Modularity constraint based symmetric non-negative matrix three-factor decomposition community detection method and system

By introducing modularity-constrained symmetric nonnegative matrix three-factor decomposition into the community detection method, the problems of high computational complexity and insufficient recognition accuracy of existing methods in large-scale networks are solved, and efficient and stable community structure recognition and continuous membership matrix representation are achieved.

CN122364625APending Publication Date: 2026-07-10BEIJING JIAOTONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610454548.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-08
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Existing community detection methods have high computational complexity in large-scale networks, making it difficult to accurately identify community structures at both local and global scales. Furthermore, they lack continuous membership matrix representations, resulting in insufficient recognition accuracy and stability.

Method used

A three-factor decomposition method for symmetric nonnegative matrices based on modularity constraints is constructed. By integrating the adjacency matrix reconstruction error and modularity maximization in the same objective function through a joint optimization framework, a consistency regularization mechanism is adopted to coordinate the community representation matrix, and a multiplication update rule that satisfies the KKT optimality condition is designed.

Benefits of technology

It improves the accuracy and stability of community detection, enhances the expressive power of global modular features, realizes continuous membership matrix representation, and reduces computational complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122364625A_ABST
    Figure CN122364625A_ABST
Patent Text Reader

Abstract

This invention provides a community detection method and system based on symmetric nonnegative matrix three-factor decomposition with modularity constraints, belonging to the field of network graph data analysis technology. This invention constructs a unified joint optimization framework that simultaneously integrates adjacency matrix reconstruction error and modularity maximization objective within the same objective function. A consistency regularization mechanism is designed to achieve coupling and coordination between different community representation matrices, thereby enhancing the expressive power of global modular features while preserving local network structural information. By constructing multiplicative update rules that satisfy nonnegativity constraints and combining Lagrange function derivation and KKT optimality condition verification, the algorithm is guaranteed to have theoretical stability and optimality, thus improving the accuracy, stability, and interpretability of community detection results. Furthermore, while ensuring computational feasibility, the applicability and engineering generalization of the model are enhanced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of network graph data analysis technology, specifically to a community detection method and system based on modularity constraints for symmetric nonnegative matrix three-factor decomposition. Background Technology

[0002] Community detection in complex networks is a core problem in graph data analysis and complex systems research. Its basic task is to identify subgroups of nodes with tightly connected internal structures and relatively sparse external connections within a network structure composed of nodes and edges. With the rapid growth of large-scale graph data such as social networks, bioinformatics networks, transportation networks, and financial networks, community detection technology plays a crucial role in recommendation systems, anomaly detection, and group behavior analysis. Mathematically, networks are typically represented as adjacency matrices, where matrix elements represent the connections between nodes. Around this matrix structure, three main technical approaches have gradually emerged in academia: one is optimization methods based on maximizing modularity; another is graph partitioning methods based on spectral decomposition; and the third is low-dimensional embedding methods based on probabilistic models or matrix decomposition.

[0003] In the aforementioned technical system, modularity, proposed by Newman et al., serves as a crucial indicator for evaluating the quality of community partitioning. Its core idea is to characterize the saliency of community structure by comparing the connectivity differences between real-world and random networks. By constructing a modularity matrix, the community detection problem can be transformed into a maximization problem. Existing implementations typically employ spectral decomposition, greedy heuristic search, or semidefinite programming methods to optimize modularity. While these methods can effectively reflect the global topology of the network, their optimization processes often involve discrete label assignment, making it difficult to generate continuous membership representations. Furthermore, they suffer from high computational complexity in large-scale networks and are difficult to unify with other embedding models.

[0004] Meanwhile, Nonnegative Matrix Factorization (NMF) and its extended models have been increasingly applied to community detection problems. NMF decomposes a nonnegative matrix X into the product of two or three low-dimensional nonnegative matrices, enabling low-dimensional representation of the network and community membership modeling. Among these, Asymmetric Nonnegative Matrix Tri-Factorization (ANMF) assumes that the network's adjacency matrix satisfies a generative model and solves for parameters by minimizing the Frobenius norm. This method offers advantages such as strong interpretability, the ability to output soft membership relationships, and the guarantee of nonnegativity constraints through multiplicative update rules. However, the optimization objective of these methods primarily revolves around reconstruction error; their essence is a matrix approximation problem, without explicitly considering the global modular structure of the network. Therefore, in some structurally complex or weakly communityd networks, they are prone to local optima or insufficient structural identification.

[0005] Existing research has also attempted to incorporate the concept of modularity into the matrix factorization framework, such as adding a regularization term related to modularity to the objective function, or indirectly enhancing community structure information by imposing constraints on the membership matrix. However, these methods typically only treat modularity as an auxiliary constraint term and do not theoretically establish a unified joint optimization framework, nor do they systematically model the consistency relationship between the two community representation matrices.

[0006] Furthermore, most existing schemes are based on empirically constructed update rules and lack rigorous derivation of the Lagrangian function and verification of KKT optimality conditions, thus still having shortcomings in terms of theoretical convergence and optimality guarantees.

[0007] Existing community detection methods based on nonnegative matrix factorization or asymmetric nonnegative matrix three-factor factorization can obtain low-dimensional embedding representations of the network by minimizing the adjacency matrix reconstruction error. However, their optimization objectives are mainly focused on matrix approximation accuracy, limiting their ability to characterize the overall network topology. In particular, they fail to explicitly consider the modularity of community partitioning at the global scale, thus easily leading to decreased recognition accuracy or structural bias in complex network structures or ambiguous community boundaries. Furthermore, these methods typically rely solely on reconstruction error as the only optimization criterion, lacking an intrinsic connection with classic modularity evaluation metrics. This makes it theoretically difficult to guarantee a high modularity value for the obtained community membership matrix, potentially causing the results to deviate from the actual community quality in terms of evaluation metrics. On the other hand, while pure modularity maximization methods can optimize community partitioning from the perspective of overall structure, their implementation often employs discrete label assignment or spectral decomposition techniques, making it difficult to output continuous soft membership matrix representations. This limits their application in subsequent embedding analysis or machine learning tasks, and their computational complexity is high in large-scale networks, resulting in efficiency and scalability issues. Existing schemes that attempt to combine modularity with matrix factorization mostly simply add modularity terms to the objective function. They lack systematic modeling of the consistency between different community representations and have not formed a rigorous joint optimization derivation framework. At the algorithm level, they usually lack theoretical proofs of nonnegativity constraints and optimality conditions, thus resulting in uncertainties in convergence and stability. Summary of the Invention

[0008] The purpose of this invention is to provide a community detection method and system based on modularity constraints for symmetric nonnegative matrix three-factor decomposition, so as to solve at least one of the technical problems existing in the background art.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] In a first aspect, the present invention provides a community detection method for symmetric nonnegative matrix three-factor factorization based on modularity constraints, comprising:

[0011] S1: Read the raw network data and construct the input matrix X based on the set of nodes and the set of edges;

[0012] S2: Calculate the node degree and the total number of edges in the network based on the input matrix X, and construct the modularity matrix;

[0013] S3: For matrices W, H and... Perform non-negative initialization, where W represents the node-to-community membership matrix, and H represents the inter-community relationship matrix. This represents the auxiliary membership matrix used for modularity optimization;

[0014] S4: Construct a joint optimization objective function, and under non-negativity constraints, employ an alternating iterative optimization strategy to sequentially update W, H, and H according to a preset update rule. ;

[0015] S5: After each update, Perform row normalization so that the sum of the elements in each row is 1;

[0016] S6: Determine if the convergence condition is met; if it is, output the final community membership matrix W; otherwise, return to step S4 to continue iterating.

[0017] As a further limitation of the first aspect of the present invention, the original network data includes node set and edge set information; the read edge data is converted into an adjacency matrix. In particular, for unauthorized networks, when a node With nodes When there is a connection, ,otherwise For weighted networks, Represents a node With nodes The connection weights between them; where the matrix elements This indicates whether there is a connection or connection weight between node i and node j.

[0018] As a further limitation of the first aspect of the invention, by calculating the degree of each node... and the total number of network edges Construct the modularity matrix ,in ,in The modularity matrix is ​​used to characterize the difference between the actual connection structure of the network and the random network structure.

[0019] As a further limitation of the first aspect of the present invention, the constructed objective function is:

[0020] Where μ is the regularization parameter; the first term ( ) is used to reconstruct the adjacency matrix, thereby preserving the local connectivity structure of the network; the second term ( ) is used to maximize the modularity metric to enhance the global community structure characteristics; the third item ( ) used to constrain W and The consistency between them makes the two community representations tend to be coordinated; the fourth item ( This is used to enhance the discriminative power of the W matrix, making community divisions clearer; during the optimization process, constraints include... , , as well as This ensures the non-negativity of the matrix and The probabilistic interpretation of a matrix.

[0021] As a further limitation of the first aspect of the present invention, in each iteration, two variables are fixed and treated as constants, and the remaining variables are optimized and updated; the update process is derived by constructing a Lagrange function and introducing non-negative Lagrange multipliers, and finally obtaining a multiplication update rule that satisfies the KKT optimality condition.

[0022] As a further limitation of the first aspect of the invention, in updating When updating the matrix, a multiplication update is performed based on the numerator and denominator expressions containing adjacency matrix items and consistency regularization terms, and the update result is multiplied by 1 / 4 to ensure numerical stability; during the update... When updating the matrix, it is done according to the multiplication rules of standard nonnegative matrix decomposition; during the update... When updating a matrix, both modularity matrix items and consistency items are considered, and updates are performed using a multiplication update rule with an exponent of 1 / 2.

[0023] As a further limitation of the first aspect of the present invention, after each update, the system performs... The matrix is ​​row normalized so that the sum of the elements in each row equals 1, thus satisfying the constraints.

[0024] In a second aspect, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the community detection method for symmetric nonnegative matrix three-factor decomposition based on modularity constraints as described in the first aspect.

[0025] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the community detection method of symmetric nonnegative matrix three-factor decomposition based on modularity constraints as described in the first aspect.

[0026] Fourthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the community detection method of symmetric nonnegative matrix three-factor decomposition based on modularity constraints as described in the first aspect.

[0027] Terminology Explanation: A "complex network" refers to a graph-structured data model consisting of nodes and edges, where nodes represent entities and edges represent relationships or connections between entities. An "adjacency matrix" is a matrix representation used to represent network structures, denoted as... , where matrix elements This represents the connection or weight between node i and node j. "Nonnegative Matrix Factorization" (NMF) refers to the process of decomposing a matrix into a product of several low-dimensional nonnegative matrices, provided all elements are nonnegative. "Asymmetric Nonnegative Matrix Tri-Factorization" (ANMF) refers to representing a matrix as... A three-factor decomposition model of the form where each factor matrix is ​​a non-negative matrix. "Modularity" is a metric for measuring the quality of network community partitioning, reflecting the degree of difference between the actual network connection structure and the random connection model. The "Frobenius norm" is the square root of the sum of the squares of all elements of a matrix, denoted as . The "trace operation" refers to the sum of the diagonal elements of a matrix, denoted as . The Karush-Kuhn-Tucker conditions (KKT conditions) are optimality criteria used to constrain optimization problems. A membership matrix is ​​a matrix representing the degree of membership of a node to different communities, where each element represents the weight or probability of a node belonging to a particular community.

[0028] The beneficial effects of this invention are as follows: It constructs a unified joint optimization framework that simultaneously integrates the adjacency matrix reconstruction error and the modularity maximization objective within the same objective function. A consistency regularization mechanism is designed to achieve coupling and coordination between different community representation matrices, thereby enhancing the expressive power of global modular features while preserving local network structural information. By constructing multiplicative update rules that satisfy non-negativity constraints and combining the derivation of the Lagrange function with KKT optimality verification, the algorithm is guaranteed to have theoretical stability and optimality, thus improving the accuracy, stability, and interpretability of community detection results. Furthermore, it enhances the model's applicability and engineering generalization while ensuring computational feasibility.

[0029] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a flowchart of the community detection method for symmetric nonnegative matrix three-factor decomposition based on modularity constraints according to an embodiment of the present invention. Detailed Implementation

[0032] 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 denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0033] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0034] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.

[0035] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0036] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0037] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0038] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0039] To address the problems of existing community detection methods, such as the separation of matrix factorization models and modularity maximization methods, inconsistent optimization objectives, insufficient utilization of structural information, and lack of consistency constraints and theoretical convergence guarantees, this invention constructs a community detection method that can simultaneously integrate local network reconstruction information and global topological structure information within a unified mathematical framework. Specifically, existing methods based on non-negative matrix factorization or asymmetric non-negative matrix three-factor factorization mainly achieve low-dimensional network representation by minimizing the reconstruction error of the adjacency matrix. However, these methods ignore the global modular structure characteristics of the network, resulting in insufficient consistency in the global structure of the obtained community partitions. While modularity maximization methods can characterize community quality from the perspective of overall structure, they typically rely on discrete partitioning or spectral decomposition techniques, making it difficult to combine with continuous matrix factorization representations and unable to naturally generate interpretable soft community membership matrices. Therefore, this invention aims to solve how to simultaneously introduce matrix reconstruction error and modularity optimization terms into the same optimization objective function, coordinate the relationship between different community representations by constructing a consistency regularization mechanism, and design a stable iterative update algorithm that satisfies non-negativity constraints and KKT optimality conditions. This will enable a more accurate, stable, and theoretically guaranteed detection method for complex network community structures.

[0040] Example 1

[0041] In this embodiment 1, a community detection system based on nonnegative matrix three-factor decomposition with modularity constraints is first provided. The system includes stages such as network data input, adjacency matrix construction, modularity matrix calculation, parameter initialization, joint optimization iterative calculation, constraint normalization processing, and community result output. Each stage runs in a predetermined sequence and is controlled by the processor to iterate until the convergence condition is met.

[0042] like Figure 1 As shown, in this embodiment, a nonnegative matrix three-factor decomposition community detection method based on modularity constraints is implemented using the above-described system. This method constructs a unified joint optimization model to simultaneously optimize the reconstruction error of the network adjacency matrix and the modularity objective function within the same framework. It also achieves coordination and coupling between different community membership representations through a consistency regularization mechanism, thereby enhancing the expressive power of global topological features while maintaining local network structural information.

[0043] The functional framework of the joint optimization model includes: a network data input module, an adjacency matrix construction module, a module degree matrix calculation module, a parameter initialization module, a joint optimization iteration module, and a community result output module. Specifically, the method includes the following iterative steps: S1: Read the original network data and construct an input matrix X based on the node set and edge set; S2: Calculate the node degree and the total number of edges in the network based on the input matrix X, and further construct the module degree matrix; S3: Calculate the module degree matrix based on matrices W, H, and... Perform non-negative initialization, where W represents the node-to-community membership matrix, and H represents the inter-community relationship matrix. S4: Construct the joint optimization objective function and, under non-negativity constraints, employ an alternating iterative optimization strategy to update W, H, and W sequentially according to a preset update rule. S5: After each update, [the system will perform a certain action]. Perform row normalization to make the sum of the elements in each row equal to 1; S6: Determine if the convergence condition is met; if it is met, output the final community membership matrix W, otherwise return to step S4 to continue iterating.

[0044] First, during the initial stage of system operation, the network data input module reads raw graph data. This graph data can originate from a database, text file, or real-time data stream, and its format includes node sets and edge sets. The system then converts the read edge data into an adjacency matrix. In particular, for unauthorized networks, when a node With nodes When there is a connection, ,otherwise For weighted networks, Represents a node With nodes The connection weights between them. Where the matrix elements... This indicates whether a connection exists or the connection weight exists between node i and node j. The system then enters the module degree matrix construction phase, which calculates the degree of each node. and the total number of network edges Construct the modularity matrix ,in ,in This modularity matrix is ​​used to characterize the difference between the actual network connection structure and the random network structure, providing a foundation for subsequent modularity optimization.

[0045] After completing matrix construction, the system enters the parameter initialization phase, initializing the nonnegative matrices W, H, and... Initialization is performed, where W represents the membership matrix from nodes to communities, and H represents the relationship matrix between communities. This represents the community membership matrix used for modularity optimization. All matrices are initialized to non-negative values ​​to ensure that the subsequent multiplication update rule holds mathematically. The system then enters the core joint optimization computation phase, where the objective function is constructed as follows: Where μ is the regularization parameter. The first term ( ) is used to reconstruct the adjacency matrix, thereby preserving the local connectivity structure of the network; the second term ( ) is used to maximize the modularity metric to enhance the global community structure characteristics; the third item ( ) used to constrain W and The consistency between them makes the two community representations tend to be coordinated; the fourth item ( This is used to enhance the discriminative power of the W matrix, making community divisions clearer. Constraints during the optimization process include... , , as well as This ensures the non-negativity of the matrix and The probabilistic interpretation of a matrix.

[0046] Since the objective function is non-convex, this embodiment employs an alternating iterative optimization strategy to solve it. Specifically, in each iteration, the system fixes two variables, treating them as constants, and optimizes and updates the remaining variables. The update process is derived by constructing a Lagrange function and introducing non-negative constrained Lagrange multipliers, ultimately obtaining a multiplicative update rule that satisfies the KKT optimality condition. When updating the matrix, the system performs multiplication updates based on the numerator and denominator expressions containing adjacency matrix items and consistency regularization terms, and takes an exponent of 1 / 4 on the update result to ensure numerical stability; during the update... When updating the matrix, it is done according to the multiplication rules of standard nonnegative matrix decomposition; during the update... When updating the matrix, both modularity matrix items and consistency items are considered, and updates are performed using a multiplication update rule with an exponent of 1 / 2. After each update, the system... The matrix is ​​row normalized so that the sum of the elements in each row equals 1, thus satisfying the constraints.

[0047] In terms of timing control, the entire algorithm follows the steps of initialization, calculation of the modularity matrix, and updating. ,renew ,renew The normalization process and convergence judgment are executed sequentially and cyclically. Convergence judgment can be based on the objective function's change being less than a preset threshold or reaching the maximum number of iterations. When the convergence condition is met, the system stops iterating and outputs the final community membership matrix. The output format can be a continuous membership probability matrix, or it can be converted to discrete community labels according to the maximum membership principle.

[0048] In terms of system implementation, this invention can be implemented using a computing device including a processor, a memory, and input / output interfaces. The memory stores the adjacency matrix, the modularity matrix, and various optimization variables. The processor performs matrix multiplication, trace operation, norm calculation, and the multiplication update formula, as follows:

[0049]

[0050] in This is a hyperparameter.

[0051] The control unit is responsible for loop control and convergence determination, and the output interface is used to output the community detection results. Since the update rule adopts a multiplicative form, if the initial matrix is ​​non-negative, it will remain non-negative throughout the entire iteration process. The final converged solution satisfies the KKT optimality condition, thus ensuring the theoretical correctness and engineering feasibility of the algorithm.

[0052] The above technical solution achieves a unified integration of modularity maximization and non-negative matrix three-factor decomposition. It coordinates local structure reconstruction and global community quality optimization under the same optimization framework, which not only ensures the interpretability and continuous representation capability of the model, but also enhances the accuracy and stability of community structure identification.

[0053] Example 2

[0054] This embodiment 2 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the community detection method based on modularity constraints of a symmetric nonnegative matrix three-factor decomposition as described above.

[0055] Example 3

[0056] This embodiment 3 provides a computer device, including a memory and a processor. The processor and the memory communicate with each other. The memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute the community detection method based on modularity constraints of symmetric nonnegative matrix three-factor decomposition as described above.

[0057] Example 4

[0058] This embodiment 4 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions to implement the community detection method of symmetric nonnegative matrix three-factor decomposition based on modularity constraints as described above.

[0059] In summary, this invention constructs a joint optimization framework that unifies the asymmetric nonnegative matrix three-factor decomposition model and the modularity maximization objective into a single optimization function. By simultaneously introducing an adjacency matrix reconstruction error term and a modularity matrix trace function term into the objective function, it achieves synchronous optimization of local network structure information and global topological structure features. This unified modeling approach constitutes the core protection point of this invention. A consistency regularization term is also introduced. This design addresses the coupling relationship between the community membership matrix obtained from constraint decomposition and the membership matrix obtained from modularity optimization, ensuring consistency and coordination between the two representations at the structural level. This specific form of regularization design is a key innovation. Under the conditions of non-negativity and row normalization constraints, a multiplicative update rule satisfying the KKT optimality condition is derived: the update of matrix W considers both the adjacency matrix reconstruction term and the consistency regularization term, and its update adopts a multiplicative update form with an exponent of 1 / 4; the update of matrix H is based on the multiplicative update rule of standard non-negative matrix three-factor decomposition; and the update of matrix... The update simultaneously introduces modularity matrix terms and consistency constraint terms, and its update adopts an exponential 1 / 2 multiplication update form; after the update is completed, Perform row normalization to make the sum of each row's elements equal to 1, thus ensuring its probabilistic interpretation.

[0060] The algorithm employs different exponential update strategies for different variables to ensure stability and convergence. This update mechanism and its theoretical derivation are important protected content. The construction method of the modularity matrix and its specific implementation path embedded in the matrix decomposition framework constitute key technical features. The whole system forms an engineering-featured community detection system structure, including a complete process of data input, modularity matrix construction, alternating iterative optimization and constraint handling. This systematic implementation scheme also falls within the scope of the technical solution to be protected.

[0061] Without altering the core idea of ​​this invention, other alternative solutions can be used to achieve the same inventive objective. For example, in the objective function, the adjacency matrix reconstruction term can be replaced by the Frobenius norm with other difference measures, such as KL divergence; regarding consistency constraints, other forms of regularization methods can also be used to constrain the differences between two membership matrices. In terms of algorithm implementation, in addition to the multiplication update rule, projection gradient descent or other optimization methods that satisfy non-negativity constraints can also be used for solving the problem; in terms of system architecture, matrix operations can also be implemented through distributed computing or hardware acceleration. Therefore, the above alternative solutions can also achieve the technical objective of this invention without changing the joint optimization idea.

[0062] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0066] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. A community detection method based on modularity constraints for symmetric nonnegative matrix three-factor decomposition, characterized in that, include: S1: Read the raw network data and construct the input matrix X based on the set of nodes and the set of edges; S2: Calculate the node degree and the total number of edges in the network based on the input matrix X, and construct the modularity matrix; S3: For matrices W, H and... Perform non-negative initialization, where W represents the node-to-community membership matrix, and H represents the inter-community relationship matrix. This represents the auxiliary membership matrix used for modularity optimization; S4: Construct a joint optimization objective function, and under non-negativity constraints, employ an alternating iterative optimization strategy to sequentially update W, H, and H according to a preset update rule. ; S5: After each update, Perform row normalization so that the sum of the elements in each row is 1; S6: Determine whether the convergence condition is met; If the condition is met, output the final community membership matrix W; otherwise, return to step S4 to continue iterating.

2. The community detection method for symmetric nonnegative matrix three-factor decomposition based on modularity constraints according to claim 1, characterized in that, The raw network data includes node and edge sets; the read edge data is then converted into an adjacency matrix. In particular, for unauthorized networks, when a node With nodes When there is a connection, ,otherwise For weighted networks, Represents a node With nodes The connection weights between them; where the matrix elements This indicates whether there is a connection or connection weight between node i and node j.

3. The community detection method for symmetric nonnegative matrix three-factor decomposition based on modularity constraints according to claim 2, characterized in that, By calculating the degree of each node and the total number of edges in the network Construct the modularity matrix ,in ,in The modularity matrix is ​​used to characterize the difference between the actual connection structure of the network and the random network structure.

4. The community detection method for symmetric nonnegative matrix three-factor decomposition based on modularity constraints according to claim 3, characterized in that, The objective function constructed is: Where μ is the regularization parameter; the first term ( ) is used to reconstruct the adjacency matrix, thereby preserving the local connectivity structure of the network; the second term ( This is used to maximize the modularity metric to enhance the global community structure characteristics; The third item ( ) used to constrain W and The consistency between them makes the two community representations tend to be coordinated; the fourth item ( This is used to enhance the discriminative power of the W matrix, making community divisions clearer; during the optimization process, constraints include... , , as well as This ensures the non-negativity of the matrix and The probabilistic interpretation of a matrix.

5. The community detection method for symmetric nonnegative matrix three-factor decomposition based on modularity constraints according to claim 4, characterized in that, In each iteration, two variables are fixed and treated as constants, while the remaining variables are optimized and updated. The update process is derived by constructing a Lagrange function and introducing Lagrange multipliers with non-negativity constraints, ultimately obtaining a multiplication update rule that satisfies the KKT optimality condition.

6. The community detection method for symmetric nonnegative matrix three-factor decomposition based on modularity constraints according to claim 5, characterized in that, In update When updating the matrix, a multiplication update is performed based on the numerator and denominator expressions containing adjacency matrix items and consistency regularization terms, and the update result is multiplied by 1 / 4 to ensure numerical stability; during the update... When updating the matrix, it is done according to the multiplication rules of standard nonnegative matrix decomposition; during the update... When updating a matrix, both modularity matrix items and consistency items are considered, and updates are performed using a multiplication update rule with an exponent of 1 / 2.

7. The community detection method for symmetric nonnegative matrix three-factor decomposition based on modularity constraints according to claim 6, characterized in that, After each update, the system performs... The matrix is ​​row normalized so that the sum of the elements in each row equals 1, thus satisfying the constraints.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the community detection method based on modularity constraints of a symmetric nonnegative matrix three-factor decomposition as described in any one of claims 1-6.

9. A computer device, characterized in that, The method includes a memory and a processor, the processor and the memory communicating with each other, the memory storing program instructions executable by the processor, and the processor calling the program instructions to execute the community detection method based on modularity constraints of a symmetric nonnegative matrix three-factor decomposition as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the community detection method of symmetric nonnegative matrix three-factor decomposition based on modularity constraints as described in any one of claims 1-6.