Label propagation community detection method and system based on grouping proximity

Abstract

The invention discloses a label propagation community detection method and system based on grouping closeness, and the method comprises the steps: abstracting a social network into a network topology structure composed of a plurality of nodes, distributing an independent label for each node to represent a community to which the node belongs, and grouping the nodes based on the importance of the nodes and the similarity between the nodes, constructing an initial community structure; the method comprises the following steps of: accessing nodes according to the importance of the nodes from large to small on the basis of the closeness of the nodes and surrounding groups, iteratively updating node labels with the degree greater than 1 in a network, and then updating the labels of the nodes with the degree of 1; the node tags located at the community boundary are corrected by using a predetermined tag updating sequence; and dividing the nodes with the same label in the network into the same community to complete community division. The method does not need to depend on an adjustable parameter or an optimized objective function, successfully eliminates the randomness problem existing in a traditional label propagation algorithm, and remarkably improves the accuracy of a community detection result at the same time.

Description

Technical Field

[0001] The present invention belongs to the technical field of social network analysis, and particularly relates to a label propagation community detection method and system based on group proximity. Background Art

[0002] Social networks reflect the dynamic evolution of real social relationships. Individual users spontaneously construct their own social circles in the network based on various factors such as interests, geographical locations, and social backgrounds, which makes the overall network show obvious modular and hierarchical structures. Through these structures, the aggregation behavior of interpersonal relationships and the path of information dissemination can be intuitively observed. Using social network analysis technology, not only can the implicit aggregation behavior be extracted from the seemingly chaotic network, but also the functional characteristics and information dissemination rules behind the network can be revealed. This provides an important scientific tool for understanding human social behavior and interaction patterns.

[0003] Although the Label Propagation Algorithm (LPA) has been widely used in community detection due to its high efficiency, its disadvantages are also very obvious: the random update process easily leads to unstable results and may fall into local optimal solutions; the simple majority voting mechanism is difficult to accurately capture the fine-grained structures in complex networks; at the same time, the sensitivity of the algorithm to network noise and complex topologies reduces the detection accuracy to a certain extent. Summary of the Invention

[0004] Object of the Invention: To solve the problems of instability and low accuracy existing in the existing Label Propagation Algorithm (LPA) in community detection, the present invention proposes a label propagation community detection method and system based on group proximity, so as to achieve more accurate community division of social networks.

[0005] Technical Solution: To achieve the above object, a label propagation community detection method based on group proximity according to the present invention includes the following steps:

[0006] S1. Abstract the social network into a network topology structure composed of multiple nodes, and assign an independent label to each node, where the label is used to represent the community to which the node belongs;

[0007] S2. Group the nodes in the network of S1 based on the importance of the nodes and the similarity between the nodes to construct an initial community structure, and update the node labels with a degree greater than 1 in descending order of node importance during the grouping process;

[0008] S3. Based on the initial community structure obtained in S2, use the proximity between the node and its surrounding groups and the label update order obtained in S2 when updating the node labels to iteratively update the node labels with a degree greater than 1 in the network, and then update the labels of the nodes with a degree of 1;

[0009] S4. Use the label update order obtained in S2 when updating node labels to correct the node labels at the community boundaries in the community structure obtained in S3;

[0010] S5. Divide the nodes with the same label in the network corrected in S4 into the same community to complete the community division.

[0011] Among them, the method for constructing the initial community structure in S2 is as follows:

[0012] For a selected node with a degree greater than 1, select the neighbor nodes that are most similar to the selected node and whose importance is not less than that of the selected node to construct a group; if the importance of the most similar neighbor node is greater than or equal to the importance of the selected node, the selected node obtains the label of the neighbor node, otherwise the label of the selected node remains unchanged; if the similarity between the selected node and its most similar neighbor is zero, the selected node will obtain the label of the neighbor with the highest degree among the neighbor nodes; the node label update order is carried out in descending order of importance to ensure that the labels of high-importance nodes are propagated first; when the label update of all nodes with a degree greater than 1 is completed, the initial community structure is formed accordingly, and this structure is composed of different groups, denoted as Group1, Group2 represent different groups, v i represents a node, represents the degree of node v i , that is, the number of neighbor nodes of node v i , and V represents the set of nodes.

[0013] Among them, the method for calculating the importance of the node is as follows:

[0014]

[0015] Among them, I(v i ) represents the importance value of node v i , represents the degree of node v i , that is, the number of neighbor nodes of node v i , represents the number of triangles formed by node v i and its neighbor nodes;

[0016] Further normalize the node importance to [0, 1]:

[0017]

[0018] Among them, IM(v i ) represents the normalized importance value of node v i , min l∈V I(vl ) and max k∈V I(v k ) respectively represent the minimum and maximum values among the importance levels of all nodes in the node set V.

[0019] Among them, for the node v i and v j The similarity calculation method between them is as follows:

[0020]

[0021] Among them, Sim(v i , v j ) represents the similarity value between adjacent nodes v i and v j . N(v i ) and N(v j ) respectively represent the neighbor node sets of v i and v j . N(v i ) ∩ N(v j ) represents the common neighbor set of node v i and node v j .

[0022] Among them, in S3, according to the label update order, after iteratively updating the labels of nodes with a degree greater than 1 in the network, the method for updating the labels of nodes with a degree of 1 is as follows: Traverse all nodes with a degree greater than 1 in the network according to the label update order, and update the label of the selected node to the label of the group with the highest proximity based on the proximity between the selected node and each surrounding group, until the labels of all nodes with a degree greater than 1 no longer change; updating the label of a node with a degree of 1 is to update the label of the node to the label of its unique neighbor.

[0023] Among them, the calculation method for the proximity between the node and the surrounding groups is as follows:

[0024]

[0025] [[ID=**54**]]Among them, pro(Group h , v i ) represents the proximity between group Group h and the selected node v i . v j is the intersection of the neighbor node set N(v i ) of node v i and Group h . v j has the label of group Group h . IM(v j ) represents the normalized node vj The importance, Sim(v i , v j ) represents the similarity between nodes v i and v j , where v z is the neighbor node of node v i , and v z may come from different groups.

[0026] Among them, the method for correcting the node labels located at the community boundary in S4 is as follows: for each selected node located at the community boundary, update the selected node label to the group label with the highest frequency of occurrence among the neighbor nodes of the selected node. If there are multiple group labels with the same maximum frequency, calculate the membership degree of the node for each group, and update the selected label to the label of the group with the highest membership degree until the labels of each node no longer change and stop the correction.

[0027] Among them, the calculation method of the membership degree of the node to the group is as follows:

[0028]

[0029] Among them, Group h is one of the groups with the highest frequency. IM(v j ) represents the importance of node v j after normalization. v j is the neighbor node set N(v i ) of node v i intersected with Group h . This calculation method measures the membership degree of node v i to group Group h by aggregating the importance indicators of the nodes belonging to group Group i among the neighbors of node v h .

[0030] A label propagation community detection system based on group proximity according to the present invention includes a network initialization module, an initial community division module based on node importance, a label selection module based on group proximity, a label correction module, and a community division module.

[0031] The network initialization module: abstracts the social network into a network topology structure composed of multiple nodes, and assigns an independent label to each node, and this label is used to represent the community to which the node belongs;

[0032] The initial community division module based on node importance: groups nodes in the social network based on node importance and similarity between nodes to construct an initial community structure. During the grouping process, the node labels with a degree greater than 1 are updated in descending order of node importance.

[0033] The group affinity-based label selection module: based on the affinity between the node and its surrounding groups, uses the label propagation order obtained by the initial community division module when updating the node label, iteratively updates the labels of nodes with a degree greater than 1 in the network, and then updates the labels of nodes with a degree of 1;

[0034] The label correction module uses the label propagation order obtained by the initial community division module when updating the node labels to correct the node labels at the community boundary in the community structure obtained by the label selection module;

[0035] The community division module divides the nodes with the same label in the network into the same community to complete the community division.

[0036] Beneficial effects: The present invention has the following advantages:

[0037] 1. This paper proposes a label propagation community detection method based on group affinity. Compared with existing community detection technologies, this method has higher robustness and practicality. It abandons the reliance on adjustable parameters or predefined objective functions and cleverly integrates the importance of nodes in social networks with the similarity information between nodes. It rationally groups nodes in the network and accurately quantifies the affinity relationship between nodes and each group. Through this mechanism, the accuracy of community division is significantly improved, providing an efficient and reliable solution for community detection in complex network structures.

[0038] 2. To address the problem that traditional label propagation algorithms are easily affected by the order of label updates, resulting in unstable results, this paper proposes an improved strategy: by prioritizing access to nodes with high importance and introducing a proximity indicator between nodes and groups, the iterative update process of labels is accurately guided. This method effectively enhances the stability of the label propagation process and significantly improves the reliability of community detection results.

[0039] 3. The label propagation community detection method based on group proximity disclosed in the present invention can continuously analyze and mine the community group structure in a multi-node network, thereby deeply revealing the intrinsic correspondence and deep-level characteristics between network topology and function. This method has high stability and accuracy, and can effectively identify important group structures in the network, providing strong support for subsequent group recommendations and interest mining, thus showing significant application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is the flowchart of the present invention;

[0041] Figure 2 is an example diagram of a social network provided by an embodiment of the present invention;

[0042] Figure 3 is an example diagram of community division of a social network provided by an embodiment of the present invention;

[0043] Figures 4 - 6 is based on the Figure 3 embodiment for the flowchart of community division;

[0044] Figure 7 is a schematic structural diagram of a label propagation community detection system based on group proximity of the present invention. Detailed implementation manners

[0045] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Embodiment 1

[0047] The present invention provides a label propagation community detection method based on group proximity, including three processes: initial community division based on node importance, label selection based on group proximity, and label correction process.

[0048] As Figure 1 , Figure 4 , Figure 5 , Figure 6 shown, an embodiment of the present invention provides a label propagation community detection method based on group proximity, including the following steps:

[0049] S1. Abstract the social network into a network topology structure composed of multiple nodes, and assign an independent label to each node, where the label is used to represent the community to which the node belongs;

[0050] Model the Figure 2 social network as a network topology structure composed of multiple nodes as shown in Figure 3 . Generally, concepts in graph theory are used to represent the social network, where nodes and edges represent users and social relationships between users respectively.

[0051] S2. Group the nodes in the network of S1 based on the importance of the nodes and the similarity between the nodes to construct an initial community structure, and update the node labels with a degree greater than 1 in descending order of node importance during the grouping process; the specific process is as follows:

[0052] First, calculate the importance of nodes and the similarity between nodes. For nodes with degree 1, they do not participate in grouping because the label of such a node is affected by its only neighbor; for selected nodes with degree greater than 1, select neighbor nodes that are most similar to the selected node and have importance not less than that of the selected node to construct a group. If the importance of the most similar neighbor node is greater than or equal to the importance of the selected node, then the selected node will obtain the label of that neighbor node; otherwise, the label of the selected node remains unchanged. Specifically, if the similarity between the selected node and its most similar neighbor is zero, then the selected node will obtain the label of the neighbor with the highest degree among the neighbor nodes. The order of updating node labels is in descending order of importance, ensuring that the labels of high-importance nodes are propagated first. When the label update of all nodes with degree greater than 1 is completed, the initial community structure is formed accordingly. This structure consists of different groups, denoted as

[0053] For a given undirected and unweighted social network G=(V, E), in G, the importance of a node (representing an individual) may reflect the influence of this person in social interactions or the control ability of this node over the information dissemination to other nodes. The greater the degree of a node and the more triangles it participates in, the more important the node is. The calculation formula for node importance is as follows:

[0054]

[0055] where, I(v i ) represents the importance of node v i ; represents the degree of node v i , that is, the number of neighbor nodes of node v i ; represents the number of triangles formed by node v i and its neighbor nodes. To avoid a large difference in importance between different nodes and reflect the influence of importance on label propagation, the node importance is normalized to [0, 1]. The formula is as follows:

[0056]

[0057] where, min l∈V I(v l ) and max k∈V I(v k ) respectively represent the minimum and maximum values among the importance of all nodes in the node set V.

[0058] In a social network, it is very likely that the neighbor nodes of any given node are also connected to each other. These nodes often belong to the same community because they communicate closely. For two adjacent nodes v i and v j , v iand v j The similarity calculation formula between them is as follows:

[0059]

[0060] where v i and v j are two adjacent nodes, N(v i ) and N(v j ) represent the neighbor node sets of v i and v j respectively, and N(v i ) ∩ N(v j ) represents the common neighbor set of node v i and node v j .

[0061] Figure 4 Step S2 in shows the initial community partitioning process based on node importance. Through the node importance and node similarity metrics, for each node with a degree greater than 1, it will select the neighbor node that is most similar to itself and has an importance not lower than its own for grouping, and then update the node labels in the order of descending node importance, thereby obtaining a reliable initial community.

[0062] S3. Based on the initial community structure obtained in step S2 and following the label propagation order obtained when updating node labels in step S2, combined with the proximity index between nodes and groups, iteratively update the labels of nodes with a degree greater than 1 in the network until all node labels no longer change, and then update the labels of nodes with a degree of 1; the specific process is as follows: following the label propagation order obtained in S2, traverse all nodes with a degree greater than 1 in the network, and update the selected node label to the label of the group with the highest proximity according to the proximity between the selected node and each surrounding group; this process is executed in a loop until the labels of all nodes with a degree greater than 1 no longer change. Finally, for nodes with a degree of 1, directly update their labels to the labels of their unique neighbors.

[0063] The group proximity reflects the influence exerted by the nodes within the group on a given node, and combines node importance and the similarity between neighbor nodes to calculate the magnitude of the influence. Generally speaking, neighbor nodes with greater importance and similarity have a greater influence on a given node. The calculation formula for the proximity between a node and a group is as follows:

[0064]

[0065] where pro(Group h ,v i ) represents the proximity between group Group h and the selected node v iProximity to v j For node v i The set of neighbor nodes N(v i ) and the intersection with Group h v j Has grouping Group h The label of, IM(v j ) represents the importance of node v j Sim(v i , v j ) represents the similarity between node v i and v j v z For node v i The neighbor node of v z May come from different groupings.

[0066] Take Figure 5 S3 in as an example to illustrate the specific implementation of label selection based on grouping proximity in detail. First, follow the label propagation order obtained from S2, and v4 is preferentially processed due to its high importance index. When updating the label of v4, analyze the distribution of the neighbor set {v1, v2, v3, v5, v6, v7} of v4: among them, v1, v2, v3 belong to grouping Group1, while v5, v6 belong to Group2. Since v7 is not grouped, it is not included in the proximity calculation. Through the proximity formula calculation, the quantization results of pro(Group1, v4) = 1.51 and pro(Group2, v4) = 0.84 are obtained. Since Group2 shows stronger proximity, theoretically the label of v4 should be updated. However, it should be noted that v4 originally belongs to the Group2 grouping, so in fact its label remains unchanged. This calculation process verifies the rationality of this method, indicating that nodes are correctly divided into the grouping with the greatest proximity. After processing v4, continue to update the labels of the remaining nodes in the predetermined order. This iterative process continues until the node labels no longer change. At this time, the community structure has basically taken shape. Finally, process special nodes with degree 1, such as Figure 5 The v7 node shown, such nodes directly inherit the label of their only neighbor v4 due to limited connection information.

[0067] S4. Follow the label propagation order obtained in S2 and correct the labels of the nodes located at the community boundaries in the community structure obtained in S3. For each selected node, update the selected node label to the group label with the highest frequency among the neighbor nodes of the node. If there are multiple group labels with the same maximum frequency, calculate the membership degree of the node for each group, and update the label to the label of the group with the highest membership degree. This process will be repeated until the labels of all nodes no longer change. The membership degree formula of the node and the group is as follows:

[0068]

[0069] where Group h is one of the groups with the highest frequency. This formula measures the membership degree of node v i to group Group h by aggregating the importance indicators of the nodes belonging to group Group i among the neighbors of node v h .

[0070] As shown in S4 of Figure 6 , the operation process of this stage is further illustrated. According to the established node update order, first visit v4. After analysis and judgment, v4 is not located at the community boundary, so there is no need to specially process its label. Similarly, the label of v 11 also does not need to be processed. Then further visit the neighbor nodes of v8. After statistical analysis, it is found that the label of v8 is the same as the group label with the highest frequency among its neighbor nodes, so the label of v8 remains unchanged. Subsequently, visit the remaining nodes in sequence, continuously check the nodes located at the community boundaries, and judge the rationality of their labels. In the entire dynamic process of inspection and correction, as the iteration continues, the labels of the nodes in the network no longer change. At this time, stop the iteration.

[0071] S5. Divide the nodes with the same label in the network corrected in S4 into the same community, such as the two communities shown in S5 of Figure 6 .

[0072] Embodiment 2

[0073] As Figure 7 shown, an embodiment of the present invention provides a label propagation community detection system based on group proximity, including a network initialization module, an initial community division module based on node importance, a label selection module based on group proximity, a label correction module, and a community division module.

[0074] The network initialization module: abstracts the social network as a network topology structure composed of multiple nodes, and assigns an independent label to each node, which is used to represent the community to which the node belongs;

[0075] The initial community division module based on node importance: groups the nodes in the social network based on the importance of the nodes and the similarity between the nodes to construct an initial community structure. During the grouping process, the labels of the nodes with a degree greater than 1 are updated in descending order of node importance in sequence;

[0076] The label selection module based on grouping proximity: based on the proximity of the node to its surrounding groups, and using the label propagation order obtained by the initial community division module when updating the node labels, iteratively updates the labels of the nodes with a degree greater than 1 in the network, and then updates the labels of the nodes with a degree of 1;

[0077] The label correction module: uses the label propagation order obtained by the initial community division module when updating the node labels to correct the labels of the nodes located at the community boundary in the community structure obtained by the label selection module;

[0078] The community division module: divides the nodes with the same label in the network into the same community to complete the community division.

[0079] The initial community division module based on node importance performs the following operations:

[0080] Calculate the importance of the nodes and the similarity between the nodes. Based on these two metrics, group the nodes in the network; update the node labels in descending order of node importance to construct the initial community. For the nodes with a degree of 1, they do not participate in the grouping process. For the selected nodes with a degree greater than 1, select the neighbor nodes that are most similar to the selected node and whose importance is not less than that of the selected node to construct a group. If the importance of the most similar neighbor node is greater than or equal to the importance of the selected node, then the selected node will obtain the label of the neighbor node, otherwise the label of the selected node remains unchanged. In particular, if the similarity between the selected node and its most similar neighbor is zero, then the selected node will obtain the label of the neighbor with the highest degree among the neighbor nodes. The node label update order is carried out in descending order of importance in sequence to ensure that the labels of the high-importance nodes are propagated first. When the label update of all nodes with a degree greater than 1 is completed, the initial community structure is formed accordingly.

[0081] The label selection module based on grouping proximity performs the following operations:

[0082] Traverse all nodes in the network with degrees greater than 1 according to the predetermined node update order, and update the label of the target node to the label of the group with the highest proximity according to the proximity between the target node and each surrounding group. This process is executed in a loop until the labels of all nodes with degrees greater than 1 no longer change. Finally, for nodes with degree 1, directly update their labels to the labels of their unique neighbors.

[0083] The label correction module performs the following operations:

[0084] According to the predetermined node update order, correct the labels of the nodes located at the community boundary in the community structure. For each selected node, update its label to the label of the group with the highest frequency of occurrence among its neighbor nodes. If there are multiple labels with the same maximum frequency, calculate the membership degree of the node for each group, and update the label to the label of the group with the highest membership degree. This process will be carried out in a loop until the label of each node no longer changes.

Claims

1. A label propagation community detection method based on grouping proximity, characterized in that It includes the following steps: S1. Abstract the social network into a network topology composed of multiple nodes, and assign an independent label to each node, where the label is used to represent the community to which the node belongs; S2. Group the nodes in the network in S1 based on the importance of the nodes and the similarity between the nodes to construct an initial community structure. During the grouping process, the labels of the nodes with a degree greater than 1 are updated in descending order of node importance; S3. Based on the initial community structure obtained in S2, use the proximity between the node and its surrounding groups and the label update order obtained in S2 when updating the node labels to iteratively update the labels of the nodes with a degree greater than 1 in the network, and then update the labels of the nodes with a degree of 1; S4. Use the label update order obtained in S2 to correct the labels of the nodes located at the community boundaries in the community structure obtained in S3; S5. Divide the nodes with the same label in the network corrected in S4 into the same community to complete the community division.

2. The method for community detection by label propagation based on group proximity according to claim 1, wherein The method for constructing the initial community structure described in S2 is: For the selected node with a degree greater than 1, select the neighbor nodes that are most similar to the selected node and whose importance is not less than that of the selected node to construct a group; if the importance of the most similar neighbor node is greater than or equal to the importance of the selected node, the selected node obtains the label of the neighbor node, otherwise the label of the selected node remains unchanged; If the similarity between the selected node and the neighbor that is most similar to the selected node is zero, the selected node will obtain the label of the neighbor with the highest degree among the neighbor nodes; The node label update order is carried out in descending order of importance to ensure the priority propagation of labels of high-importance nodes; when the label updates of all nodes with a degree greater than 1 are completed, the initial community structure is formed accordingly. This structure consists of different groups, denoted as Group1 and Group2 represent different groups, and v i represents a node, denotes the degree of node v i i.e., the number of neighbor nodes of node v i and V represents the set of nodes.

3. The method for community detection by label propagation based on group proximity according to claim 2, wherein The method for calculating the importance of the node is: Among them, I(v i ) represents the importance value of node v i . represents the degree of node v i , that is, the number of neighbor nodes of node v i . represents the number of triangles formed by node v i and its neighbor nodes; Further normalize the node importance to [0, 1]: Among them, IM(v i ) represents the importance value of the normalized node v i , min l∈V I(v l ) and max k∈V I(v k ) respectively represent the minimum and maximum values among the importance values of all nodes in the node set V.

4. The method for community detection of label propagation based on grouping proximity according to claim 2, characterized in that The described node v i and v j Similarity calculation method between: Among them, Sim(v i , v j ) represents the similarity value between adjacent nodes v i and v j . N(v i ) and N(v j ) respectively represent the sets of neighbor nodes of v i and v j . N(v i ) ∩ N(v j ) represents the set of common neighbor nodes of node v i and node v j .

5. The method for community detection by label propagation based on grouping proximity according to claim 1, characterized in that, The method for iteratively updating the labels of the nodes with a degree greater than 1 in the network according to the label update order in S3 and then updating the labels of the nodes with a degree of 1 is: traverse all the nodes with a degree greater than 1 in the network according to the label update order, and update the selected node label to the label of the group with the highest proximity according to the proximity between the selected node and each surrounding group until the labels of all the nodes with a degree greater than 1 no longer change; the label update for the node with a degree of 1 is to update the node label to the label of the only neighbor.

6. The method for label propagation community detection based on group proximity according to claim 5, wherein The method for calculating the proximity between the node and the surrounding groups is: Among them, pro(Group h , v i ) represents the proximity of the grouping Group h and the selected node v i . v j is the node v i . The neighbor node set N(v i ) is the intersection of Group h . v j has the label of the grouping Group h . IM(v j ) represents the importance of the normalized node v j . Sim(v i , v j ) represents the similarity between the node v i and v j . v z is the neighbor node of the node v i . v z may come from different groupings.

7. The method for community detection of label propagation based on group proximity according to claim 1, characterized in that The method for correcting the labels of the nodes located at the community boundaries described in S3 in S4 is: for each selected node located at the community boundary, update the selected node label to the group label with the highest occurrence frequency among the neighbor nodes of the node. If there are multiple group labels with the same maximum frequency, calculate the membership degree of the node for each group, and update the selected label to the label of the group with the highest membership degree until the label of each node no longer changes to stop the correction.

8. The method for community detection of label propagation based on group proximity according to claim 7, characterized in that The method for calculating the membership degree of the node to the group is: Among them, Group h is one of the groups with the highest frequency. IM(v j ) represents the importance of node v j after normalization. v j is the set of neighbor nodes N(v i ) of node v i intersected with Group h . This calculation method measures the membership degree of node v i to the group Group h by aggregating the importance indicators of the nodes in the neighborhood of node v i that belong to the group Group h .

9. A label propagation community detection system based on grouping proximity, characterized in that, It includes a network initialization module, an initial community division module based on node importance, a label selection module based on group proximity, a label correction module, and a community division module. The network initialization module: abstract the social network into a network topology composed of multiple nodes, and assign an independent label to each node, where the label is used to represent the community to which the node belongs; The initial community division module based on node importance: Group the nodes in the social network based on the importance of the nodes and the similarity between nodes to construct an initial community structure. During the grouping process, the node labels with a degree greater than 1 are updated in descending order of node importance one by one; The label selection module based on grouping proximity: Based on the proximity of a node to its surrounding groups, and using the label propagation order obtained by the initial community division module when updating node labels, the node labels with a degree greater than 1 in the network are iteratively updated, and then the labels of the nodes with a degree of 1 are updated; The label correction module: Using the label propagation order obtained by the initial community division module when updating node labels, correct the node labels located at the community boundary in the community structure obtained by the label selection module; The community division module: Divide the nodes with the same label in the network into the same community to complete the community division.

Images