Community network attack method based on non-redundant community structure learning
Through a matrix decomposition algorithm based on non-redundant community structure learning, the perturbation matrix is optimized and applied to the network, and the problems of poor attack effects and redundant interference in the prior art are solved, and precise interference to complex community structures and effective interference to community detection results are achieved.
Patent Information
- Application Number
- CN202411710245.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-27
- Publication Date
- 2025-02-28
AI Technical Summary
Existing cyber attack methods fail to fully utilize the complex community structure characteristics, resulting in unsatisfactory attack effects and may cause redundant interference to non-target areas.
The matrix decomposition algorithm based on non-redundant community structure learning is adopted, and by constructing the original community network membership matrix, the post-attack community network membership matrix and the perturbation matrix, iterative optimization is performed to obtain the optimized perturbation matrix and apply it to the network to achieve precise interference.
It significantly reduces the similarity of community structures in the network, effectively interferes with community detection results, while maintaining the rationality of the overall structure of the network, and minimizing redundant interference to non-target areas.
Smart Images
Figure 00000000_0000_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of social network privacy protection, and in particular to a community network attack method based on non-redundant community structure learning, which is suitable for attack scenarios in complex network environments. Background Art
[0002] In the context of the big data era, complex network analysis has become a hot topic in interdisciplinary research, and its research content covers many fields such as social science, biology, and information science. As a core tool for network structure analysis, community detection technology can reveal the close connections between nodes in the network and the functional modules they constitute, which has a far-reaching impact on network analysis, recommendation systems, social media and other fields. This technology aims to identify highly interconnected subsets of nodes in the network, namely communities, which are characterized by the fact that the connection strength between internal nodes is much higher than the connection with external nodes. This structural feature is crucial for understanding the functional modules and hierarchical structure of the network. However, community detection algorithms may be abused in the field of network security, thus causing privacy and security issues. Since the algorithm may expose the secret connections of individuals or groups in the network, studying how to interfere with community detection algorithms through attack methods to hide or disguise the true structure of the network has become a key direction of network security research. At present, most network attack methods focus on interfering with the weaknesses of specific community detection algorithms, but these methods often fail to fully utilize the community structure characteristics within the network, resulting in unsatisfactory attack effects. Therefore, developing a network attack method that can effectively interfere with complex community structures is not only innovative in theory, but also has important strategic value in practical applications. Therefore, the present invention proposes a community network attack method based on non-redundant community structure learning. Summary of the invention
[0003] The purpose of the present invention is to provide a community network attack method based on non-redundant community structure learning, which uses community structure information to design a matrix decomposition algorithm to generate a disturbed network, thereby achieving precise interference with the community structure in a complex network. It can not only significantly reduce the similarity of the community structure in the network, but also effectively interfere with the community detection results while ensuring the overall structure of the network.
[0004] To achieve the above object, the present invention provides the following solutions:
[0005] Community network attack methods based on non-redundant community structure learning include:
[0006] Obtain the adjacency matrix of the target community network;
[0007] Constructing an original community network membership matrix, a community network membership matrix after the attack, and a disturbance matrix of the target community network according to the adjacency matrix, and obtaining an optimized disturbance matrix by iteratively optimizing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix;
[0008] The optimized disturbance matrix is applied to the adjacency matrix to obtain the adjacency matrix after the attack, thereby completing the attack on the target community network.
[0009] Optionally, constructing an original community network membership matrix, a post-attack community network membership matrix, and a disturbance matrix of the target community network according to the adjacency matrix includes:
[0010] Several random matrices are set, and they are iteratively decomposed to obtain the initialized original community network membership matrix, the attacked community network membership matrix, and the disturbance matrix.
[0011] Optionally, the random matrix includes a first random matrix, a second random matrix, and a third random matrix. The first random matrix is used to iteratively decompose with the adjacency matrix to obtain an initialized original community network membership matrix, the second random matrix is used to iteratively decompose with the adjacency matrix and the third random matrix to obtain an initialized post-attack community network membership matrix, and the third random matrix is used to iteratively decompose with the adjacency matrix and the second random matrix to obtain an initialized perturbation matrix.
[0012] Optionally, an alternating optimization strategy is used to iteratively optimize the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix.
[0013] To further achieve the above objectives, the present invention also provides a community network attack system based on non-redundant community structure learning, comprising:
[0014] A matrix acquisition module is used to obtain the adjacency matrix of the target community network;
[0015] A matrix processing module, used to construct an original community network membership matrix, a community network membership matrix after the attack, and a disturbance matrix of the target community network according to the adjacency matrix, and obtain an optimized disturbance matrix by iteratively optimizing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix;
[0016] The network attack module is used to apply the optimized disturbance matrix to the adjacency matrix, obtain the adjacency matrix after the attack, and complete the attack on the target community network.
[0017] Optionally, the matrix processing module is specifically used to construct the original community network membership matrix, the attacked community network membership matrix, and the disturbance matrix of the target community network according to the adjacency matrix, including:
[0018] Several random matrices are set, and they are iteratively decomposed to obtain the initialized original community network membership matrix, the attacked community network membership matrix, and the disturbance matrix.
[0019] Optionally, the random matrix includes a first random matrix, a second random matrix, and a third random matrix. The first random matrix is used to iteratively decompose with the adjacency matrix to obtain an initialized original community network membership matrix, the second random matrix is used to iteratively decompose with the adjacency matrix and the third random matrix to obtain an initialized post-attack community network membership matrix, and the third random matrix is used to iteratively decompose with the adjacency matrix and the second random matrix to obtain an initialized perturbation matrix.
[0020] Optionally, the matrix processing module is specifically used to iteratively optimize the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix using an alternating optimization strategy.
[0021] To further achieve the above-mentioned purpose, the present invention also provides a processor, which is used to run a program, wherein the community network attack method based on non-redundant community structure learning is executed when the program is running.
[0022] To further achieve the above-mentioned purpose, the present invention also provides an electronic device, comprising one or more memories and a processor, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the community network attack method based on non-redundant community structure learning.
[0023] The beneficial effects of the present invention are:
[0024] (1) The present invention adopts a matrix decomposition strategy based on non-redundant community structure, which can more accurately identify the key community structure in the network and achieve effective interference with the community detection results by controlling the update of the perturbation matrix.
[0025] (2) The present invention minimizes redundant interference to non-target areas by optimizing the community membership matrix, effectively maintaining the rationality of the overall network structure while interfering with community detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0027] Figure 1 This is a flow chart of a community network attack method based on non-redundant community structure learning according to an embodiment of the present invention. DETAILED DESCRIPTION
[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0029] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0030] This embodiment provides a community network attack method based on non-redundant community structure learning, including:
[0031] Obtain the adjacency matrix of the target community network;
[0032] Constructing an original community network membership matrix, a community network membership matrix after the attack, and a disturbance matrix of the target community network according to the adjacency matrix, and obtaining an optimized disturbance matrix by iteratively optimizing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix;
[0033] The optimized disturbance matrix is applied to the adjacency matrix to obtain the adjacency matrix after the attack, thereby completing the attack on the target community network.
[0034] Specifically, this embodiment adopts a matrix decomposition strategy based on non-redundant community structure, which can more accurately identify the key community structure in the network, and effectively interfere with the community detection results by controlling the update of the perturbation matrix. By optimizing the community membership matrix, the redundant interference to non-target areas is minimized, and the rationality of the overall network structure is effectively maintained while interfering with community detection.
[0035] Furthermore, constructing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix of the target community network according to the adjacency matrix includes:
[0036] Several random matrices are set, and they are iteratively decomposed to obtain the initialized original community network membership matrix, the attacked community network membership matrix, and the disturbance matrix.
[0037] Furthermore, the random matrix includes a first random matrix, a second random matrix, and a third random matrix. The first random matrix is used to iteratively decompose with the adjacency matrix to obtain an initialized original community network membership matrix, the second random matrix is used to iteratively decompose with the adjacency matrix and the third random matrix to obtain an initialized post-attack community network membership matrix, and the third random matrix is used to iteratively decompose with the adjacency matrix and the second random matrix to obtain an initialized perturbation matrix.
[0038] Furthermore, an alternating optimization strategy is adopted to iteratively optimize the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix.
[0039] Combine the following Figure 1 The community network attack method based on non-redundant community structure learning provided in this embodiment is described in detail. This embodiment uses matrix decomposition technology to achieve precise attacks on complex networks by constructing an optimization model based on non-redundant community structure learning. First, the adjacency matrix A is obtained through the original topological structure of the target community network, and the original community network membership matrix U and the community network membership matrix U after the attack are constructed. P , represents the relationship between the network node and its community before and after the attack. Then, the network is disturbed by the perturbation matrix P, adding or deleting edges in the network to achieve the destruction of the community structure. The goal of this optimization method is to minimize the similarity between the original community structure and the community structure after the attack, while controlling the intensity and quantity of the perturbation, so as to maximize the interference effect on community detection while maintaining the overall structure of the network. Through the alternating optimization strategy, the perturbation matrix P, the original community network membership matrix U and the community network membership matrix U after the attack are gradually optimized. P , so that the attack can achieve the best effect within the specified budget range. The specific steps include:
[0040] Step 1: Obtain the adjacency matrix based on the original topological structure of the target community network Where n represents the number of nodes in the network. The element in the adjacency matrix A is 1 if there is an edge between the two nodes at the corresponding position, and 0 if there is no edge. The specific representation is as follows:
[0041]
[0042] Among them, A ij is the element in the i-th row and j-th column.
[0043] Step 2: Perform 20 iterations of matrix decomposition on the adjacency matrix A to obtain the initialized original community network membership matrix And the initialized community network membership matrix after the attack And according to the initialized attack community network membership matrix U P Generate the initialized perturbation matrix Where k represents the number of communities. The specific steps are as follows:
[0044] 2a) Initialize a random matrix U0, where all elements are not less than 0; iterate 20 times to obtain the original community network membership matrix U, using the following formula:
[0045]
[0046] Among them, ⊙ represents element-by-element multiplication (i.e., Hadamard product), (*) T Represents the transposed matrix of *.
[0047] 2b) Set up a random matrix and a matrix of all zeros U P0 All elements are not less than 0; iterate 20 times to get the initialized community network membership matrix U after the attack P And the method of initializing the perturbation matrix P uses the following formula:
[0048]
[0049] Among them, ⊙ represents element-by-element multiplication (i.e., Hadamard product), min(A,AU P0 U P0 T ) means taking matrix A and matrix AU P0 U P0 T The minimum value between corresponding elements.
[0050] Step 3: Set the number of iterations t, adopt the alternating optimization strategy, and iteratively update the perturbation matrix P, the original community network membership matrix U, and the community network membership matrix U after the attack. P , until the number of iterations is reached, and the optimized perturbation matrix P is obtained ′ ;
[0051] 3a) Set the number of links e of the community network attack, and use the original community network membership matrix U and the community network membership matrix U after the attack P , the formula for updating the perturbation matrix P is as follows:
[0052]
[0053] 3b) Using the perturbation matrix P and the community network membership matrix U after the attack P , the formula for optimizing the original community network membership matrix U is as follows:
[0054]
[0055] Among them, β is the weight coefficient of the similarity between the original community structure and the community structure after the attack, β>0.
[0056] 3c) Using the original community network membership matrix U and the perturbation matrix P, optimize the community network membership matrix U after the attack P The formula is as follows:
[0057]
[0058] Among them, α is the weight coefficient of the target attack constraint, which is used to control the intensity of the attack, α>0.
[0059] Step 4: Use the adjacency matrix A and the optimized perturbation matrix P ′ , get the adjacency matrix A of the community network after the attack * , the specific formula is as follows:
[0060] A * =A+P ′ .
[0061] To further optimize the technical solution, this embodiment also provides a community network attack system based on non-redundant community structure learning, including:
[0062] A matrix acquisition module is used to obtain the adjacency matrix of the target community network;
[0063] A matrix processing module, used to construct an original community network membership matrix, a community network membership matrix after the attack, and a disturbance matrix of the target community network according to the adjacency matrix, and obtain an optimized disturbance matrix by iteratively optimizing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix;
[0064] The network attack module is used to apply the optimized disturbance matrix to the adjacency matrix, obtain the adjacency matrix after the attack, and complete the attack on the target community network.
[0065] Furthermore, the matrix processing module is specifically used to construct the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix of the target community network according to the adjacency matrix, including:
[0066] Several random matrices are set, and they are iteratively decomposed to obtain the initialized original community network membership matrix, the attacked community network membership matrix, and the disturbance matrix.
[0067] The random matrix includes a first random matrix, a second random matrix, and a third random matrix. The first random matrix is used to iteratively decompose with the adjacency matrix to obtain an initialized original community network membership matrix. The second random matrix is used to iteratively decompose with the adjacency matrix and the third random matrix to obtain an initialized post-attack community network membership matrix. The third random matrix is used to iteratively decompose with the adjacency matrix and the second random matrix to obtain an initialized perturbation matrix.
[0068] Furthermore, the matrix processing module is specifically used to iteratively optimize the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix using an alternating optimization strategy.
[0069] Specifically, in this embodiment, the matrix acquisition module acquires the adjacency matrix of the target community network including:
[0070] According to the original topological structure of the target community network, obtain the adjacency matrix Where n represents the number of nodes in the network. An element in A is 1 if there is an edge between the two nodes at the corresponding position, and 0 if there is no edge. The specific representation is as follows:
[0071]
[0072] Among them, A ij is the element in the i-th row and j-th column.
[0073] In this embodiment, the matrix processing module constructs the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix of the target community network according to the adjacency matrix, including:
[0074] The adjacency matrix A is subjected to 20 iterations of matrix decomposition update strategy to obtain the initialized original community network membership matrix And the initialized community network membership matrix after the attack And according to the initialized attack community network membership matrix U P Generate the initialized perturbation matrix Where k represents the number of communities. The specific steps are as follows:
[0075] (1) Initialize a random matrix U0, where all elements are not less than 0; iterate 20 times to obtain the original community network membership matrix U, using the following formula:
[0076]
[0077] Among them, ⊙ represents element-by-element multiplication (i.e., Hadamard product), (*) T Represents the transposed matrix of *.
[0078] (2) Set up a random matrix and a matrix of all zeros U P0 All elements are not less than 0; iterate 20 times to get the initialized community network membership matrix U after the attack P And the method of initializing the perturbation matrix P uses the following formula:
[0079]
[0080] Among them, ⊙ represents element-by-element multiplication (i.e., Hadamard product), min(A,AU P0 U P0 T ) means taking matrix A and matrix AU P0 U P0 T The minimum value between corresponding elements.
[0081] In this embodiment, the matrix processing module iteratively optimizes the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix to obtain the optimized disturbance matrix, which includes:
[0082] Set the number of iterations t, adopt the alternating optimization strategy, and iteratively update the perturbation matrix P, the original community network membership matrix U, and the community network membership matrix U after the attack. P , until the number of iterations is reached, and the optimized perturbation matrix P is obtained ′ , specifically including the following:
[0083] (1) Set the number of links e of the community network attack, and use the original community network membership matrix U and the community network membership matrix U after the attack P , the formula for updating the perturbation matrix P is as follows:
[0084]
[0085] (2) Using the perturbation matrix P and the community network membership matrix U after the attack P , the formula for optimizing the original community network membership matrix U is as follows:
[0086]
[0087] Among them, β is the weight coefficient of the similarity between the original community structure and the community structure after the attack, β>0.
[0088] (3) Using the original community network membership matrix U and the perturbation matrix P, optimize the community network membership matrix U after the attack P The formula is as follows:
[0089]
[0090] Among them, α is the weight coefficient of the target attack constraint, which is used to control the intensity of the attack, α>0.
[0091] In this embodiment, the network attack module applies the optimized disturbance matrix to the adjacency matrix to obtain the adjacency matrix after the attack, and completes the attack on the target community network, including:
[0092] Using the adjacency matrix A and the optimized perturbation matrix P ′ , get the adjacency matrix A of the community network after the attack * , the specific formula is as follows:
[0093] A * =A+P ′ .
[0094] In order to further optimize the technical solution, this embodiment further provides a processor, which is used to run a program, wherein the community network attack method based on non-redundant community structure learning is executed when the program is running.
[0095] To further optimize the technical solution, this embodiment also provides an electronic device, including one or more memories and a processor, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the community network attack method based on non-redundant community structure learning.
[0096] Specifically, the electronic device and the program stored and executed in the memory and processor included therein include the following method steps and contents:
[0097] Obtain the adjacency matrix of the target community network;
[0098] Constructing an original community network membership matrix, a community network membership matrix after the attack, and a disturbance matrix of the target community network according to the adjacency matrix, and obtaining an optimized disturbance matrix by iteratively optimizing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix;
[0099] The optimized disturbance matrix is applied to the adjacency matrix to obtain the adjacency matrix after the attack, thereby completing the attack on the target community network.
[0100] Furthermore, constructing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix of the target community network according to the adjacency matrix includes:
[0101] Several random matrices are set, and they are iteratively decomposed to obtain the initialized original community network membership matrix, the attacked community network membership matrix, and the disturbance matrix.
[0102] Furthermore, the random matrix includes a first random matrix, a second random matrix, and a third random matrix. The first random matrix is used to iteratively decompose with the adjacency matrix to obtain an initialized original community network membership matrix, the second random matrix is used to iteratively decompose with the adjacency matrix and the third random matrix to obtain an initialized post-attack community network membership matrix, and the third random matrix is used to iteratively decompose with the adjacency matrix and the second random matrix to obtain an initialized perturbation matrix.
[0103] Furthermore, an alternating optimization strategy is adopted to iteratively optimize the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix.
[0104] The effect of the method proposed in this embodiment is verified as follows:
[0105] 1. Simulation conditions
[0106] This example tests three network data sets, Zachary's Karate Club network (Karate), Dolphins network (Dolphins) and American political books network (Polbooks), and fixes the attack budget for each network data set, and sets the budget for the Karate network, Dophins network and Polbooks network to 10%, 10% and 4% of the edges respectively. The detailed network information is shown in Table 1.
[0107] Table 1
[0108] Network dataset Number of nodes Number of edges Attack Budget Karate 34 78 8(10％) Dolphins 62 159 16(10％) Polbooks 105 441 18(4％)
[0109] The experiment used MATLAB R2022b software and the computer configuration was Intel Core i7-13700F / 2.10GHz / 32G.
[0110] 2. Simulation method
[0111] We selected a random attack method, a community detection attack method, a degree-based attack method, a genetic algorithm-based attack heuristic method, a degree-based attack method, and a local structure-based community detection attack heuristic method. These six community detection attack methods were compared with the results of the method proposed in this embodiment to verify the effectiveness of the method proposed in this embodiment. The six representative methods are:
[0112] Method 1, Method 2, Method 3, and Method 4 are methods proposed by Chen et al. For specific references, please refer to “GA-based Q-attack on community detection[J].IEEE Transactions on Computational Social Systems, 2019, 6(3): 491-503.”;
[0113] Method 5 is proposed by Zhao et al., and the specific reference is "Obfuscating community structure in complex network with evolutionary divide-and-conquer strategy[J].IEEE Transactions on Evolutionary Computation, 2023, 27(6): 1926-1940.";
[0114] Method 6 is proposed by Yang et al. For details, please refer to “LSHA: ALocal Structure-Based Community Detection Attack Heuristic Approach[J]. IEEE Transactions on Computational Social Systems, 2024, 11(2): 2966-2978.”
[0115] The community detection method uses the classic community detection algorithm Walktrap proposed by Pons P to divide the network data set into communities. For specific reference, please refer to "Computing communities in large networks using randomwalks[J].Journal of Graph Algorithms and Applications,2006,10(2):191-218.".
[0116] 3. Simulation content
[0117] In simulation 1, the Walktrap community detection algorithm is first used to perform initial community division on the Karate network, and then methods 1-6 and the method proposed in this embodiment are used to attack the Karate network to generate the attacked network. Then, the Walktrap algorithm is used again to perform community division on the attacked network, and finally the changes in the community structure of the network before and after the attack are compared.
[0118] In simulation 2, the Walktrap community detection algorithm is first used to perform initial community division on the Dophins network, and then methods 1-6 and the method proposed in this embodiment are used to attack the Dophins network to generate the attacked network. Then, the Walktrap algorithm is used again to perform community division on the attacked network, and finally the changes in the community structure of the network before and after the attack are compared.
[0119] In simulation 3, the Walktrap community detection algorithm is first used to perform initial community division on the Polbooks network, and then the methods 1-6 and the method proposed in this embodiment are used to attack the Polbooks network to generate the network after the attack. Next, the Walktrap algorithm is used again to perform community division on the network after the attack, and finally the changes in the community structure of the network before and after the attack are compared.
[0120] The normalized mutual information (NMI) indicator is used to measure the similarity between the two partitions of the community, that is, the NMI between the community structures obtained by the detection algorithm before and after the attack is used to measure the effectiveness of the attack. The higher the value of the NMI indicator, the higher the similarity between the two community partitions. Therefore, the lower the value of the NMI means the better the effect of the attack. The comparison results of the NMI of different community detection attack algorithms are shown in Table 2.
[0121] Table 2
[0122] Network dataset Method 1 Method 2 Method 3 Method 4 Method 5 Method 6 The method proposed in this embodiment Karate 0.7202 0.6591 0.6897 0.695 0.8057 0.6298 0.278442 Dolphins 0.6591 0.5809 0.5854 0.5276 0.6005 0.5251 0.180882 Polbooks 0.9077 0.8796 0.842 0.7054 0.8452 0.8315 0.431076
[0123] As can be seen from Table 2, the method proposed in this embodiment has the lowest NMI value on the three network data sets, which is significantly higher than other methods. For the Karate network data set, the NMI value of the method proposed in this embodiment is 0.2784, which is 44.2%, 38.1%, 41.1%, 41.7%, 52.7% and 35.1% lower than that of method 1, method 2, method 3, method 4, method 5 and method 6, respectively. For the Dophins network data set, the NMI value of the method proposed in this embodiment is 0.1809, which is 47.8%, 40.0%, 40.5%, 34.7%, 42.0% and 34.4% lower than that of method 1, method 2, method 3, method 4, method 5 and method 6, respectively. For the Polbooks network dataset, the NMI value of the method proposed in this embodiment is 0.4311, which is 47.7%, 44.9%, 41.1%, 27.4%, 41.4% and 40.0% lower than that of method 1, method 2, method 3, method 4, method 5 and method 6, respectively.
[0124] In summary, the network attack method based on non-redundant community structure proposed in this embodiment can effectively reduce the similarity of community partitions, and the NMI value on multiple network data sets is significantly lower than other methods, showing an excellent community attack effect. By controlling the disturbance intensity and optimizing the strategy, the integrity of the network structure is maintained while ensuring the attack effect, and it has a strong generalization ability.
[0125] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.
Claims
1. A community network attack method based on non-redundant community structure learning, characterized in that: include: Obtain the adjacency matrix of the target community network; Constructing an original community network membership matrix, a community network membership matrix after the attack, and a disturbance matrix of the target community network according to the adjacency matrix, and obtaining an optimized disturbance matrix by iteratively optimizing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix; The optimized disturbance matrix is applied to the adjacency matrix to obtain the adjacency matrix after the attack, thereby completing the attack on the target community network.
2. The community network attack method based on non-redundant community structure learning according to claim 1 is characterized in that: Constructing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix of the target community network according to the adjacency matrix includes: Several random matrices are set, and they are iteratively decomposed to obtain the initialized original community network membership matrix, the attacked community network membership matrix, and the disturbance matrix.
3. The community network attack method based on non-redundant community structure learning according to claim 2 is characterized in that: The random matrix includes a first random matrix, a second random matrix, and a third random matrix. The first random matrix is used to iteratively decompose with the adjacency matrix to obtain an initialized original community network membership matrix. The second random matrix is used to iteratively decompose with the adjacency matrix and the third random matrix to obtain an initialized post-attack community network membership matrix. The third random matrix is used to iteratively decompose with the adjacency matrix and the second random matrix to obtain an initialized perturbation matrix.
4. The community network attack method based on non-redundant community structure learning according to claim 1 is characterized in that: The original community network membership matrix, the attacked community network membership matrix, and the disturbance matrix are iteratively optimized using an alternating optimization strategy.
5. A community network attack system based on non-redundant community structure learning, characterized in that: include: A matrix acquisition module is used to obtain the adjacency matrix of the target community network; A matrix processing module, used to construct an original community network membership matrix, a community network membership matrix after the attack, and a disturbance matrix of the target community network according to the adjacency matrix, and obtain an optimized disturbance matrix by iteratively optimizing the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix; The network attack module is used to apply the optimized disturbance matrix to the adjacency matrix, obtain the adjacency matrix after the attack, and complete the attack on the target community network.
6. The community network attack system based on non-redundant community structure learning according to claim 5 is characterized in that: The matrix processing module is specifically used to construct the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix of the target community network according to the adjacency matrix, including: Several random matrices are set, and they are iteratively decomposed to obtain the initialized original community network membership matrix, the attacked community network membership matrix, and the disturbance matrix.
7. The community network attack system based on non-redundant community structure learning according to claim 6 is characterized in that: The random matrix includes a first random matrix, a second random matrix, and a third random matrix. The first random matrix is used to iteratively decompose with the adjacency matrix to obtain an initialized original community network membership matrix. The second random matrix is used to iteratively decompose with the adjacency matrix and the third random matrix to obtain an initialized post-attack community network membership matrix. The third random matrix is used to iteratively decompose with the adjacency matrix and the second random matrix to obtain an initialized perturbation matrix.
8. The community network attack system based on non-redundant community structure learning according to claim 5 is characterized in that: The matrix processing module is specifically used to iteratively optimize the original community network membership matrix, the community network membership matrix after the attack, and the disturbance matrix using an alternating optimization strategy.
9. A processor, characterized in that: The processor is used to run a program, wherein the program executes the method according to any one of claims 1 to 4 when running.
10. An electronic device, characterized in that: The invention comprises one or more memories and processors, wherein the memories are used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 4.