High-influence information communicator identification method based on circle graph random walk

Abstract

The invention relates to the technical field of graph processing, in particular to a high-influence information communicator identification method based on circle graph random walk. According to the method, the user relation network and the circle relation network are constructed, the circle graph is formed, and the node importance score is calculated in combination with multi-layer random walk, so that high-influence information propagators can be effectively identified. The method has the advantages that the user relation and the circle structure are deeply fused, the node propagation capacity is analyzed from the perspective of a multi-dimensional network, compared with a traditional single network model, the method can better capture the cooperative effect between the complex path of information propagation and the nodes in a social scene, and the accuracy and comprehensiveness of high-influence propagation node recognition are improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of graph processing, and particularly relates to a high-influence information disseminator identification method based on circle graph random walk. BACKGROUND

[0002] High-influence information disseminator identification is a key task in social network analysis, and the target is to accurately identify a high-influence information disseminator node set with strong propagation efficiency in the information diffusion process in a large-scale network. With the continuous expansion of the social network scale and the increasing complexity of the network structure, how to efficiently and accurately identify high-influence information disseminator nodes with extensive propagation potential from a large and heterogeneous node set has become an important challenge in current research.

[0003] Current mainstream methods can be divided into three categories: 1. Methods based on node centrality (such as degree centrality, core number centrality): this kind of method measures the importance of nodes through local or global topological features, but it is difficult to distinguish the high-order interaction differences between nodes, and it is easy to lead to the convergence of node importance scores and the phenomenon of "pseudo-key node redundancy". For example, the method based on degree centrality often makes high-degree nodes gather to form a "rich club" in social networks, resulting in overlapping of propagation influence.

[0004] 2. Dynamic propagation models based on random walk (such as PageRank, HITS, etc.): this kind of method evaluates the influence of nodes by simulating the information diffusion process, but the traditional single-layer random walk is difficult to capture the high-order structure in the network, leading to the failure to identify long-range associations and hidden propagation hubs.

[0005] 3. End-to-end models based on machine learning (such as RCNN, MRCNN, etc.): although this kind of method can adaptively extract network features, its black box characteristics make it difficult to explicitly model the coupling relationship between low-order structure (direct connection between nodes) and circle structure (multi-node cooperation), and the generalization ability is limited in sparse networks or dynamic networks.

[0006] The common defects of the above methods are that they fail to systematically integrate the multi-scale structure information of the network. These problems directly lead to the existence of bottlenecks such as limited coverage, high propagation redundancy, and insufficient robustness in the identification results of high-influence information disseminators.

[0007] Therefore, it is urgent to build an identification method that can uniformly model multi-type structure dependency relationships and accurately identify high-influence propagation nodes, so as to improve the accuracy, stability and application effectiveness of disseminator identification in complex networks. SUMMARY

[0008] In order to avoid and overcome the technical problems existing in the prior art, the application provides a high-influence information disseminator identification method based on circle graph random walk.

[0009] In order to achieve the above-mentioned purpose, the application provides the following technical scheme. A high-influence information disseminator identification method based on circle graph random walk comprises the following identification steps: S1, acquiring a user list and relationship data between users in a current social scene, and constructing a user relationship network without direction and weight by taking the users as original nodes and the relationship between the users as edges; S2, acquiring all basic circles in the user relationship network, and mapping all the basic circles to corresponding circle nodes; S3, adding edges between the circle nodes by using a circle similarity matrix, and constructing a circle relationship network by using the circle nodes and the edges; S4, connecting the circle nodes and each original node contained in the circle nodes by interlayer edges, and calculating the weight of each interlayer edge, so that the user relationship network and the circle relationship network are connected to form a circle graph; S5, calculating the importance score of each original node based on the weight of each interlayer edge in the circle graph; S6, selecting the original nodes with high importance scores to form a high-influence information disseminator cluster. K

[0010] As a further scheme of the application, the construction process of the user relationship network is as follows: S11, calling an open interface of a social platform to acquire a user list and relationship data between users; S12, structurally cleaning the relationship data, and performing network modeling by taking the users as original nodes and the relationship between the users as original edges, to form a user relationship network .

[0011] As a further scheme of the application, the acquisition process of the circle nodes is as follows: S21, selecting the original node with the maximum intermediate degree as a root node, and generating an acyclic tree by using a breadth-first search algorithm ; identifying all original edges not contained in the acyclic tree, and marking the original edges as non-tree edges; for each non-tree edge, finding a unique path of two end points in the acyclic tree, and forming a basic circle with the non-tree edge; and finding all the basic circles in the acyclic tree in this way. ​​​​S22. By directly considering the basic circle as a mapping node, all basic circles are mapped to corresponding nodes, which are called circle nodes.

[0012] As a further solution of the present invention, the process of constructing the circle relationship network is as follows: S31, respectively calculating the circle adjacency similarity between two circle nodes, and establishing a circle adjacency similarity matrix; S32, negate the circle adjacency similarity matrix, and use each element in the circle adjacency similarity matrix after the negation operation as a weight to establish a weighted undirected graph composed of circle nodes; under the premise of not forming a loop, construct a minimum spanning tree connecting all circle nodes in the weighted undirected graph, and select the minimum spanning tree with the smallest sum of weights as the circle relationship network .

[0013] As a further solution of the present invention: the calculation formula of circle adjacency similarity is as follows: ; Where, Indicates the circle nodes; Indicates the circle nodes; express and The circle adjacency similarity between them; express The edge set consisting of all the original edges included, express The edge set consisting of all the original edges included; express and The number of identical original edges contained in ; express and The total number of categories of the original edges contained in .

[0014] As a further solution of the present invention: the weight calculation formula of the inter-layer connection edge is as follows: ; Where, Indicates the connection Original nodes and The weight of the inter-layer edges; express The number of basic circles involved; express The corresponding basic circle; express The number of original nodes in ; express The average number of basic circles formed by all original nodes in .

[0015] As a further solution of the present invention: the calculation process of the importance score is as follows: S51. Formulate and The walking modes between layers include intra-layer random walk and cross-layer random walk; intra-layer random walk is the or Internally, nodes transfer randomly to each other; cross-layer random walks are divided into arrive The upward random transfer and Back to Downward random transfer; S52, calculated The adjacency transfer matrix of : exist In the example, let the random walker be currently located at , then its next step will be to randomly transfer to Neighbor set A circle node in ; Calculate the state transition probability and use the state transition probability as the element The adjacency transfer matrix ; S53, similarly, according to the content in step S52, calculate and obtain The adjacency transfer matrix ; S54. Construct an adjacency up-shift matrix for random up-shift across layers: from arrive When performing an upward random transfer, assume that the random walker is currently located at , then its next step will be to randomly transfer to ; Calculate the uplink transfer probability and use the uplink transfer probability as the element to form arrive The adjacency up-shift matrix ; S55. Construct the adjacent downward shift matrix for cross-layer downward random transfer: from arrive When performing a random downlink transfer, assume that the random walker is currently at , then its next step will be randomly transferred to the corresponding downlink transfer probability Calculate the downlink transition probability, and use the downlink transition probability as an element to form arrive adjacent down-shift matrix ; S56, combining the intra-layer random walk and the cross-layer random walk to form a multi-layer random walk: by fusing the adjacent transition matrix, the adjacent up-shift matrix and the adjacent down-shift matrix, a complete multi-layer random walk transition matrix is obtained: ; S57, set as the visiting probability vector of the time , the visiting probability vector of the time is obtained by the following formula: ; S58, judging whether to converge and outputting the importance score of the original node: if the convergence condition is met, the iteration is terminated, otherwise the iteration is looped; the second norm of the difference between and , represents the convergence threshold; the visiting probability vector of the original node obtained after convergence is taken as the importance score of the original node.

[0016] As a further scheme of the present application: the state transition probability is represented as follows: ; In the formula, represents the state transition probability from randomly to ; represents the set of all edges in .

[0017] As a further scheme of the present application: the up-shift transition probability is represented as follows: ; In the formula, represents the up-shift transition probability from randomly to ; represents the set of all circle nodes connected with through the inter-layer edge; represents the weight of the inter-layer edge connecting and the th circle node .

[0018] As a further scheme of the present application: the down-shift transition probability is represented as follows: ; In the formula,​​ denotes the transition from randomly transitions to downward transition probability; denotes the original node ; denotes the weight of the inter-layer edge connecting and .

[0019] Compared with the prior art, the present application has the following advantages: 1. The present application can effectively identify high-influence information disseminators by constructing user relationship networks, circle relationship networks and forming circle graphs, and combining multi-layer random walk to calculate node importance scores. The advantage is that the user relationship and circle structure are deeply integrated, the node dissemination capacity is analyzed from a multi-dimensional network perspective, the complex path of information dissemination and the synergistic effect between nodes in the social scene are better captured compared with the traditional single network model, and the identification accuracy and comprehensiveness are improved.

[0020] 2. In the process of constructing the user relationship network, the data is obtained by calling the social platform interface and structured cleaning, ensuring the authenticity and standardization of the data. The advantage of this step is to model based on real social relationships, avoid data bias, lay a reliable foundation for subsequent circle structure analysis, make the constructed network accurately reflect the actual interaction relationship of users, and enhance the practicality and operability of the method.

[0021] 3. The breadth-first spanning tree search algorithm is used to identify basic circles and map them to circle nodes. The advantage of this process is that the circle structure in the network is efficiently and systematically extracted. By combining spanning trees and non-tree edges, all basic circles can be found completely and without redundancy, providing accurate basic units for constructing circle relationship networks, so that the circle nodes can truly reflect the cyclic association characteristics in the network topology.

[0022] 4. The circle relationship network is constructed based on the circle adjacency similarity matrix. The intersection and union of the edge set are calculated to quantify the similarity between circles, and the circle nodes are connected by the minimum weight spanning tree. The advantage is that the association degree of the circle is described from the perspective of edge sharing, so that the circle relationship network can reflect the topological similarity and dependency relationship of the circle structure, provide a reasonable inter-circle transition logic for cross-layer random walk, and enhance the expression ability of the model to the network structure.

[0023] 5. The inter-layer edge weight calculation formula considers the number of basic circles participated by the original node, the average number of circle nodes participated by the nodes in the circle, and the number of nodes in the circle. The advantage is that the connection strength of the original node and the circle node is quantified in multiple dimensions. The weight can reflect the participation depth of the node in the circle structure and the scale influence of the circle, so that the cross-layer transition probability calculation is more consistent with the actual influence distribution of the node in the network, and provides key weight support for accurately evaluating the node importance.

[0024] 6. By defining the intra-layer and cross-layer random walk mode, constructing the transition matrix and iteratively calculating the access probability until convergence, the converged probability is taken as the node importance score. The advantage is that the interaction between the original node network and the circle node network is fully considered by means of the multi-layer random walk model, the propagation law of information between different network layers can be dynamically captured, the importance score calculation has the comprehensive consideration of the local features and global structure of the network, and the scientificity and reliability of the high-influence node identification are improved. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The flowchart of the present application.

[0026] Figure 2 The user relationship network in the present application.

[0027] Figure 3 The circle relationship network in the present application.

[0028] Figure 4 The circle graph in the present application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0030] Please refer to Figure 1 , a high-influence information disseminator identification method based on circle graph random walk in the embodiments of the present application includes the following contents: I. Constructing a user relationship network The user list and the relationship data between users in the current social scene are obtained, and a user relationship network without direction and weight is constructed with the user as the original node and the relationship between users as the edge. The specific steps are as follows: S11, call the social platform open interface to obtain user attribute data (user ID, registration time, geographic location) and explicit social relationship data (attention / fan list, friend relationship chain); S12, structure the collected original data (such as removing duplicates, correcting errors, filling missing values), abstract the user as a node, abstract the relationship between users as an edge for network modeling, form a user relationship network , and serve as an input layer. The original node used in this embodiment is , which constitutes a as shown in Figure 2 .

[0031] 2. Get circle nodes

[0032] Obtain all basic circles in the user relationship network and map all basic circles to corresponding circle nodes; the specific steps are as follows: S21. Selection The original node with the largest degree is used as the root node, and a cycle-free tree is generated using the breadth-first spanning tree search algorithm. ; Identification All not included in The original edges in , and marked as non-tree edges; for each non-tree edge, Find the unique path between its two endpoints, which forms a basic cycle with the non-tree edge; All the basic circles in . The basic circles formed are as follows Figure 3 As shown, they are .

[0033] S22. By directly considering the basic circle as a mapping node, all basic circles are mapped to corresponding nodes, which are called circle nodes.

[0034] Will Each basic circle in is mapped to a new node; Mapped as circle nodes , Mapped as circle nodes , Mapped as circle nodes , Mapped as circle nodes , Mapped as circle nodes , specifically Figure 3 shown.

[0035] 3. Build a circle relationship network

[0036] Use the circle similarity matrix to add edges between circle nodes, and use the circle nodes and the edges to build a circle relationship network; the specific steps are as follows: S31. Calculate the circle adjacency similarity between two circle nodes respectively and establish the circle adjacency similarity matrix The calculation formula of circle adjacency similarity is as follows: ; The circle adjacency similarity matrix is ​​specifically expressed as follows, and the circle node orderings corresponding to the horizontal columns are: : ; S32, a negative operation is performed on the circle adjacency similarity matrix, and a weighted undirected graph composed of circle nodes is established with each element in the circle adjacency similarity matrix after the negative operation as a weight; a minimum spanning tree connecting all circle nodes is constructed in the weighted undirected graph on the premise of not forming a loop, and the minimum spanning tree with the smallest sum of weights is selected as the circle relationship network . Specifically as follows: The maximum spanning tree is constructed as a backbone connection, and a negative operation is performed on the circle adjacency similarity matrix The purpose of the operation is to convert the matrix originally indicating that the larger the element value is, the higher the similarity is, into the form indicating that the smaller the element value is, the lower the connection cost is, so as to meet the setting requirement of the edge weight of the minimum spanning tree algorithm.

[0037] Subsequently, the minimum spanning tree algorithm is applied to the weighted undirected graph corresponding to the matrix The core idea of the algorithm is that in a weighted undirected graph, a set of edges is selected to connect all nodes on the premise of not forming a loop, and the sum of the weights of all selected edges is minimized. Since the edge weight here comes from , that is, the negative value of the similarity, the algorithm is equivalent to maximizing the overall circle adjacency similarity in the graph on the premise of maintaining the overall connectivity of the basic circle. In other words, the minimum spanning tree algorithm can preferentially retain the part of edges with the strongest structural similarity among all possible circle pair connection modes, remove redundant connections, and extract a high correlation skeleton.

[0038] Finally, the edges in the minimum spanning tree are added to the circle nodes as their basic connection structure, so as to construct the circle relationship network characterized by high circle adjacency similarity while ensuring connectivity, and as a representation layer.

[0039] Four, constructing a circle graph

[0040] The circle nodes and each original node contained therein are connected by interlayer edges, and the weights of each interlayer edge are calculated, so that the user relationship network and the circle relationship network are connected to form a circle graph; the specific steps are as follows: S41, interlayer edges between and are established, such as mentioned in step S22, mapping to circle node , mapping to circle node , mapping to circle node , mapping to circle node , mapping to circle node . Therefore, ​in the middle of respectively, and the rest of the circle nodes and the original nodes are connected in the same way, thereby forming the inter-layer edges between and The connection of the rest of the circle nodes and the original nodes is the same, thereby forming the inter-layer edges between and The constructed circle graph As shown in Figure 4 , the inter-layer edges are represented by the dashed lines connecting the input layer and the representation layer .

[0041] S42, add weights to the inter-layer edges, and calculate according to the following formula: ; Take the circle node Figure 4 in the representation layer and the original node in the input layer as an example, the calculation process of the inter-layer edge weight is as follows: First, count the number of basic circles in which the original node participates, which is 2; the basic circle to which the circle node belongs is , in which the number of original nodes is 3; and the average number of basic circles in which all nodes , , , and participate is , so the inter-layer edge weight of the circle node and the original node is 0.5926.

[0042] By analogy, the weights of all inter-layer edges in Figure 4 can be calculated.

[0043] Five, calculate the importance score

[0044] Based on the weights of each inter-layer edge in the circle graph, calculate the importance score of each original node; the specific steps are as follows: S51, formulate the walk mode between and , including intra-layer random walk and cross-layer random walk; the intra-layer random walk is the random transfer between nodes within or ; the cross-layer random walk is divided into upward random transfer from to and downward random transfer from back to ; S52, the adjacency transition matrix of is calculated ; In , suppose the random walker is currently located at , then its next step will randomly transfer to a node in the neighborhood set of with equal state transition probability ; the state transition probability is calculated, and the adjacency transition matrix of is constructed with the state transition probability as the element ; The state transition probability is represented as follows: ; ; S53, similarly, the adjacency transition matrix of is calculated according to the content in step S52 ; ; S54, construct the adjacency up-matrix of cross-layer up random transfer: When performing up random transfer from to , suppose the random walker is currently located at , then its next step will randomly transfer to with the corresponding up transfer probability; the up transfer probability is calculated, and the adjacency up-matrix from to is constructed with the up transfer probability as the element ; The up transfer probability is represented as follows: ; S55, construct the adjacency down-matrix of cross-layer down random transfer: When performing down random transfer from to , suppose the random walker is currently located at , then its next step will randomly transfer to with the corresponding down transfer probability; the down transfer probability is calculated, and the adjacency down-matrix from to is constructed with the down transfer probability as the element ; The down transfer probability is represented as follows: ; S56, combine the intra-layer random walk and cross-layer random walk to form a multi-layer random walk: by fusing the adjacency transition matrix, the adjacency up-matrix and the adjacency down-matrix, the complete multi-layer random walk transition matrix is obtained ; S57, suppose is​ Time In the access probability vector of the original node, then its access probability vector at time is represented as follows: ; S58, judge whether to converge and output the importance score of the original node: if the convergence condition is met , the iteration is terminated, otherwise the loop iteration is performed; The second norm of the difference between and is represented as follows: The convergence threshold is represented as follows: 0.001; The access probability vector of the original node obtained after convergence is used as the importance score of the original node.

[0045] Six, obtain a high-impact information propagation node cluster Select the top K original nodes with high importance scores to form a high-impact information propagation node cluster.

[0046] The above is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can make equivalent replacements or changes to the technical solution and the inventive concept of the present application within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for identifying high-influence information disseminators based on random walks in a circle graph, characterized by: The identification steps include: S1. Obtain the user list and user relationship data in the current social scenario, and construct an undirected and unweighted user relationship network with users as original nodes and relationships between users as edges; S2. Obtain all basic circles in the user relationship network and map all basic circles to corresponding circle nodes; S3. Use the circle similarity matrix to add edges between circle nodes, and construct a circle relationship network using circle nodes and corresponding edges; S4. Connect the circle nodes and the original nodes contained therein through inter-layer edges, and calculate the weight of each inter-layer edge, so that the user relationship network and the circle relationship network are connected to each other to form a circle graph; S5. Calculate the importance score of each original node based on the weight of the edges between each layer in the circle graph; S6. Select the top ranking importance scores K The original nodes constitute a high-influence information dissemination node cluster.

2. The method for identifying high-influence information communicators based on random walks in a circle graph according to claim 1 is characterized in that: The process of building a user relationship network is as follows: S11. Call the open interface of the social platform to obtain the user list and the relationship data between users; S12. Structural cleaning of relational data is performed, and network modeling is performed with users as original nodes and relationships between users as original edges to form a user relationship network. .

3. The method for identifying high-influence information communicators based on random walks in a circle graph according to claim 2 is characterized in that: The process of obtaining circle nodes is as follows: S21. Selection The original node with the largest moderate centrality is used as the root node, and the breadth-first spanning tree search algorithm is used to generate an acyclic tree. ; Identification All not included in The original edges in , and marked as non-tree edges; for each non-tree edge, Find the unique path connecting its two endpoints, which forms a basic cycle with the non-tree edge; find All the basic circles in S22. By directly considering the basic circle as a mapping node, all basic circles are mapped to corresponding nodes, which are called circle nodes.

4. The method for identifying high-influence information communicators based on random walks in a circle graph according to claim 3 is characterized in that: The process of building a circle relationship network is as follows: S31, respectively calculating the circle adjacency similarity between two circle nodes, and establishing a circle adjacency similarity matrix; S32, negate the circle adjacency similarity matrix, and use each element in the circle adjacency similarity matrix after the negation operation as a weight to establish a weighted undirected graph composed of circle nodes; under the premise of not forming a loop, construct a minimum spanning tree connecting all circle nodes in the weighted undirected graph, and select the minimum spanning tree with the smallest sum of weights as the circle relationship network .

5. The method for identifying high-influence information communicators based on random walks in a circle graph according to claim 4 is characterized in that: The calculation formula of circle adjacency similarity is as follows: ； Where, Indicates the circle nodes; Indicates the circle nodes; express and The circle adjacency similarity between them; express The edge set consisting of all the original edges included, express The edge set consisting of all the original edges included; express and The number of identical original edges contained in ; express and The total number of categories of the original edges contained in .

6. A method for identifying highly influential information communicators based on random walks in a circle graph according to any one of claims 1 to 5, characterized in that: The weight calculation formula of the inter-layer edge is as follows: ； Where, Indicates connection Hedi Original nodes The weight of the inter-layer edges; express The number of basic circles involved; express The corresponding basic circle; express The number of original nodes in ; express The average number of basic circles formed by all original nodes in .

7. The method for identifying high-influence information communicators based on random walks in a circle graph according to claim 6 is characterized in that: The importance score is calculated as follows: S51. Formulate and The walking modes between layers include intra-layer random walk and cross-layer random walk; intra-layer random walk is the or Internally, nodes transfer randomly to each other; cross-layer random walks are divided into arrive The upward random transfer and Back to Downward random transfer; S52, calculated The adjacency transfer matrix of : exist In the example, let the random walker be currently located at , then its next step will be to randomly transfer to Neighbor set A circle node in , calculate the state transition probability, and use the state transition probability as the element The adjacency transfer matrix ; S53, similarly, according to the content in step S52, calculate and obtain The adjacency transfer matrix ; S54. Construct an adjacency up-shift matrix for random up-shift across layers: from arrive When performing an upward random transfer, assume that the random walker is currently located at , then its next step will be to randomly transfer to , calculate the uplink transition probability, and use the uplink transition probability as the element to form arrive The adjacency up-shift matrix ; S55. Construct the adjacent downward shift matrix for cross-layer downward random transfer: from arrive When performing a random downlink transfer, assume that the random walker is currently at , then its next step will be randomly transferred to the corresponding downlink transfer probability , calculate the downlink transition probability, and use the downlink transition probability as an element to form arrive The adjacency down-shift matrix ; S56. Combine the intra-layer random walk and the cross-layer random walk to form a multi-layer random walk: By fusing the adjacency transfer matrix, the adjacency up-shift matrix, and the adjacency down-shift matrix, we get the complete multi-layer random walk transfer matrix: ; S57, set for time middle The access probability vector of The access probability vector at time It is expressed as follows: ； S58, determine whether convergence and output the importance score of the original node: if the convergence condition is met , the iteration is terminated, otherwise the loop iteration is continued; express and The second-order norm of the difference between represents the convergence threshold; The access probability vector of the original node obtained after convergence is used as the importance score of the original node.

8. The method for identifying high-influence information communicators based on random walks in a circle graph according to claim 7 is characterized in that: The state transition probability is expressed as follows: ； Where, Indicates from Randomly transferred to The state transition probability of express The set of all edges in .

9. The method for identifying high-influence information communicators based on random walks in a circle graph according to claim 8, characterized in that: The uplink transition probability is expressed as follows: ； Where, Indicates from Randomly transferred to The upward transition probability of Indicates that the edges between layers are connected The set of all connected circle nodes; Indicates connection Hedi Circle nodes The weight of the inter-layer edges.

10. The method for identifying high-influence information communicators based on random walks in a circle graph according to claim 9, characterized in that: The downlink transition probability is expressed as follows: ； Where, Indicates from Randomly transferred to The downward transition probability of Indicates the Original nodes; Indicates connection and The weight of the inter-layer edges.

Images