, simple operation and easy to use.

[0016] Furthermore, the edge generation method in the initial graph structure is as follows:

[0017] For the edge E between node i and node j in the graph structure G ij , using the probability calculation formula for edge E ij Whether there is control:

[0018] p(E ij )=f(popul i ×popul j )×g(dist ij )

[0019] Among them, popul i represents the influence factor of node i, and the influence factors of all nodes obey a random variable X that conforms to Poisson distribution or normal distribution, popul i ∈X;popul j represents the influence factor of node j, f and g are two independent activation functions, the activation function f(x) is the sigmoid function that normalizes x to (0, 1), or the 01 activation function that shifts the origin to (0, 1); Represents two functions, and the activation function g is distij represents the normalized distance between node i and node j;

[0020] Whether there is an edge between node i and node j is mainly determined by the influence factor and distance of the two nodes. f() and normalized () are the activation functions that control the number of edges of the entire graph G. Nodes are distinguished by influence factors that obey Poisson distribution or normal distribution to simulate the power-law distribution of actual networks.

[0021] For the generation of graph structure edges, the distance factor between nodes is introduced to simulate the actual network for easy operation.

[0022] Furthermore, the method for obtaining the neighbor node set of each node is as follows:

[0023] Define the neighbor set S of node i i , for all edges E ij =1 and the union of the set S of nodes j and node i itself. For each neighbor node set S of node i, it can be expressed by the following formula:

[0024] S={s|E sj =1, for every j if E ij =1}.

[0025] Get the neighbor node set of each node. The operation is simple and easy to use.

[0026] Furthermore, the method of adding edges to the graph by influence factor is as follows:

[0027] Connecting isolated nodes:

[0028] After the initial graph structure is created in step S2, the isolated nodes are traversed in a loop. The number of edges of the isolated nodes is 0, and the number of edges of the connected nodes is greater than 0. The isolated nodes are connected to the connected nodes with probability p1=f(popul i ×popul j )×β generates an edge, β∈(0,1), or with probability p2=f(popul i ×popul j )×g(dist ij ) Generate an edge; if an edge is generated, skip the current isolated node; popul i represents the influence factor of node i, dist ij represents the normalized distance between node i and node j, and f and g are two independent activation functions;

[0029] Traversal of neighbor sets of node neighbors:

[0030] Traverse each neighbor node j of node i, V is the node set, for each node s in the neighbor node set S of j, with probability p = f (popul i ×popul j )×σ to generate edges, σ∈(0,1).

[0031] Simple operation and easy to use.

[0032] Furthermore, the characteristics of the graph network are used to determine whether the graph structure needs to be reconstructed;

[0033] The current random graph network features are combined with the actual network to make a judgment. The graph network features include clustering coefficient and average path length. When the clustering coefficient is greater than the preset value or the average path length is less than the preset value, the graph structure needs to be rebuilt.

[0034] In an undirected network, the clustering coefficient is defined as:

[0035] C = 2*n / (k*(k-1)),

[0036] Where n represents the number of edges between all k neighbors of node v;

[0037] The average path length is:

[0038]

[0039] Where N is the number of network nodes, d ij is the shortest path between nodes i and j.

[0040] The judging process is simple to operate and easy to use.

[0041] Further, the method of reconstructing the graph structure is as follows:

[0042] The number of common neighbors between nodes is used to determine whether to retain the edges between nodes. The edges whose number of common neighbors ranks in the top a% in the entire network, a∈(0,100) are retained, and the edges that do not meet the conditions are divided with a probability of p∈(0,1);

[0043] Other reconstruction methods include, but are not limited to: randomly deleting a preset ratio of edges, deleting edges between nodes whose overlap ratio is lower than a preset threshold;

[0044] The calculation formula of the overlap rate between two nodes i and j is:

[0045] Where N s is the number of elements in the neighbor set S, is the common neighbor set S of node i and node j i The number of elements of is the neighbor set S of node ii The number of elements contained, is the neighbor set S of node j i The number of elements contained.

[0046] Since the interaction phase is cyclical, the number of edges in the entire random graph network is constantly growing. Therefore, it is necessary to reconstruct the graph when it grows to a certain extent. The operation is simple and easy to use.

[0047] The present invention also provides a network analysis method using a dynamic random graph model, comprising the following steps:

[0048] Model the real relationship network, and by comparing the characteristics of the real relationship network with the model graph network, determine the development stage of the relationship and predict whether the relationship network will change significantly within a certain period of time;

[0049] The edges of the model are regarded as paths for information propagation. Individuals with high influence in real life are mapped to nodes with high influence factor scores in the model graph network for modeling, and the impact of information propagation through this individual on the entire graph network is predicted.

[0050] Through the dynamic random graph model, the dynamic changes of the real-world network are simulated, which is convenient for use.

[0051] The present invention also provides a system for establishing a dynamic random graph model based on the method described in the present invention, including a creation module, an interaction module and a reconstruction module. The creation module is used to create an initial graph structure and distinguish nodes. The interaction module is used to obtain a neighbor node set for each node and generate a graph structure edge between the node and the nodes in the corresponding neighbor node set. The reconstruction module is used to reconstruct the graph structure.

[0052] The system has a simple structure and can simulate the power-law distribution characteristics of real-world relationship networks, which is easy to use.

[0053] The present invention also provides a network analysis system using a dynamic random graph model based on the method of the present invention, including an information collection unit and a modeling unit. The information collection unit is used to obtain real-life relationship network information. The output end of the information collection unit is connected to the input end of the modeling unit. The modeling unit performs modeling based on the real-life relationship network and predicts whether the relationship network will have significant changes within a certain period of time, as well as predicts the impact of the individual's propagated information on the entire graph network information propagation.

[0054] This system can simulate various relationship networks in the real world, understand, analyze and explore the laws and evolution methods therein, and is easy to use. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1It is a flow chart of the method for establishing a dynamic random graph model according to the present invention. DETAILED DESCRIPTION

[0056] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0057] In the description of the present invention, it is necessary to understand that the terms "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0058] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0059] The present invention discloses a method for establishing a dynamic random graph model to simulate the power law distribution characteristics of a real-world relationship network, and to simulate the dynamic changes of a real-world network through a dynamic random graph model. Figure 1 As shown, the method comprises the following steps:

[0060] S1, assuming that the number of nodes in the graph network is fixed and their positions remain unchanged;

[0061] S2, calculate the distance between nodes by node positions, and obtain the standardized distance between each node; the method for obtaining the standardized distance between each node is: calculate the Euclidean distance between two nodes by the coordinates of the two nodes, and after obtaining the distances between all nodes, normalize the entire distance sample to (0,1);

[0062] Assign an influence factor to each node in the initial graph structure. The influence factor set conforms to the right half of the Poisson distribution or the normal distribution, or the power-law distribution, and simulates the power-law distribution of the actual network through the subsequent edge generation method;

[0063] Loop through each node, generate edges between nodes and form the initial graph structure;

[0064] S3, the neighbor node set obtained through the initial graph structure, further traverses the neighbor node set of its neighbor nodes for each node, and further adds edges to the graph through the influence factor;

[0065] S4: If the characteristics of the graph structure reach a preset range (such as a clustering coefficient greater than or equal to 0.3 or an average path length less than or equal to 6), the graph structure is rebuilt.

[0066] After the graph structure is reconstructed, steps S2-S3 are repeated on the reconstructed graph structure until the model reaches the expected similarity with the real network in terms of expected network characteristics, such as the similarity between the edge distribution of the real network and the simulated network is greater than 80%.

[0067] In a preferred embodiment of the present invention, the edge generation method in the initial graph structure is as follows:

[0068] Creation Phase: Define the edge E in the graph structure G(V, E) ij is the edge between nodes i and j. If this edge exists, then E ij = 1. For the edge E between node i and node j in the graph structure G ij , using the probability calculation formula for edge E ij Whether there is control:

[0069] p(E ij )=f(popul i ×popul j )×g(dist ij )

[0070] If p(E ij )=0.9, it means that there is a 90% probability that nodes i and j will form an edge; i - represents the influence factor of node i, and the influence factors of all nodes obey a random variable X that conforms to Poisson distribution or normal distribution, popul i ∈X;popul j represents the influence factor of node j, f and g are two independent activation functions. The appropriate activation function f(x) can be a sigmoid function that normalizes x to (0, 1) or a 01 activation function that shifts the origin to (0, 1). Representing the above two functions, the appropriate activation function g can be dist ij represents the normalized distance between node i and node j;

[0071] Whether there is an edge between node i and node j is mainly determined by the influence factor and distance of the two nodes. f() and g() are activation functions that control the number of edges of the entire graph G. The graph in the creation stage meets the realistic conditions that "individuals with greater influence are more likely to have social relationships" and "individuals with closer distances are more likely to get to know each other". Nodes are distinguished by influence factors that obey Poisson distribution or normal distribution to simulate the power-law distribution of actual networks.

[0072] In a preferred embodiment of the present invention, the method for obtaining the neighbor node set of each node is as follows:

[0073] Mingle Phrase: Define the neighbor set S of node i i , for all edges E ij =1 (there is an edge between i and j) and the union of the set S of nodes j and node i itself. For each neighbor node set S of node i’s neighbors, it is expressed by the following formula:

[0074] S={s|E sj =1, for every j if E ij =1}.

[0075] In a preferred embodiment of the present invention, a method for further adding edges to a graph by using influence factors is as follows:

[0076] Connecting isolated nodes:

[0077] After the initial graph structure is created in step S2, the isolated nodes are traversed in a loop. The number of edges of the isolated nodes is 0, and the number of edges of the connected nodes is greater than 0. The isolated nodes are connected to the connected nodes with probability p1=f(popul i ×popul j )×β generates an edge, β∈(0,1), or with probability p2=f(popul i ×popul j )×g(dist ij ) Generate an edge; if an edge is generated, skip the current isolated node; popul i represents the influence factor of node i, dist ij represents the normalized distance between node i and node j, and f and g are two independent activation functions;

[0078] Traversal of neighbor sets of node neighbors:

[0079] Traverse each neighbor node j of node i, V is the node set, for each node s in the neighbor node set S of j, with probability p = f (popul i ×popul j)×σ to generate edges, σ∈(0,1).

[0080] In a preferred embodiment of the present invention, the reconstruction phase (Reconstruct Phrase):

[0081] Determine whether the graph structure needs to be rebuilt based on the characteristics of the graph network. Since the interaction phase is cyclical, the number of edges in the entire random graph network is constantly growing, so reconstruction is required when the graph grows to a certain size.

[0082] The current random graph network features are judged in combination with the specific actual network. The graph network features include clustering coefficient and average path length. When the clustering coefficient is greater than the preset value or the average path length is less than the preset value (for example, the clustering coefficient is greater than or equal to 0.3 or the average path length is less than or equal to 6), the graph structure needs to be rebuilt.

[0083] Clustering coefficient calculation method: In an undirected network, the clustering coefficient is defined as:

[0084] C = 2*n / (k*(k-1)),

[0085] Where n represents the number of edges between all k neighbors of node v;

[0086] The average path length is:

[0087]

[0088] Where N is the number of network nodes, d ij is the shortest path between nodes i and j.

[0089] In a preferred embodiment of the present invention, the method for reconstructing the graph structure is as follows:

[0090] By judging whether to keep the edges between nodes based on the number of common neighbors between nodes, keep the edges whose number of common neighbors ranks in the top a% in the entire network, a∈(0,100), and remove the edges that do not meet the conditions with a probability of p∈(0,1). For example, keep the edges whose number of common neighbors is greater than 3, and delete the edges that do not meet the conditions with a certain probability.

[0091] Other reconstruction methods include, but are not limited to: randomly deleting edges of a preset ratio (such as 5%, 7%, 10%, etc.), deleting edges between nodes whose overlap ratio is lower than a preset ratio value (such as 30%);

[0092] The calculation formula of the overlap rate between two nodes i and j is:

[0093]

[0094] Among them, N s is the number of elements in the neighbor set S, is the common neighbor set S of node i and node j i The number of elements of is the neighbor set S of node i i The number of elements contained, is the neighbor set S of node j i The number of elements contained.

[0095] The present invention also provides a network analysis method using a dynamic random graph model, comprising the following steps:

[0096] Model the real relationship network, and by comparing the characteristics of the real relationship network with the model graph network, determine the development stage of the relationship and predict whether the relationship network will undergo significant changes (experience a reconstruction phase) within a certain period of time;

[0097] The edges of the model are regarded as paths for information diffusion. Individuals with high influence in real life are mapped to nodes with high influence factor scores in the model graph network for modeling, and the impact of information dissemination through the individual on the entire graph network is predicted. Through the dynamic random graph model, the dynamic changes of the relationship network in the real world are simulated, and various relationship networks in the real world are simulated at a lower cost to understand, analyze and explore the laws and evolution methods therein.

[0098] The present invention also provides a system for establishing a dynamic random graph model based on the method described in the present invention, including a creation module, an interaction module and a reconstruction module. The creation module is used to create an initial graph structure and distinguish nodes. The interaction module is used to obtain a neighbor node set for each node and generate a graph structure edge between the node and the nodes in the corresponding neighbor node set. The reconstruction module is used to reconstruct the graph structure.

[0099] This scheme dynamically simulates the relationship network in the real world through the multi-stage generation and deletion of edges, and divides the establishment of the relationship network into multiple stages; it distinguishes nodes through influence factors that obey the Poisson distribution (or normal distribution) to simulate the power-law distribution of the actual network; for the generation of graph structure edges, the distance factor between nodes is introduced to simulate the actual network.

[0100] The present invention also provides a network analysis system using a dynamic random graph model based on the method of the present invention, including an information collection unit and a modeling unit, wherein the information collection unit is used to obtain real relationship network information, the output end of the information collection unit is connected to the input end of the modeling unit, and the modeling unit is modeled according to the real relationship network, and predicts whether the relationship network will have a large change within a certain period of time, and predicts the impact of the individual propagation information on the entire graph network information propagation. This system can simulate various relationship networks in the real world, understand, analyze and explore the laws and evolution methods therein.

[0101] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0102] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for establishing a dynamic random graph model, characterized in that: The steps include: S1, assuming that the number of nodes in the graph network is fixed and their positions remain unchanged; S2, calculate the distance between nodes through node positions and obtain the standardized distance between each node; Assign an influence factor to each node in the initial graph structure. The influence factor set conforms to the right half of the Poisson distribution or the normal distribution, or the power-law distribution, and simulates the power-law distribution of the actual network through the subsequent edge generation method; Loop through each node, generate edges between nodes and form the initial graph structure; S3, the neighbor node set obtained through the initial graph structure, further traverses the neighbor node set of its neighbor nodes for each node, and further adds edges to the graph through the influence factor; The method to further add edges to the graph by influence factor is as follows: Connecting isolated nodes: After the initial graph structure is created in step S2, the isolated nodes are traversed in a loop. The number of edges of the isolated nodes is 0, and the number of edges of the connected nodes is greater than 0. The isolated nodes are connected to the connected nodes with probability p1=f(popul i ×popul j )×β generates an edge, β∈(0,1), or with probability p2=f(popul i ×popul j )×g(dist ij ) Generate an edge; if an edge is generated, skip the current isolated node; popul i represents the influence factor of node i, dist ij represents the normalized distance between node i and node j, f and g are two independent activation functions; popul j represents the influence factor of node j; Traversal of neighbor sets of node neighbors: Traverse each neighbor node j of node i, V is the node set, for each node s in the neighbor node set S of j, with probability p = f (popul i ×popul j )×σ generates an edge, σ∈(0,1); S4, if the characteristics of the graph structure reach a preset range, the graph structure is reconstructed; The method to reconstruct the graph structure is as follows: The number of common neighbors between nodes is used to determine whether to retain the edges between nodes. The edges whose number of common neighbors ranks in the top a% in the entire network, a∈(0,100) are retained, and the edges that do not meet the conditions are divided with a probability of p∈(0,1); Other reconstruction methods include, but are not limited to: randomly deleting a preset ratio of edges, deleting edges between nodes whose overlap ratio is lower than a preset threshold; The calculation formula of the overlap rate between two nodes i and j is: Where N s is the number of elements in the neighbor set S, is the common neighbor set S of node i and node j i The number of elements of is the neighbor set S of node i i The number of elements contained, is the neighbor set S of node j i The number of elements contained.

2. The method for establishing a dynamic random graph model according to claim 1, characterized in that: After reconstructing the graph structure, steps S2-S3 are repeated on the reconstructed graph structure until the model reaches the expected similarity with the real network in terms of expected network features.

3. The method for establishing a dynamic random graph model according to claim 1, wherein: The method to generate edges between nodes is as follows: For the edge E between node i and node j in the graph structure G ij , using the probability calculation formula for edge E ij Whether there is control: on ij )=f(people i ×people j )×g(dist ij ) Among them, popul i represents the influence factor of node i, and the influence factors of all nodes obey a random variable X that conforms to Poisson distribution or normal distribution, popul i ∈X;popul j represents the influence factor of node j, f and g are two independent activation functions; dist ij represents the normalized distance between node i and node j; Whether there is an edge between node i and node j is mainly determined by the influence factor and distance of the two nodes. f() and g() are activation functions that control the number of edges of the entire graph G. Nodes are distinguished by influence factors that obey Poisson distribution or normal distribution to simulate the power-law distribution of actual networks.

4. The method for establishing a dynamic random graph model according to claim 1, wherein: The method to obtain the neighbor node set of each node is as follows: Define the neighbor set S of node i i , for all edges E ij =1 and the union of the set S of nodes j and node i itself. For each neighbor node set S of node i, it can be expressed by the following formula: S={s|E sj =1,for every j if E ij =1}。 5. The method for establishing a dynamic random graph model according to claim 1, wherein: Determine whether the graph structure needs to be reconstructed based on the characteristics of the graph network; The current random graph network features are combined with the actual network to make a judgment. The graph network features include clustering coefficient and average path length. When the clustering coefficient is greater than the preset value or the average path length is less than the preset value, the graph structure needs to be rebuilt. In an undirected network, the clustering coefficient is defined as: C = 2*n / (k*(k-1)), Where n represents the number of edges between all k neighbors of node v; The average path length is: Where N is the number of network nodes, d ij is the shortest path between nodes i and j.

6. A network analysis method using a dynamic random graph model according to any one of claims 1 to 5, characterized in that: The steps include: Model the real relationship network, and by comparing the characteristics of the real relationship network with the model graph network, determine the development stage of the relationship and predict whether the relationship network will change significantly within a certain period of time; The edges of the model are regarded as paths for information propagation. Individuals with high influence in real life are mapped to nodes with high influence factor scores in the model graph network for modeling, and the impact of information propagation through this individual on the entire graph network is predicted.

7. A system for establishing a dynamic random graph model based on the method according to any one of claims 1 to 5, characterized in that: It includes a creation module, an interaction module and a reconstruction module. The creation module is used to create an initial graph structure and distinguish nodes. The interaction module is used to obtain the neighbor node set of each node and generate graph structure edges between the node and the nodes in the corresponding neighbor node set. The reconstruction module is used to reconstruct the graph structure.

8. A network analysis system using a dynamic random graph model based on the method of claim 6, characterized in that: It includes an information collection unit and a modeling unit. The information collection unit is used to obtain real relationship network information. The output end of the information collection unit is connected to the input end of the modeling unit. The modeling unit models the real relationship network and predicts whether the relationship network will have a large change within a certain period of time, and predicts the impact of the individual propagation information on the entire graph network information propagation.

Images