Community detection micro-scale adversarial method and system based on node relative importance perturbation

Abstract

The present application relates to network data processing technical field, especially a kind of community detection microscale countermeasure method and system based on node relative importance disturbance, add false user node in the community topology structure of target node belonging to, and according to the link relationship of false user node and original node in community establish false node relative importance candidate link library;Node in community is divided into inferior node and preparatory node, and get inferior gene and preparatory gene;By pruning operation, eliminate inferior gene, retain preparatory gene, use node relative importance as fitness, and filter excellent gene by roulette selection method;With false user node as core, use cross selection to obtain excellent gene chromosome, and the optimal relative importance false link generated by chromosome is added to the community topology structure of target node belonging to.The present application can reach counteracting effect to overlapping community detection and non-overlapping community detection, and can better realize the purpose of hidden target node or community.

Description

Technical Field

[0001] This invention relates to the field of network data processing technology, and in particular to a microscale adversarial method and system for community detection based on node relative importance perturbation. Background Technology

[0002] As community detection algorithms are increasingly sophisticated and applied across various fields, people's information is becoming more and more exposed. Some organizations or individuals may exploit these algorithms to steal others' private information. For example, a key figure in a company with confidential professional information can be precisely located by community detection methods. To prevent such problems, countering community detection methods has become a focus for protecting privacy.

[0003] Adversarial community detection involves altering the community structure through various means or methods to obfuscate detection results and protect the community from incorrect detection, thus addressing privacy concerns. Currently popular adversarial community detection methods primarily focus on community hiding or individual hiding. Most research mainly involves adjusting the connections between nodes—deleting original links and adding new ones—to disrupt the community structure and hide individuals or even the entire community. However, this approach ignores real-world problems. First, most methods only target the relationship between a specific node and its surrounding nodes during the adversarial process, failing to fully utilize the characteristics of a node's association with its community, making the effectiveness of the adversarial approach questionable. Second, deleting actual connections between users is difficult and costly; methods that only consider concealment or adversarial effectiveness without considering the cost cannot fully demonstrate the merits of the strategy. Third, the purpose of adversarial methods is to protect privacy, but deleting connections between users can lead to the loss of potentially important information, violating the original intention of protecting user privacy. Summary of the Invention

[0004] To address this, the present invention provides a micro-scale adversarial method and system for community detection based on node relative importance perturbation, which solves the problems in existing community detection adversarial methods, such as the reduced status of user nodes during the adversarial process due to the lack of consideration of node importance, and the high cost / information loss caused by deleting user connections.

[0005] According to the design scheme provided by this invention, on the one hand, a micro-scale adversarial method for community detection based on node relative importance perturbation is provided, comprising:

[0006] Obtain the community topology of the target node in the social network, add fake user nodes to the community topology of the target node, and establish a candidate link library of relative importance of fake nodes based on the link relationship between fake user nodes and original nodes in the community. The candidate link library of relative importance is used to store the expected links between fake user nodes and all nodes in the community based on the relative importance of nodes. The relative importance of nodes is based on the size of node importance to represent the relative importance of nodes in the community.

[0007] Based on the importance of all nodes in the community, nodes are divided into inferior nodes and reserve nodes. Links in the relative importance candidate link library are divided into inferior genes and reserve genes. Inferior nodes are used to represent nodes whose importance is greater than a threshold, reserve nodes are used to represent nodes whose importance is less than a threshold, inferior genes are used to represent links that are expected to exist between fake user nodes and inferior nodes, and reserve genes are used to represent links that are expected to exist between fake user nodes and reserve nodes.

[0008] The pruning operation removes inferior genes from the candidate link library of relative importance, retains reserve genes, and forms an important gene library. The relative importance of nodes is used as fitness, and excellent genes are selected from the important gene library based on the fitness of user nodes and by roulette wheel selection.

[0009] Using fake user nodes as the core, crossover selection in a genetic algorithm is used to obtain chromosomes formed by excellent genes, and the optimal relative importance fake links generated by the chromosomes are added to the community topology of the target node.

[0010] As a micro-scale adversarial method for community detection based on node relative importance perturbation in this invention, further, obtaining the community topology to which the target node belongs in the social network includes:

[0011] Representing the topology of a social network using undirected graphs;

[0012] By performing community detection on the social network, the undirected graph of the social network is divided into several community subgraphs, and each community subgraph represents a corresponding community population.

[0013] Select the target node according to the user's needs and obtain the community population to which the target node belongs.

[0014] As a micro-scale adversarial method for community detection based on node relative importance perturbation in this invention, further, nodes within the community are divided into inferior nodes and reserve nodes according to the importance of all nodes in the community, including:

[0015] The importance of target nodes in the community is set as a threshold, and the importance of nodes is calculated based on the degree of nodes in the topology.

[0016] Nodes in the community that are not lower than the threshold are classified as inferior nodes, and nodes that are lower than the threshold are classified as reserve nodes.

[0017] As a micro-scale adversarial method for community detection based on node relative importance perturbation in this invention, further, based on user node fitness and using a roulette wheel selection method to screen for superior genes from an importance gene pool, it includes:

[0018] The normalized relative importance function of nodes is used as the node fitness function. The normalized relative importance function of nodes is constructed using the relative importance of nodes, the maximum value of relative importance of nodes in the community and the minimum value of relative importance of nodes. The relative importance of nodes is represented by the ratio of the node importance to the number of nodes in the community that are not less than the node importance.

[0019] The fitness of each node is calculated using the node fitness function, and the total fitness of all nodes in the community is obtained by summing the fitness of each node.

[0020] The ratio of the fitness of a user node to the total fitness is used as the probability of each user node being selected, and the cumulative probability of each user node before it is selected is obtained based on the probability of the user node being selected.

[0021] Random numbers are generated, and user nodes are selected based on cumulative probability and random numbers. The expected links between the selected user nodes and fake nodes are screened from the importance gene library as excellent genes.

[0022] As a micro-scale adversarial method for community detection based on node relative importance perturbation in this invention, it further utilizes crossover selection in genetic algorithms to obtain chromosomes with excellent genes, including:

[0023] Using spurious nodes as the core, excellent genes are interconnected to form a new gene pool;

[0024] Genes in a new gene pool are used to form chromosomes by using single-point crossover and multi-point crossover. The single-point crossover is a crossover consisting of two genes, and the multi-point crossover is a crossover consisting of two or more genes.

[0025] As a micro-scale adversarial method for community detection based on node relative importance perturbation in this invention, the optimal relative importance spurious links generated by chromosomes are further added to the community topology of the target node, including:

[0026] The chromosome is inserted into the community topology, and the target node hiding effect index is calculated based on the relative importance of the target node before and after the confrontation, the number of the same nodes before and after the confrontation, and the number of all nodes before and after the confrontation contained in the community to which the target node belongs.

[0027] The overall effectiveness of the confrontation is calculated based on the target node hiding effect index and the difference in community division before and after the confrontation.

[0028] Based on the overall effect of the adversarial process, the optimal chromosome is selected, and the spurious links corresponding to the optimal chromosome are taken as the optimal relative importance spurious links. The optimal relative importance spurious links are then added to the community topology of the target node.

[0029] As a micro-scale adversarial method for community detection based on node relative importance perturbation in this invention, the calculation process of the target node hiding effect index is further expressed as follows: Where α and β are weighting coefficients, and α+β=1, C(n) represents the change in the relative importance of the target node n before and after the confrontation, Q(m) represents the number of the same nodes before and after the confrontation, and Q(n) represents the number of all nodes included in the community to which the target node belongs before and after the confrontation.

[0030] On another front, this invention also provides a micro-scale adversarial system for community detection based on node relative importance perturbation, comprising: a community detection module, a node partitioning module, a link filtering module, and a link adding module, wherein...

[0031] The community detection module is used to obtain the community topology to which the target node belongs in the social network, add fake user nodes to the community topology structure to which the target node belongs, and establish a candidate link library of relative importance of fake nodes based on the link relationship between the fake user nodes and the original nodes in the community. The candidate link library of relative importance is used to store the links that are expected to exist between the fake user nodes and all nodes in the community based on the relative importance of the nodes. The relative importance of the nodes is based on the size of the node importance to represent the relative importance of the node in the community.

[0032] The node partitioning module is used to divide nodes in the community into inferior nodes and prospective nodes based on the importance of all nodes in the community, and to divide links in the relative importance candidate link library into inferior genes and prospective genes. The inferior nodes are used to represent nodes whose importance is greater than a threshold, the prospective nodes are used to represent nodes whose importance is less than a threshold, the inferior genes are used to represent links that are expected to exist between fake user nodes and inferior nodes, and the prospective genes are used to represent links that are expected to exist between fake user nodes and prospective nodes.

[0033] The link filtering module is used to remove inferior genes from the candidate link library of relative importance through pruning operations, retain reserve genes, and form an important gene library. The relative importance of nodes is used as fitness, and excellent genes are selected from the important gene library based on the fitness of user nodes and through roulette wheel selection.

[0034] The link addition module is used to obtain chromosomes with excellent genes by using crossover selection in genetic algorithms, with fake user nodes as the core, and add the best relative importance fake links generated by the chromosomes to the community topology of the target node.

[0035] The beneficial effects of this invention are:

[0036] This invention, based on genetic algorithms, reduces the relative importance of target nodes at minimal cost by adding only a small number of spurious nodes and links within the community without deleting the original structure. This alters the community partitioning result, establishes a candidate link library based on node relative importance to avoid confusion with real links, uses pruning principles to remove inferior genes composed of high-importance nodes and spurious nodes within the community, and then employs a roulette wheel method for double screening of reserve genes. The reserve genes are then combined and cross-crossed to find the optimal spurious links, improving the accuracy of gene selection and minimizing the relative importance of target users. This changes the target node community partitioning result, thereby protecting the privacy information of target users. Furthermore, this invention conducts experimental tests on multiple real-world network datasets using evaluation metrics, comparing various community detection algorithms with four existing adversarial community detection methods. Experimental results demonstrate that the proposed GIM scheme significantly outperforms traditional baseline methods, achieving adversarial effects against both overlapping and non-overlapping community detection, and shows promising application prospects in the field of community detection. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the micro-scale adversarial process for community detection based on the perturbation of the relative importance of nodes in the embodiment;

[0038] Figure 2 This example illustrates the relative importance of nodes within the community and the effectiveness of adversarial actions.

[0039] Figure 3 This is a schematic diagram of the GIM algorithm flow in this embodiment;

[0040] Figure 4 This example illustrates the spurious node candidate link library and pruning principle.

[0041] Figure 5 This is a schematic diagram illustrating the combination of cross-processing and iterative selection in the embodiment;

[0042] Figure 6 This is a comparison of the hiding effect parameters in the embodiment;

[0043] Figure 7 This example illustrates the performance of various methods on different datasets for NMI values. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer and more understandable, the invention will be further described in detail below with reference to the accompanying drawings and technical solutions.

[0045] User importance is one of the key criteria for community monitoring. Different users exhibit varying degrees of importance within a community. However, social networks often aim to highlight users, and deleting connections between users cannot diminish a user's visibility. (See embodiments of this invention.) Figure 1 As shown, a micro-scale adversarial method for community detection based on node relative importance perturbation is provided, comprising:

[0046] S101. Obtain the community topology of the target node in the social network, add fake user nodes to the community topology of the target node, and establish a candidate link library of relative importance of fake nodes based on the link relationship between the fake user nodes and the original nodes in the community. The candidate link library of relative importance is used to store the expected links between the fake user nodes and all nodes in the community based on the relative importance of the nodes. The relative importance of the nodes is based on the size of the node importance to represent the relative importance of the node in the community.

[0047] The importance of nodes within a community not only affects the community partitioning result but also influences the community's overall structure, function, and dynamic characteristics. Therefore, in-depth analysis of node importance is a key factor in improving the quality and accuracy of partitioning. It can be defined and measured from multiple perspectives. Common measurement methods include:

[0048] 1. Centrality

[0049] (1) Degree centrality: The number of connections a node has. The more connections a node has, the more important it is in the community.

[0050] (2) Betweenness centrality: the frequency with which a node appears in the shortest path between other nodes. The more frequent the occurrence, the more important the node is.

[0051] (3) Proximity centrality: the average distance between a node and other nodes. The shorter the distance, the higher the importance.

[0052] 2. Community influence

[0053] (1) Leader node: A node that can influence other nodes or play a guiding role in the community.

[0054] (2) Bridging nodes: Nodes that connect different communities and play a key role in information flow.

[0055] 3. Degree of aggregation

[0056] The aggregation degree of a node within a community, that is, the connection strength with its neighbors, reflects the degree of closeness of the node within the community.

[0057] In addition to the three methods mentioned above, there are also similarity, dissemination ability, stability, and so on.

[0058] Without compromising the importance of users, this solution utilizes the relative importance of nodes within the community to assess the relative importance of nodes in the community.

[0059] The process of calculating the relative importance of nodes can be summarized as: within this community, the ratio of a node's importance to the number of nodes with an importance no less than that node. Let I(n) represent this.

[0060]

[0061] Where n represents a user node, i(n) represents the importance of the node, and N(n) represents the number of nodes in the community where the node is located that are no less important than node i(n).

[0062] like Figure 2 As shown in the blue community on the left, assuming degree as importance, node 1 has i(1) = 4, N(1) = 1, I(1) = 4; node 2 has i(2) = 3, N(2) = 2, I(2) = 1.5.

[0063] The relative importance of a node effectively reflects its relative significance within a community and influences the community segmentation results. Nodes with high relative importance typically play a crucial role in community detection and may become core or bridging nodes. Community detection algorithms often prioritize these nodes to ensure the segmented community structure is reasonable and representative. Therefore, analyzing the relative importance of nodes allows for a better understanding of the functions and dynamics within a community, helping to optimize community management strategies and resource allocation. The magnitude of a node's relative importance plays a significant role in community segmentation, and a decrease in a node's relative importance does not diminish a user's original status or importance within the social network.

[0064] There are two methods to reduce the relative importance of a node: one is to delete some links of the target node to reduce the connection between the target node and other nodes, which does not conform to the original intention of protection; the other is to reduce the relative importance of the target node by increasing the importance of nodes with lower importance, which conforms to the original intention of the solution in this case embodiment.

[0065] In obtaining the community topology of a target node in a social network, an undirected graph can be used to represent the social network topology. By performing community detection on the social network, the undirected graph of the social network is divided into several community subgraphs, and each community subgraph represents a corresponding community population. The target node is selected according to the user's needs, and the community population to which the target node belongs is obtained.

[0066] like Figure 3 As shown, the entire social network structure G is treated as a single community. First, a community detection algorithm is used to divide the network into communities, and target nodes are selected based on user needs. Fake user nodes are added within the communities of the target nodes. A candidate link library based on relative importance is established using the possible links between the fake nodes and the original nodes in the communities. A pruning principle is used to remove low-quality links created by high-importance nodes and fake nodes. The remaining candidate links are then further screened using a roulette wheel approach, followed by a comprehensive crossover process to form chromosomes. Finally, a dual-index constraint method is used to generate the optimal relative importance fake links, which are then inserted into the original structure. This reduces the relative importance of the target node at minimal cost, affecting the community division results and thus achieving target node hiding.

[0067] Let G = (N, E) represent a complete set of real-world networks. Using the community detection algorithm Aa, the original graph G = (N, E) is divided into K communities C. i =(N i E i ), i = 1, 2, ..., K, where C i ={C1,C2,...,C a} represents a set of divided communities, with each community representing a population. N i ={n1,n2,...,n α} represents the set of nodes in the community, E i ={e1,e2,...,e β} represents a set of links between nodes. The meanings of each symbol are shown in Table 1.

[0068] Table 1 Symbol Table

[0069]

[0070]

[0071] The approach involves adding only fake nodes N within the community. f and link E f This approach counters the structural imbalance by adding dummy nodes within the target node's community. These communities are not limited to overlapping or non-overlapping communities; they are solely the target node's community. After adding these dummy nodes, the target node's community is updated, which can be represented as:

[0072] C t =C t ∪{N f} (2)

[0073] Adding a fake node N f At the same time, this node may interact with all nodes N within the community. jThe link is connected, and this possible link is set as gene G. f The set of genes formed in this way is defined as the relative importance candidate link library (RIL).

[0074] RIL={(N f N j )|N j ∈C t} (3)

[0075] This library is a collection based on a set of possible links. This definition method provides more comprehensive coverage of all user nodes within the community, preventing the loss of important user link information and avoiding confusion between real links that could affect the selection of the optimal individual. Figure 4 As shown, assuming node 1 is the target node and n is a dummy node, n may be connected to nodes 2, 3, 4, 5, 6, 7, 8, and 9. Such possible links are defined as the relative importance candidate link library of node 1.

[0076] S102. Based on the importance of all nodes in the community, the nodes in the community are divided into inferior nodes and reserve nodes, and the links in the relative importance candidate link library are divided into inferior genes and reserve genes. The inferior nodes are used to represent nodes whose importance is greater than a threshold, the reserve nodes are used to represent nodes whose importance is less than a threshold, the inferior genes are used to represent links that are expected to exist between fake user nodes and inferior nodes, and the reserve genes are used to represent links that are expected to exist between fake user nodes and reserve nodes.

[0077] Specifically, the importance I(n) of all nodes in the community can be calculated, and the importance T of the target node can be set as a threshold. Nodes are then categorized into low-quality nodes and potential nodes based on the threshold. Nodes not lower than the threshold are defined as low-quality nodes, and nodes lower than the threshold are defined as potential nodes.

[0078]

[0079] The link between spurious nodes and low-quality nodes is called the low-quality gene I. bad The link with the preparatory node is called the preparatory gene I. pre .

[0080]

[0081] S103. Through pruning operations, inferior genes in the candidate link library of relative importance are removed, reserve genes are retained, and an important gene library is formed. The relative importance of nodes is used as fitness, and excellent genes are screened from the important gene library based on the fitness of user nodes and through roulette wheel selection.

[0082] The principle of pruning is used to remove inferior genes and retain reserve genes. Therefore, the important gene pool RIL after pruning can be updated.

[0083] RIL'=I pre (6)

[0084] like Figure 4 As shown, nodes 3, 5, 8, and 9 are more important than target node 1 within their communities. These nodes are called inferior nodes, and any links between false nodes and these nodes are called inferior genes. The pruning principle is used to remove these inferior genes, retaining the potential genes.

[0085] Using the pruning principle to initially screen for genes can reduce the workload of genetic algorithms in selecting optimal nodes and links, thereby improving selection efficiency and accuracy.

[0086] Among them, the selection of superior genes from the importance gene pool based on user node fitness and using a roulette wheel selection method can be designed to include:

[0087] The normalized relative importance function of nodes is used as the node fitness function. The normalized relative importance function of nodes is constructed using the relative importance of nodes, the maximum value of relative importance of nodes in the community and the minimum value of relative importance of nodes. The relative importance of nodes is represented by the ratio of the node importance to the number of nodes in the community that are not less than the node importance.

[0088] The fitness of each node is calculated using the node fitness function, and the total fitness of all nodes in the community is obtained by summing the fitness of each node.

[0089] The ratio of the fitness of a user node to the total fitness is used as the probability of each user node being selected, and the cumulative probability of each user node before it is selected is obtained based on the probability of the user node being selected.

[0090] Random numbers are generated, and user nodes are selected based on cumulative probability and random numbers. The expected links between the selected user nodes and fake nodes are screened from the importance gene library as excellent genes.

[0091] The normalized relative importance function is used as the fitness function. I norm (n) represents.

[0092]

[0093] Where I(n) represents the relative importance of the target node, I max (n) represents the value of the node with the highest relative importance in this community, I min (n) represents the value of the node with the lowest relative importance in the community.

[0094] The process of secondary gene screening is also a process of selection. In this case, roulette wheel selection was used for further gene selection, assigning each individual a "sector" whose area is proportional to its fitness value. Then, by rotating the roulette wheel and randomly selecting a point, the selection of which sector is chosen is determined. The steps of the roulette wheel selection method can be summarized as follows:

[0095] (1) First, calculate the fitness I of each user node. norm (i). Assuming there are n user nodes, their fitness is I. norm (1) I norm (2),...,I norm (n).

[0096] (2) Next, calculate the total fitness F of all users in the community.

[0097]

[0098] (3) Calculate the selection probability P for each user node. i This is the ratio of the fitness of a user node to the total fitness.

[0099]

[0100] (4) Calculate the cumulative probability Q i .

[0101]

[0102] Where Q1 = P1, Q2 = P1 + P2, and so on.

[0103] (5) Generate a random number r in [0,1], and then find the first one that satisfies Q. i If the value of i is greater than or equal to r, then individual i is selected.

[0104] Individuals with higher fitness occupy a larger sector area, and therefore have a higher probability of being selected.

[0105] By employing a two-stage selection process, the probability of selecting superior genes is increased. This two-stage selection method is more efficient than ordinary selection methods, effectively reducing the complexity of the entire process and improving GIM's effectiveness against genetic engineering.

[0106] S104. Using fake user nodes as the core, cross-selection in the genetic algorithm is used to obtain chromosomes formed by excellent genes, and the optimal relative importance fake links generated by the chromosomes are added to the community topology of the target node.

[0107] Depending on the number of nodes and links to be added, this embodiment combines single-point and multi-point crossover methods to form chromosomes from the new gene pool. The change in degree of the original target node caused by the crossover chromosome is calculated, and the most suitable offspring chromosome is retained. This crossover process uses dummy nodes N... f Using the selected candidate genes as the core, the selected genes are interconnected. A crossover involving two genes is called a single-point crossover, while a crossover involving more than two genes is called a multi-point crossover. This can be represented as:

[0108] E = {(G qualityi N f ),(N f G qualityj (11)

[0109] When i = j = 1, it means that the crossover has only two genes with spurious node N. f With the core single-point intersection, when This indicates that the crossover occurs when at least three genes are associated with spurious nodes N. f Multiple intersections centered around a core element. For example... Figure 5 As shown.

[0110] This method is more versatile, allowing for the selection of a more suitable crossover method based on the number of nodes and links added. Compared to a typical single crossover method, this approach can be applied to a wider range of scenarios and situations, and is also more effective in generating optimal chromosomes.

[0111] The process of adding the optimal relative importance spurious links generated by chromosomes to the community topology of the target node can include:

[0112] The chromosome is inserted into the community topology, and the target node hiding effect index is calculated based on the relative importance of the target node before and after the confrontation, the number of the same nodes before and after the confrontation, and the number of all nodes before and after the confrontation contained in the community to which the target node belongs.

[0113] The overall effectiveness of the confrontation is calculated based on the target node concealment effect index and the differences in community division before and after the confrontation.

[0114] Based on the overall effect of the adversarial process, the optimal chromosome is selected, and the spurious links corresponding to the optimal chromosome are taken as the optimal relative importance spurious links. The optimal relative importance spurious links are then added to the community topology of the target node.

[0115] The selected chromosomes are inserted into the original structure, and community detection is performed again to calculate the hiding effect R(n). The number of iterations t and the threshold r for the hiding effect are set. A qualified chromosome is defined as one whose hiding effect R(n) is not less than the threshold r. The termination condition is reaching the constant number of iterations t. This iterative selection can further narrow down the range of high-quality chromosomes.

[0116] Let C(n) represent the change in relative importance, where I'(n) represents the relative importance of the target node after the adversarial process.

[0117]

[0118] Let R(n) represent the hiding effect of the target node. After the adversarial process, the nodes in the target node's community are compared with the original community nodes. The sum of the changes in the proportion of the same nodes and their relative importance represents the hiding effect.

[0119]

[0120] Where α+β=1, Q(m) represents the number of identical nodes m after comparison, and Q(n) represents the number of all nodes in the target node community before and after the confrontation.

[0121] In this embodiment, the hiding effect metric D(n) is used to reflect the changes in the number and relative importance of the original nodes within the target node's community in the graph structure after adversarial processing. The more nodes that are not in the original community within the community, and the greater the decrease in relative importance, the better the hiding effect. NMI is used to constrain the number of added nodes, minimizing changes in the graph structure and achieving stealth in adversarial processing.

[0122] D(n)=γR(n)+δNMI (14)

[0123] Here, γ+δ=1, and the values ​​of γ and δ can be adjusted according to the actual concealment effect and the proportion of the strategy's superiority or inferiority based on NMI. NMI represents Normalized Mutual Information, used to evaluate the difference between the two community divisions before and after the adversarial process, and also represents the concealment of the adversarial process. The value of D(n) is (0,1), the closer it is to 0, the worse the overall adversarial effect, and conversely, the closer it is to 1, the better the overall adversarial effect.

[0124] The high-quality genes obtained from the above process are inserted into the original structure to generate an adversarial network G' = (N', E'). The same detection algorithm Aa is used again to partition the graph to obtain a new structural graph, resulting in a new community C' = (c′1, c′2, ..., c′). a ). Contains target node N t The community is represented as Cx'=(n j N t ,e j ), where n i ≠n j e i ≠e jThe relative importance of the target node is I'(x), and I'(x) < I(x). After the adversarial process, the number of community nodes and edges containing the target node changes, and the relative importance of the target node decreases.

[0125] B(n) can be used to represent the cost of the strategy, and A(n) can be used to represent the superiority or inferiority of the strategy. This indicator overcomes the shortcomings of traditional evaluation indicators by incorporating three factors: concealment effect, structural influence, and adversarial cost, thus providing a more comprehensive assessment of the strategy's merits. Furthermore, this indicator formula can be applied to various scenarios for different adversarial testing strategies, demonstrating its universality.

[0126] The cost of adding and deleting edges can be expressed as:

[0127]

[0128] Where N(E) + ) represents the number of edges added, N(E) - The denot represents the number of edges to be deleted, and N is the total number of edges in the entire network structure. k is a constant because deleting edges is difficult in reality, and the greater the cost, the more difficult the edge deletion becomes. Therefore, the value of k can be used to adjust the difficulty parameter of edge deletion.

[0129] The overall evaluation index can be defined as:

[0130] A (n) =D (n) -B (n) (16)

[0131] B (n) This is a purely burdensome behavior parameter used to measure the cost of adversarial actions. This formula can be used to evaluate the merits of a method.

[0132] Furthermore, based on the above method, this embodiment of the invention also provides a micro-scale adversarial system for community detection based on node relative importance perturbation, comprising: a community detection module, a node partitioning module, a link filtering module, and a link adding module, wherein,

[0133] The community detection module is used to obtain the community topology to which the target node belongs in the social network, add fake user nodes to the community topology structure to which the target node belongs, and establish a candidate link library of relative importance of fake nodes based on the link relationship between the fake user nodes and the original nodes in the community. The candidate link library of relative importance is used to store the links that are expected to exist between the fake user nodes and all nodes in the community based on the relative importance of the nodes. The relative importance of the nodes is based on the size of the node importance to represent the relative importance of the node in the community.

[0134] The node partitioning module is used to divide nodes in the community into inferior nodes and prospective nodes based on the importance of all nodes in the community, and to divide links in the relative importance candidate link library into inferior genes and prospective genes. The inferior nodes are used to represent nodes whose importance is greater than a threshold, the prospective nodes are used to represent nodes whose importance is less than a threshold, the inferior genes are used to represent links that are expected to exist between fake user nodes and inferior nodes, and the prospective genes are used to represent links that are expected to exist between fake user nodes and prospective nodes.

[0135] The link filtering module is used to remove inferior genes from the candidate link library of relative importance through pruning operations, retain reserve genes, and form an important gene library. The relative importance of nodes is used as fitness, and excellent genes are selected from the important gene library based on the fitness of user nodes and through roulette wheel selection.

[0136] The link addition module is used to obtain chromosomes with excellent genes by using crossover selection in genetic algorithms, with fake user nodes as the core, and add the best relative importance fake links generated by the chromosomes to the community topology of the target node.

[0137] To verify the effectiveness of this solution, the following explanation is based on experimental data:

[0138] The proposed algorithm GIM was tested on four real-world networks using three common community detection methods to examine the hiding effect of target nodes.

[0139] Among them, the Louvain algorithm is a modularity-based community detection algorithm that aims to identify modular structures in a network and divide communities based on the identification of modular structures. The LPA algorithm is a label propagation-based community detection algorithm that assigns labels to nodes and allows these labels to propagate throughout the network. Once stable, nodes with the same labels are grouped into a community. The CPM algorithm is an overlapping community detection algorithm that detects communities by searching for adjacent cliques. Its basic idea is to discover community structures by finding maximal complete subgraphs (cliques) in the network.

[0140] The dataset and the results of overlapping community partitioning are shown in Table 2.

[0141] Table 2. Number of Real Social Networks and Community Divisions

[0142]

[0143] The algorithm GIM in this case is compared with the following four baseline methods.

[0144] Safeness-Based Attacks (SBA): By identifying a certain number of community member links and perturbing the community structure by deleting edges within communities and adding new links between communities, the target node's community is hidden.

[0145] Q-Attack: This method uses modularity Q to design a fitness function and employs a genetic algorithm to delete old links and reconnect a small number of new links to attack community detection methods, thereby achieving community deception and countering community detection.

[0146] DICE: This method uses the idea of ​​heuristic rewiring. By deleting specified internal links and connecting neighboring nodes in the community to distant nodes, the target node becomes invisible and is hidden in the community.

[0147] ProHiCo: This method introduces the idea of ​​likelihood minimization, randomly and fairly allocates perturbation resources, and then selects appropriate edges for perturbation through the likelihood minimization method.

[0148] The hiding effect metric R(n) incorporates changes in relative importance C(n) and the community of the target node. The impact of different parameter values ​​α and β on the hiding effect can be compared by adding different edges. Figure 6 As shown, the effect of adding different numbers of edges on the hiding effect follows a normal distribution. Therefore, it can be deduced that adding different numbers of edges, with α = β = 0.5, achieves the optimal hiding effect. Thus, in the experiment, parameters α and β were set to 0.5.

[0149] The experiment aims to achieve a good hiding effect and optimal overall performance with minimal cost and fewer added nodes and edges. Therefore, it is necessary to determine the optimal number of edges to add. After partitioning different real-world networks using different community detection algorithms, a genetic algorithm is used to add varying numbers of optimal nodes and edges to calculate the number of communities to which the target node belongs and its impact on the structure. The NMI value is set to 0-1, with values ​​closer to 1 indicating greater similarity between the two community structures. Figure 7The bar chart shows significant differences in NMI values ​​across these real-world network datasets, with the Q-attack method exhibiting the most pronounced variation. This indicates that Q-attack has the greatest impact on the structure. GIM, because it needs to consider the adversarial effect, the impact on the structure, and the cost of the adversarial process, doesn't have a completely outstanding NMI value in this dataset comparison, but it still performs relatively well. To balance the concealment effect and the impact on the structure, both γ and δ are set to 0.5. This setting ensures that the concealment effect is achieved while also considering the impact on the structure. If γ > δ, the emphasis is on the concealment effect, with a lighter weight given to the impact on the structure, neglecting the concealment of the overall adversarial process in the real world, resulting in a significant deviation in the overall community structure. If γ < δ, the emphasis is on the impact on the structure, which, while achieving a smaller change in the structure, fails to achieve the required concealment effect on target user nodes. Therefore, setting γ = δ = 0.5 achieves the best overall effect by maximizing the concealment of target users while minimizing the impact on the overall community structure.

[0150] Different nodes were selected as target user nodes, and the optimal number of edges used for each dataset and algorithm adversarial are shown in Table 3.

[0151] Table 3 Optimal number of edges used in adversarial scenarios against different algorithms and datasets (1) Optimal number of edges required for the target user (node ​​13) in adversarial scenarios against different community detection algorithms

[0152]

[0153] (2) The optimal number of edges to be added for the target user, node 1, in the adversarial process of different community detection algorithms.

[0154]

[0155]

[0156] (3) The optimal number of edges to be added for the target user, node 2, in the adversarial process of different community detection algorithms.

[0157]

[0158] During the experiments, adversarial exercises were conducted against different target nodes, revealing significant differences in results across different datasets. For communities with similar degrees and denser connections, fewer edges were used in the adversarial process. Conversely, adversarial exercises against communities with significantly different degrees required more links and incurred greater costs.

[0159] The NMI value ranges from 0 to 1; the closer it is to 1, the more similar the two community structures are. Figure 7The bar chart shows that the NMI values ​​of these real network datasets vary greatly, with the Q-attack method showing the most significant difference, indicating that this method has the greatest impact on the structure.

[0160] Because GIM needs to consider the countermeasure effect, the impact on the structure, and the cost of countermeasure, the NMI value is not completely outstanding in this set of data comparisons, but it can still show relatively good performance.

[0161] Based on the experimental results, the overall evaluation A(n) was calculated to assess the effectiveness of the countermeasure method in this case, and compared with four overlapping community detection methods. (See Table 4.)

[0162] Table 4 Overall evaluation indicators for different methods

[0163]

[0164] The value of A(n) ranges from (0,1). The closer A(n) is to 0, the worse the overall effect of the strategy on the target node, community structure changes, and costs; conversely, the closer it is to 1, the better. The data in the table clearly shows that in the adversarial test against different community detection methods on real network data, the GIM scheme in this case outperforms the optimal SBA scheme by 17%, significantly outperforming other adversarial methods. During the adversarial process, it was found that other methods exhibited significant errors when adversarial against the CPM overlapping community detection algorithm. This indicates that these methods are primarily designed for adversarial testing against non-overlapping community detection algorithms, suggesting that the GIM method is also applicable to adversarial testing against overlapping community detection algorithms. Therefore, this method is applicable to both overlapping and non-overlapping community detection.

[0165] The experimental results above show that the proposed solution takes the target node as an opportunity and reconstructs the data structure by adding fake nodes and edges to target overlapping communities. This approach is more realistic, less costly, and can achieve the desired effect against both overlapping and non-overlapping community detection, thereby achieving the goal of hiding the target node or community.

[0166] Unless otherwise specifically stated, the relative steps, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention.

[0167] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0168] The units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations are not considered to be beyond the scope of this invention.

[0169] Those skilled in the art will understand that all or part of the steps in the above methods can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Optionally, all or part of the steps in the above embodiments can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiments can be implemented in hardware or as a software functional module. This invention is not limited to any particular combination of hardware and software.

[0170] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A micro-scale adversarial method for community detection based on node relative importance perturbation, characterized in that, Include: Obtain the community topology of the target node in the social network, add fake user nodes to the community topology of the target node, and establish a candidate link library of relative importance of fake nodes based on the link relationship between fake user nodes and original nodes in the community. The candidate link library of relative importance is used to store the expected links between fake user nodes and all nodes in the community based on the relative importance of nodes. The relative importance of nodes is based on the size of node importance to represent the relative importance of nodes in the community. Based on the importance of all nodes in the community, nodes are divided into inferior nodes and reserve nodes. Links in the relative importance candidate link library are divided into inferior genes and reserve genes. Inferior nodes are used to represent nodes whose importance is greater than a threshold, reserve nodes are used to represent nodes whose importance is less than a threshold, inferior genes are used to represent links that are expected to exist between fake user nodes and inferior nodes, and reserve genes are used to represent links that are expected to exist between fake user nodes and reserve nodes. The pruning operation removes inferior genes from the candidate link library of relative importance, retains reserve genes, and forms an important gene library. The relative importance of nodes is used as fitness, and excellent genes are selected from the important gene library based on the fitness of user nodes and by roulette wheel selection. Using fake user nodes as the core, crossover selection in a genetic algorithm is used to obtain chromosomes formed by excellent genes, and the optimal relative importance fake links generated by the chromosomes are added to the community topology of the target node.

2. The micro-scale adversarial method for community detection based on node relative importance perturbation according to claim 1, characterized in that, Retrieve the community topology of the target node in the social network, including: Representing the topology of a social network using undirected graphs; By performing community detection on the social network, the undirected graph of the social network is divided into several community subgraphs, and each community subgraph represents a corresponding community population. Select the target node according to the user's needs and obtain the community population to which the target node belongs.

3. The micro-scale adversarial method for community detection based on node relative importance perturbation as described in claim 1, characterized in that, Based on the importance of all nodes in the community, the nodes are divided into low-quality nodes and reserve nodes, including: The importance of target nodes in the community is set as a threshold, and the importance of nodes is calculated based on the degree of nodes in the topology. Nodes in the community that are not lower than the threshold are classified as inferior nodes, and nodes that are lower than the threshold are classified as reserve nodes.

4. The micro-scale adversarial method for community detection based on node relative importance perturbation according to claim 1, characterized in that, Based on user node fitness, superior genes are selected from the importance gene pool using a roulette wheel selection method, including: The normalized relative importance function of nodes is used as the node fitness function. The normalized relative importance function of nodes is constructed using the relative importance of nodes, the maximum value of relative importance of nodes in the community and the minimum value of relative importance of nodes. The relative importance of nodes is represented by the ratio of the node importance to the number of nodes in the community that are not less than the node importance. The fitness of each node is calculated using the node fitness function, and the total fitness of all nodes in the community is obtained by summing the fitness of each node. The ratio of the fitness of a user node to the total fitness is used as the probability of each user node being selected, and the cumulative probability of each user node before it is selected is obtained based on the probability of the user node being selected. Random numbers are generated, and user nodes are selected based on cumulative probability and random numbers. The expected links between the selected user nodes and fake nodes are screened from the importance gene library as excellent genes.

5. The micro-scale adversarial method for community detection based on node relative importance perturbation according to claim 1, characterized in that, Using crossover selection in genetic algorithms to obtain chromosomes with superior gene formation includes: Using spurious nodes as the core, excellent genes are interconnected to form a new gene pool; Genes in a new gene pool are used to form chromosomes by using single-point crossover and multi-point crossover. The single-point crossover is a crossover consisting of two genes, and the multi-point crossover is a crossover consisting of two or more genes.

6. The micro-scale adversarial method for community detection based on node relative importance perturbation according to claim 1 or 5, characterized in that, Add the optimal relative importance pseudo-links generated by the chromosome to the community topology of the target node, including: The chromosome is inserted into the community topology, and the target node hiding effect index is calculated based on the relative importance of the target node before and after the confrontation, the number of the same nodes before and after the confrontation, and the number of all nodes before and after the confrontation contained in the community to which the target node belongs. The overall effectiveness of the confrontation is calculated based on the target node concealment effect index and the differences in community division before and after the confrontation. Based on the overall effect of the adversarial process, the optimal chromosome is selected, and the spurious links corresponding to the optimal chromosome are taken as the optimal relative importance spurious links. The optimal relative importance spurious links are then added to the community topology of the target node.

7. The micro-scale adversarial method for community detection based on node relative importance perturbation according to claim 6, characterized in that, The calculation process for the target node hiding effect metric is expressed as follows: Where α and β are weighting coefficients, and α+β=1, C(n) represents the change in the relative importance of the target node n before and after the confrontation, Q(m) represents the number of the same node m before and after the confrontation, and Q(n) represents the number of all nodes in the target node's community before and after the confrontation.

8. A micro-scale adversarial system for community detection based on node relative importance perturbation, characterized in that, It includes: a community detection module, a node division module, a link filtering module, and a link adding module. The community detection module is used to obtain the community topology to which the target node belongs in the social network, add fake user nodes to the community topology structure to which the target node belongs, and establish a candidate link library of relative importance of fake nodes based on the link relationship between the fake user nodes and the original nodes in the community. The candidate link library of relative importance is used to store the links that are expected to exist between the fake user nodes and all nodes in the community based on the relative importance of the nodes. The relative importance of the nodes is based on the size of the node importance to represent the relative importance of the node in the community. The node partitioning module is used to divide nodes in the community into inferior nodes and prospective nodes based on the importance of all nodes in the community, and to divide links in the relative importance candidate link library into inferior genes and prospective genes. The inferior nodes are used to represent nodes whose importance is greater than a threshold, the prospective nodes are used to represent nodes whose importance is less than a threshold, the inferior genes are used to represent links that are expected to exist between fake user nodes and inferior nodes, and the prospective genes are used to represent links that are expected to exist between fake user nodes and prospective nodes. The link filtering module is used to remove inferior genes from the candidate link library of relative importance through pruning operations, retain reserve genes, and form an important gene library. The relative importance of nodes is used as fitness, and excellent genes are selected from the important gene library based on the fitness of user nodes and through roulette wheel selection. The link addition module is used to obtain chromosomes with excellent genes by using crossover selection in genetic algorithms, with fake user nodes as the core, and add the best relative importance fake links generated by the chromosomes to the community topology of the target node.

9. An electronic device, characterized in that, include: At least one processor, and a memory coupled to said at least one processor; The memory stores a computer program that can be executed by the at least one processor to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, enables the implementation of the method as described in any one of claims 1 to 7.

Images