Multilayer network community detection method based on weighted prior tensor chain decomposition

Through the methods of weighted prior tensor chain decomposition and loop optimization, the problem of unsatisfactory community detection results under multi-layer networks is solved, more accurate community detection is achieved, and the intrinsic connections of data and the integrity of community structure is maintained.

CN118917957AInactive Publication Date: 2024-11-08GUANGDONG UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410708259.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-03
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing technology has poor community detection results under multi-layer networks, mainly because network compression damages the integrity of the community structure or ignores the inherent connection of data in multi-layer networks.

Method used

A multi-layer network community detection method based on weighted prior tensor chain decomposition is adopted, and the community is extracted through weighted flattening processing, tensor chain decomposition and public community manifold learning.

Benefits of technology

Maintain the intrinsic connection of data in multi-layer networks, ensure the integrity of community structure, and improve the accuracy of community detection under multi-layer networks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917957A_ABST
    Figure CN118917957A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-layer network community detection method based on weighted prior tensor chain decomposition. The method comprises the following steps: firstly, acquiring a multi-layer network and extracting features to obtain weighted flat network features; thirdly, initializing three tensor sequence factors and performing loop optimization to obtain a final first tensor sequence factor; finally, a community is extracted from the final first tensor sequence. According to the invention, the community detection result under the multi-layer network is more accurate.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of community detection, and more specifically, to a multi-layer network community detection method based on weighted prior tensor chain decomposition. Background Art

[0002] As society and technology continue to develop, the connections between various objective things are becoming more and more valuable. Various connections form networks, and communities are a subset of these networks. The nodes in a community have denser connectivity, which allows for a better understanding of the network structure and function and promotes various applications. Network community detection is about finding communities and giving full play to their advantages.

[0003] Although some existing algorithms can achieve community detection, they are mainly aimed at single networks with a single type of node relationship, while real-world network systems often have multi-layer structures. In order to further perform community detection, some methods have been proposed, such as compressing multi-layer networks into single-layer networks, detecting layer by layer first and then aggregating, but either compressing the network and damaging the integrity of the community structure, or ignoring the inherent connection of the data in the multi-layer network, which ultimately leads to unsatisfactory results in multi-layer network detection.

[0004] The prior art discloses a network community detection method based on multi-objective dense mother calculation, which detects the community by cyclically generating new individuals, updating the network population, and locally searching the network population. The present invention uses a method based on weighted prior tensor chain decomposition for detection, and does not use the above-mentioned multi-objective dense mother calculation method for detection. Summary of the invention

[0005] Aiming at the defect of unsatisfactory community detection results in multi-layer networks in the prior art, the present invention provides a multi-layer network community detection method based on weighted prior tensor chain decomposition.

[0006] The primary purpose of the present invention is to solve the above technical problems. The technical solution of the present invention is as follows:

[0007] A multi-layer network community detection method based on weighted prior tensor chain decomposition, including:

[0008] S1: Obtain a multi-layer network, perform weighted flattening processing and extract features to obtain weighted flat network features;

[0009] S2: Randomly initialize three sequences as the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor respectively;

[0010] S3: cyclically optimizing the weighted flat network feature, the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor to obtain a final first tensor sequence factor;

[0011] S4: Perform community extraction on the final first tensor sequence factor to obtain a community.

[0012] Further, step S1 includes:

[0013] S101: Obtain a multi-layer network and convert it into a weighted planar network;

[0014] S102: Searching the weighted planarized network to obtain a plurality of dense subgraphs;

[0015] S103: constructing prior information using the plurality of dense subgraphs to obtain a priori matrix;

[0016] S104: Merge the prior matrix and the multi-layer network to obtain weighted flat network features.

[0017] Furthermore, in step S101, the formula for converting into a weighted planar network is as follows:

[0018]

[0019] Among them, N represents the number of layers in the multi-layer network, a i represents the adjacency matrix of the i-th layer of the multi-layer network, l i represents the number of edges in the i-th layer of the multi-layer network, and H represents the sum of the number of edges in the multi-layer network.

[0020] Furthermore, in step S103, the formula for constructing the priori matrix is ​​as follows:

[0021]

[0022] Wherein, a and b represent the vertex numbers of the weighted planar network, respectively, and Z a , Z b Respectively represent the vertices corresponding to a and b, C i Represents the i-th dense subgraph in a multi-layer network.

[0023] Furthermore, in step S104, the calculation formula of the weighted flat network feature is as follows:

[0024] O 1 =A+μτ(Ap)

[0025] Among them, μ represents the prior weight, τ(Ap) represents the prior matrix replicated multiple times in three-dimensional space, and A represents a multi-layer network.

[0026] Further, in step S3, the loop optimization includes:

[0027] S301: performing tensor chain decomposition on the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor to obtain tensor chain decomposition features; performing common community manifold learning on the first tensor sequence factor and the weighted flat network feature to obtain manifold learning features;

[0028] S302: constructing an optimization objective function according to the weighted flat network feature, the tensor chain decomposition feature, and the manifold learning feature;

[0029] S303: Optimizing the first tensor sequence factor according to the target optimization function to obtain an optimized first tensor sequence factor as a new first tensor sequence factor;

[0030] S304: Substituting the new first tensor sequence factor into the target optimization function, optimizing the second tensor sequence factor, and obtaining an optimized second tensor sequence factor as a new second tensor sequence factor;

[0031] S305: Substituting the new second tensor sequence factor into the target optimization function, optimizing the third tensor sequence factor, and obtaining an optimized third tensor sequence factor as a new third tensor sequence factor;

[0032] S306: Determine whether the current optimization times are less than the preset optimization times; if so, substitute the new third tensor factor into the target optimization function and execute step S303; if not, execute step S307;

[0033] S307: Use the first tensor sequence factor as the final first tensor sequence factor.

[0034] Furthermore, in step S301, the formula for tensor chain decomposition is as follows:

[0035] O 2 =G 1 *G 2 *G 3

[0036] Among them, G 1 represents the first tensor sequence factor, G 2 represents the second tensor sequence factor, G 3 Represents the third tensor sequence factor.

[0037] Furthermore, in step S301, the formula for learning the public community manifold is as follows:

[0038] O 3 =λtr((G 1 ) T LG 1)

[0039] Among them, λ represents the regularization parameter, G 1 represents the first tensor sequence factor, and L represents the result of Laplace calculation on the weighted flat network feature.

[0040] Furthermore, in step S302, the optimization objective function is as follows:

[0041]

[0042] Among them, O 1 represents the weighted flat network feature, O 2 represents the decomposition characteristics of the tensor chain, O 3 Represents manifold learning features.

[0043] A multi-layer network community detection device based on weighted prior tensor chain decomposition, comprising:

[0044] Weighted flattening processing module: obtains a multi-layer network, performs weighted flattening processing and extracts features to obtain weighted flattened network features;

[0045] Initialization module: randomly initialize three sequences as the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor respectively;

[0046] A loop optimization module: loop optimizing the weighted flat network feature, the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor to obtain a final first tensor sequence factor;

[0047] Community extraction module: extracts the community from the final first tensor sequence factor to obtain a community.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention not only maintains the intrinsic connection of data in the multi-layer network, but also ensures the integrity of the community structure by adopting a cyclic optimization method. The accuracy of community detection in the multi-layer network is improved by weighted flattening processing. In summary, the present invention makes the community detection results in the multi-layer network more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a multi-layer network community detection method based on weighted prior tensor chain decomposition provided in an embodiment.

[0051] Figure 2 A flowchart of weighted flattening processing and feature extraction provided in an embodiment.

[0052] Figure 3A flowchart of loop optimization provided for an embodiment.

[0053] Figure 4 A structural diagram of a multi-layer network community detection device based on weighted prior tensor chain decomposition provided in an embodiment. DETAILED DESCRIPTION

[0054] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0055] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0056] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0057] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0058] Example

[0059] like Figure 1 As shown, a multi-layer network community detection method based on weighted prior tensor chain decomposition includes:

[0060] S1: Obtain a multi-layer network, perform weighted flattening processing and extract features to obtain weighted flat network features;

[0061] S2: Randomly initialize three sequences as the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor respectively;

[0062] S3: cyclically optimizing the weighted flat network feature, the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor to obtain a final first tensor sequence factor;

[0063] S4: Perform community extraction on the final first tensor sequence factor to obtain a community.

[0064] In a specific embodiment, in step S4, the extraction formula of the community index matrix is ​​as follows:

[0065]

[0066] Among them, B represents the final first tensor sequence factor, i and j represent the rows and columns of the matrix respectively.

[0067] Furthermore, if Figure 2 As shown, step S1 includes:

[0068] S101: Obtain a multi-layer network and convert it into a weighted planar network;

[0069] S102: Searching the weighted planarized network to obtain a plurality of dense subgraphs;

[0070] S103: constructing prior information using the plurality of dense subgraphs to obtain a priori matrix;

[0071] S104: Merge the prior matrix and the multi-layer network to obtain weighted flat network features.

[0072] In a specific embodiment, in step S102, a greedy algorithm is used to search the weighted planarized network to obtain a dense subgraph, and at the same time, an optimal value of the number of communities finally extracted is determined.

[0073] Furthermore, in step S101, the formula for converting into a weighted planar network is as follows:

[0074]

[0075] Among them, N represents the number of layers in the multi-layer network, a i represents the adjacency matrix of the i-th layer of the multi-layer network, l i represents the number of edges in the i-th layer of the multi-layer network, and H represents the sum of the number of edges in the multi-layer network.

[0076] Furthermore, in step S103, the formula for constructing the priori matrix is ​​as follows:

[0077]

[0078] Wherein, a and b represent the vertex numbers of the weighted planar network, respectively, and Z a , Z b Respectively represent the vertices corresponding to a and b, C i Represents the i-th dense subgraph in a multi-layer network.

[0079] It should be noted that Ap(a,b)=1 means Z a , Z b The two vertices are more closely related.

[0080] Furthermore, in step S104, the calculation formula of the weighted flat network feature is as follows:

[0081] O 1 =A+μτ(Ap)

[0082] Among them, μ represents the prior weight, τ(Ap) represents the prior matrix replicated multiple times in three-dimensional space, and A represents a multi-layer network.

[0083] In a specific embodiment, the value of μ is 0.1.

[0084] Furthermore, if Figure 3 As shown, in step S3, the loop optimization includes:

[0085] S301: performing tensor chain decomposition on the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor to obtain tensor chain decomposition features; performing common community manifold learning on the first tensor sequence factor and the weighted flat network feature to obtain manifold learning features;

[0086] S302: constructing an optimization objective function according to the weighted flat network feature, the tensor chain decomposition feature, and the manifold learning feature;

[0087] S303: Optimizing the first tensor sequence factor according to the target optimization function to obtain an optimized first tensor sequence factor as a new first tensor sequence factor;

[0088] S304: Substituting the new first tensor sequence factor into the target optimization function, optimizing the second tensor sequence factor, and obtaining an optimized second tensor sequence factor as a new second tensor sequence factor;

[0089] S305: Substituting the new second tensor sequence factor into the target optimization function, optimizing the third tensor sequence factor, and obtaining an optimized third tensor sequence factor as a new third tensor sequence factor;

[0090] S306: Determine whether the current optimization times are less than the preset optimization times; if so, substitute the new third tensor factor into the target optimization function and execute step S303; if not, execute step S307;

[0091] S307: Use the first tensor sequence factor as the final first tensor sequence factor.

[0092] It should be noted that tensor chain decomposition is a method that is naturally suitable for processing multidimensional data. Through tensor chain decomposition, the present invention can simultaneously consider information at multiple levels in the network, thereby more accurately identifying and dividing communities. At the same time, using auxiliary information in the multi-layer network as a priori significantly improves the ability to capture communities during tensor decomposition.

[0093] It should be noted that common community manifold learning helps to assimilate the local and global structural properties inherent in the network. Common community manifold learning imposes constraints on the basic matrix and improves the accuracy of group detection methods based on matrix decomposition.

[0094] It should be noted that the manifold structure represents the community structure that frequently appears on multiple network levels. The present invention can identify and utilize these popular structures through common community manifold learning, thereby effectively capturing the communities in the network. At the same time, combined with the tensor chain decomposition method, the present invention shows higher efficiency and accuracy when processing multi-layer networks with complex community structures.

[0095] In a specific embodiment, the preset number of optimizations in S306 is 50 times.

[0096] Furthermore, in step S301, the formula for tensor chain decomposition is as follows:

[0097] O 2 =G 1 *G 2 *G 3

[0098] Among them, G 1 represents the first tensor sequence factor, G 2 represents the second tensor sequence factor, G 3 Represents the third tensor sequence factor.

[0099] Furthermore, in step S301, the formula for learning the public community manifold is as follows:

[0100] O 3 =λtr((G 1 ) T LG 1 )

[0101] Among them, λ represents the regularization parameter, G 1 represents the first tensor sequence factor, and L represents the result of Laplace calculation on the weighted flat network feature.

[0102] In a specific embodiment, the value of λ is 0.01.

[0103] Furthermore, in step S302, the optimization objective function is as follows:

[0104]

[0105] Among them, O 1 represents the weighted flat network feature, O 2 represents the decomposition characteristics of the tensor chain, O 3 Represents manifold learning features.

[0106] It should be noted that, combined with O 1 , O 2 and O 3 , the optimization objective function is as follows:

[0107]

[0108] Among them, μ represents the prior weight, τ(Ap) represents the prior matrix replicated multiple times in three-dimensional space, A represents a multi-layer network, λ represents the regularization parameter, G 1 represents the first tensor sequence factor, G 2 represents the second tensor sequence factor, G 3 represents the third tensor sequence factor, and L represents the result of Laplace calculation on the weighted flat network feature.

[0109] In a specific embodiment, in step S303, the optimization formula for the first tensor sequence factor is as follows:

[0110]

[0111] Assume that the first tensor sequence factor G 1 is an unknown number, let We can solve for G 1 , as the new first tensor sequence factor.

[0112] Among them, (a) n represents the mode-n expansion operation of a, min represents the minimum error optimization, μ represents the prior weight, τ(Ap) represents the prior matrix replicated multiple times in three-dimensional space, A represents a multi-layer network, λ represents the regularization parameter, G 1 represents the first tensor sequence factor, G 2 represents the second tensor sequence factor, G 3 represents the third tensor sequence factor, and L represents the result of Laplace calculation on the weighted flat network feature.

[0113] In a specific embodiment, in step S304, the optimization formula for the second tensor sequence factor is as follows:

[0114]

[0115] Assume that the second tensor sequence factor G 2 is an unknown number, let We can solve for G 2 , as the new second tensor sequence factor.

[0116] Among them, (a) n represents the mode-n expansion operation on a, min represents the minimum error optimization, fold(a,b) represents the mode-b folding of a, μ represents the prior weight, τ(Ap) represents the prior matrix replicated multiple times in three-dimensional space, A represents a multi-layer network, λ represents the regularization parameter, G 1 represents the first tensor sequence factor, G 2represents the second tensor sequence factor, G 3 represents the third tensor sequence factor, and L represents the result of Laplace calculation on the weighted flat network feature.

[0117] In a specific embodiment, in step S305, the optimization formula for the third tensor sequence factor is as follows:

[0118]

[0119]

[0120] Assume that the third tensor sequence factor G 3 is an unknown number, let We can solve for G 3 , as the new third tensor sequence factor.

[0121] Among them, (a) n represents the mode-n expansion operation of a, min represents the minimum error optimization, μ represents the prior weight, τ(Ap) represents the prior matrix replicated multiple times in three-dimensional space, A represents a multi-layer network, λ represents the regularization parameter, G 1 represents the first tensor sequence factor, G 2 represents the second tensor sequence factor, G 3 represents the third tensor sequence factor, and L represents the result of Laplace calculation on the weighted flat network feature.

[0122] like Figure 4 As shown, a multi-layer network community detection device based on weighted prior tensor chain decomposition includes:

[0123] Weighted flattening processing module: obtains a multi-layer network, performs weighted flattening processing and extracts features to obtain weighted flattened network features;

[0124] Initialization module: randomly initialize three sequences as the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor respectively;

[0125] A loop optimization module: loop optimizing the weighted flat network feature, the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor to obtain a final first tensor sequence factor;

[0126] Community extraction module: extracts the community from the final first tensor sequence factor to obtain a community.

[0127] The same or similar reference numerals correspond to the same or similar components;

[0128] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;

[0129] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. Multi-layer network community detection method based on weighted prior tensor chain decomposition, It is characterized in that include: S1: Obtain a multi-layer network, perform weighted flattening processing and extract features to obtain weighted flat network features; S2: Randomly initialize three sequences as the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor respectively; S3: cyclically optimizing the weighted flat network feature, the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor to obtain a final first tensor sequence factor; S4: Perform community extraction on the final first tensor sequence factor to obtain a community.

2. According to the multi-layer network community detection method based on weighted prior tensor chain decomposition according to claim 1, It is characterized in that Step S1 includes: S101: Obtain a multi-layer network and convert it into a weighted planar network; S102: Searching the weighted planarized network to obtain a plurality of dense subgraphs; S103: constructing prior information using the plurality of dense subgraphs to obtain a priori matrix; S104: Merge the prior matrix and the multi-layer network to obtain weighted flat network features.

3. According to the multi-layer network community detection method based on weighted prior tensor chain decomposition according to claim 2, It is characterized in that In step S101, the formula for converting into a weighted planar network is as follows: Among them, N represents the number of layers in the multi-layer network, a i represents the adjacency matrix of the i-th layer of the multi-layer network, l i represents the number of edges in the i-th layer of the multi-layer network, and H represents the sum of the number of edges in the multi-layer network.

4. According to the multi-layer network community detection method based on weighted prior tensor chain decomposition according to claim 2, It is characterized in that In step S103, the formula for constructing the priori matrix is ​​as follows: Wherein, a and b represent the vertex numbers of the weighted planar network, respectively, and Z a , Z b Respectively represent the vertices corresponding to a and b, C i Represents the i-th dense subgraph in a multi-layer network.

5. According to the multi-layer network community detection method based on weighted prior tensor chain decomposition according to claim 4, It is characterized in that In step S104, the calculation formula of the weighted flat network feature is as follows: The 1 =A+μτ(Ap) Among them, μ represents the prior weight, τ(Ap) represents the prior matrix replicated multiple times in three-dimensional space, and A represents a multi-layer network.

6. According to the multi-layer network community detection method based on weighted prior tensor chain decomposition according to claim 1, It is characterized in that In step S3, loop optimization includes: S301: performing tensor chain decomposition on the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor to obtain tensor chain decomposition features; performing common community manifold learning on the first tensor sequence factor and the weighted flat network feature to obtain manifold learning features; S302: constructing an optimization objective function according to the weighted flat network feature, the tensor chain decomposition feature, and the manifold learning feature; S303: Optimizing the first tensor sequence factor according to the target optimization function to obtain an optimized first tensor sequence factor as a new first tensor sequence factor; S304: Substituting the new first tensor sequence factor into the target optimization function, optimizing the second tensor sequence factor, and obtaining an optimized second tensor sequence factor as a new second tensor sequence factor; S305: Substituting the new second tensor sequence factor into the target optimization function, optimizing the third tensor sequence factor, and obtaining an optimized third tensor sequence factor as a new third tensor sequence factor; S306: Determine whether the current optimization times are less than the preset optimization times; if so, substitute the new third tensor factor into the target optimization function and execute step S303; if not, execute step S307; S307: Use the first tensor sequence factor as the final first tensor sequence factor.

7. According to the multi-layer network community detection method based on weighted prior tensor chain decomposition according to claim 6, It is characterized in that In step S301, the formula for tensor chain decomposition is as follows: O 2 =G 1 *G 2 *G 3 Among them, G 1 represents the first tensor sequence factor, G 2 represents the second tensor sequence factor, G 3 Represents the third tensor sequence factor.

8. According to the multi-layer network community detection method based on weighted prior tensor chain decomposition according to claim 6, It is characterized in that In step S301, the formula for public community manifold learning is as follows: O 3 =λtr((G 1 ) T LG 1 ) Among them, λ represents the regularization parameter, G 1 represents the first tensor sequence factor, and L represents the result of Laplace calculation on the weighted flat network feature.

9. The multi-layer network community detection method based on weighted prior tensor chain decomposition according to claim 6, It is characterized in that In step S302, the optimization objective function is as follows: Among them, O 1 represents the weighted flat network feature, O 2 represents the decomposition characteristics of the tensor chain, O 3 Represents manifold learning features.

10. A multi-layer network community detection device based on weighted prior tensor chain decomposition, used to implement the detection method according to any one of claims 1 to 9, include: Weighted flattening processing module: obtains a multi-layer network, performs weighted flattening processing and extracts features to obtain weighted flattened network features; Initialization module: randomly initialize three sequences as the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor respectively; A loop optimization module: loop optimizing the weighted flat network feature, the first tensor sequence factor, the second tensor sequence factor, and the third tensor sequence factor to obtain a final first tensor sequence factor; Community extraction module: extracts the community from the final first tensor sequence factor to obtain a community.