Structural information-oriented social network community information hiding method and device

Abstract

The invention discloses a structural information-oriented social network community information hiding method and device, and the method comprises the steps: 1) collecting social network data, and obtaining user node topology information; 2) constructing a node-embedded dictionary NR; 3) searching an optimal network reconnection strategy by using a genetic algorithm based on a feature similarity guided bit mutation operator; and 4) generating an adversarial network # imgabs0 # to realize community hiding. According to the method, attention on local structure information is enhanced, a novel individual genotype based on feature embedding is designed, a fine-grained bit variation operator guided by feature similarity is introduced in combination with topological information, and then a multi-granularity variation strategy with coarse and fine granularity variation coupled is provided; the method can effectively balance the global exploration and local search capabilities of the algorithm, overcomes the difficulty problem that the existing method lacks an optimization guidance strategy and a solution set is easy to fall into local optimum, and achieves a community hiding target.

Description

Technical Field

[0001] The present invention relates to the fields of community detection, community hiding and evolutionary computing technology, and in particular to a method and device for hiding community information in a social network guided by structural information. Background Art

[0002] With the rapid development of the Internet and mobile communication technologies, social networks have become an indispensable part of people's daily lives. Users interact on social platforms, forming complex and rich community structures. These community structures reflect the interests, social relationships and behavior patterns of users, and are of great value for platforms to provide personalized recommendations, advertising and other services. However, the widespread sharing of social network data poses a serious challenge to user privacy protection. Malicious attackers can use community detection technology to infer users' private information. For example, by analyzing the community structure in social networks, they can infer the potential relationships between users, conduct recommendation interventions, and even carry out precision fraud, which seriously threatens Internet security. As a result, privacy protection and data security issues have received more and more attention, and community hiding technology has emerged.

[0003] Community hiding technology aims to confuse the community detection algorithm by changing the network structure, achieve the hiding of the target community structure, and protect the user's sensitive information from being leaked. Many existing studies model the adversarial network construction problem as an optimization problem of finding the best perturbation. Its goal is to disrupt the recognition results of the community detection algorithm without significantly changing the network structure, thereby achieving the hiding of community information. This problem has been proven to be an NP-hard problem, that is, it cannot be solved in polynomial time. The current mainstream method is based on evolutionary computing methods, such as genetic algorithms that search for optimal solutions by simulating natural selection and genetic mechanisms. It is particularly suitable for dealing with such problems due to its global convergence characteristics. For example, a global community hiding method based on an elite population co-evolution genetic algorithm disclosed in Publication No. CN117176377A. This method uses the method of adding and deleting edges to change the original network structure, introduces elite population strategies and co-evolution mechanisms to improve traditional genetic algorithms, and performs well in hiding performance. However, when choosing to add or delete edges, this method still views the edge relationship between nodes from a discrete perspective, which makes it difficult to capture the potential correlation information between nodes. This results in low algorithm search efficiency and slow convergence when the network scale grows, and the solution set easily falls into the local optimum. Summary of the invention

[0004] The present invention aims to overcome the above-mentioned shortcomings of the prior art and provide a method and device for hiding community information in a social network guided by structural information.

[0005] Based on the genetic algorithm, the present invention abstracts the reconnection operation of the community hiding problem in the social network into an independent individual in the population, and transforms the problem of solving the optimal perturbation into the problem of searching for individuals with the highest fitness in the population. At the same time, by strengthening the attention to local structural information, a new individual genotype based on feature embedding is designed to achieve the fusion of topological information and algorithm. In addition, the present invention introduces a fine-grained position mutation operator guided by feature similarity. When the population searches for a better individual, the new individual generated based on the constraint of feature similarity retains the similarity with the original individual to a large extent. This feature enhances the ability of individuals to introduce external dominant genes into the population through mutation operations, and provides effective guidance for the algorithm optimization process. Furthermore, the present invention proposes a multi-granularity mutation strategy that couples coarse and fine-grained mutations. The coarse-grained mutation operator has a large mutation space and can provide effective information for the population in the early stage of the algorithm. The fine-grained position mutation operator considers the potential correlation information between nodes. The combination of the two not only prevents the algorithm from falling into the local optimum too early, but also enhances the local search ability of the algorithm, so that it can fully explore the solution space and avoid missing potential dominant solutions. This strategy effectively balances the algorithm's global exploration and local search capabilities, accelerates algorithm convergence, and improves algorithm accuracy.

[0006] The first aspect of the present invention relates to a structure information-oriented social network community information hiding method, comprising the following steps:

[0007] S1: Collect social network data and obtain user node topology information.

[0008] S2: Build the node-embedding dictionary NR.

[0009] S3: Use a genetic algorithm based on feature similarity-guided mutation operators to find the optimal network reconnection strategy.

[0010] S4: Generative Adversarial Networks Achieve community concealment.

[0011] Specifically, in step S1, given a social network, the network is loaded in Represents the social network The set of all user nodes in , Indicates the total number of nodes, represents the set of connections between any two users, e ij Indicates the connection between user i and user j. If there is a connection, the value is 1, otherwise it is 0.

[0012] In the step S2, where h i represents the i-th node v iThe embedding vector of the node is obtained through the network embedding algorithm, and the values ​​of each dimension in the embedding vector of all nodes are counted to construct the distribution model F(h). For node v i , whose embedding vector can be expressed as h i ={h i [1], h i [2], ..., h i [m]}, where m is the dimension of the embedding vector, h i [m] represents the value of the embedding vector in the mth feature dimension. When a specific dimension of an embedding vector needs to be mutated later, a new random value can be generated through F(h) to ensure that the mutated embedding vector conforms to a reasonable distribution and the rationality of the newly generated embedding vector.

[0013] The step S3 specifically comprises the following steps:

[0014] S3.1: Population initialization. The population obtained by initialization is pop 0 is a collection of individual phenotypes, i.e., pop 0 ={ind_phe1, ind_phe2,..., ind_phe g}. Among them, g represents the population size, ind_phe i Indicates the phenotype of the i-th individual in the population. i ={ind_phe i [1], ind_phe i [2], ..., ind_phe i [T]}, where T represents the number of edge reconnection operations. Indicates individual ind_phe i The lth gene of , a gene represents a set of reconnection operations. i，l is the target node, represents the non-neighboring nodes of the target node, Represents the neighbor nodes of the target node. That is, in v i，l and Add an edge between them and delete v i，l and The edges between.

[0015] S3.2: Calculate individual fitness. Individual ind_phe i The fitness value of can be calculated by the modularity Q of the adversarial network as follows:

[0016] f(ind_phe i )=e -Q (1)

[0017] S3.3: Selection operation. Use the roulette wheel selection operator to select individuals from the population for subsequent crossover and mutation operations, so that individuals with larger fitness values ​​have a greater probability of being selected. i The probabilities of being selected are as follows:

[0018]

[0019] Where f(·) represents the individual fitness function.

[0020] S3.4: Crossover operation. First, a fixed-point crossover operator is used to generate a crossover site for the individual that needs to perform the crossover operation, and then the crossover probability p is used to generate a crossover site. c Exchange the gene fragments of two individuals before and after the crossover site.

[0021] S3.5: Multi-granularity mutation operation. Traverse the individuals in the population generated by the crossover operation and determine whether to perform the mutation operation based on the set mutation rate pm. For individuals that need to be mutated, the mutation rate pm is set to 0. bit Determine whether to perform coarse-grained mutation or fine-grained mutation guided by feature similarity. If it is a coarse-grained mutation, traverse the genes at each gene position of the individual phenotype and randomly select one of the three operations: edge addition mutation, edge deletion mutation, and edge reconnection mutation. Specifically, the lth gene of the phenotype of the i-th individual in the population can be expressed as When edge-increasing variation occurs, Will randomly change to the same as v i，l Any one of the non-adjacent nodes; edge deletion mutation makes Randomly change to v i,l Any of the adjacent nodes; reconnection mutation will regenerate the entire gene, and the newly generated target node is not v i，l . If it is a fine-grained bit mutation guided by feature similarity, the individual phenotype is first encoded based on the mapping operation to obtain the genotype. The embedding vector of the gene corresponding to the individual phenotype gene can be obtained by querying the node-embedding dictionary NR. Traverse each gene of the individual genotype, randomly select one of the three operations of adding edge bit mutation, deleting edge bit mutation and reconnecting edge bit mutation, and then calculate the variable range of each dimension of the gene according to different mutation types. Specifically, the lth gene of the genotype of the i-th individual in the population can be expressed as where h i,l , Respectively represent v i,l , The corresponding embedding vector is Respectively represent the embedded vectors obtained after mutation. If edge-adding mutation occurs, Each dimension in is quantized according to the dimensional mutation rate pm eChanges occur with equal probability and need to be satisfied With h i,l The feature similarity of is as high as possible. Among them, the Distance(·,·) function can be used to measure the feature similarity between two embedding vectors. The smaller the value, the more similar the two are. The specific calculation formula is as follows.

[0022]

[0023] Specifically, for the tth dimension Its variable range must satisfy the requirement of Increase. If edge deletion occurs, Each dimension in is quantized according to the dimensional mutation rate pm e changes with equal probability, where for the tth dimension Its variable range must satisfy the requirement of If the reconnection edge position variation occurs, h i,l Each dimension in is quantized according to the dimensional mutation rate pm e Changes occur with equal probability. Specifically, for the tth dimension h i,l [t], its variable interval must satisfy the requirement of Increase and Reduce. In each mutation interval, a new value of the mutation bit is randomly generated according to F(h) to obtain a new embedding vector, and then the newly generated embedding vector is decoded based on similarity, that is, the node that meets the edge condition and has the highest feature similarity is found as the new node in the individual phenotype after the mutation operation, and finally the corresponding new gene is obtained.

[0024] S3.6: Elite retention. First, traverse the individuals in the population generated by the mutation operation, calculate the fitness according to their phenotype, then sort the individual fitness from high to low, and finally select the individuals with the highest fitness in the parent generation according to the elite retention rate elite_rate, and replace the individuals with the worst fitness in the new population after the mutation operation to accelerate the convergence of the algorithm.

[0025] S3.7: Termination condition. Determine whether the number of iterations reaches the preset value N. If not, the newly generated child population becomes the parent population, and steps S3.2 to S3.6 are repeated. Otherwise, the algorithm terminates.

[0026] In step S4, the best individual is obtained from the last generation population obtained in step S3, and its reconnection operation is deployed to the network. Adversarial Network This is to confuse the community detection algorithm, making its division invalid and thus achieving community hiding.

[0027] The second aspect of the present invention relates to a structure information-guided social network community information hiding device, comprising a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the structure information-guided social network community information hiding method of the present invention.

[0028] A third aspect of the present invention relates to a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the structure information-guided social network community information hiding method of the present invention.

[0029] The present invention provides a method for hiding community information in social networks guided by structural information. Specifically, in view of the fact that existing methods lack effective guidance strategies in the process of searching for optimal perturbation substructures and are limited to exploring solution spaces using discrete perspectives, the present invention designs a new type of individual genotype based on local structural information of the network, introduces a feature similarity-guided bit mutation operator, and further proposes a multi-granularity mutation strategy that couples coarse and fine granularity mutations. Then, an improved genetic algorithm is used to search for the optimal reconnection strategy, which is deployed on the original network to construct an adversarial network, and finally achieves community hiding on social networks.

[0030] The present invention combines network topology information with genetic algorithms, designs a new type of individual genotype based on feature embedding, and introduces a fine-grained position mutation operator guided by feature similarity. When the population searches for a better individual, the new individual generated based on the constraint of feature similarity retains the similarity with the original individual to a large extent. This feature enhances the ability of individuals to introduce external superior genes into the population through mutation operations, providing effective guidance for the algorithm optimization process. In addition, the present invention proposes a multi-granularity mutation strategy that couples coarse and fine granularity mutations, which effectively improves the search efficiency and accuracy of the algorithm. Finally, the optimal reconnection strategy is obtained through the improved genetic algorithm search, and it is deployed on the original network to construct an adversarial network, realizing community hiding on social networks.

[0031] The advantages of the present invention are:

[0032] (1) The present invention proposes a structural information-guided social network community information hiding method. By strengthening the focus on local structural information, a new individual genotype based on feature embedding is designed to guide the construction of an adversarial network. This overcomes the limitation of the traditional genetic algorithm-based community information hiding method that lacks effective guidance in optimization, and effectively improves the effect of community hiding.

[0033] (2) When solving the problem using a genetic algorithm, the present invention designs a fine-grained position mutation operator guided by feature similarity and proposes a multi-grained mutation strategy that couples coarse-grained and fine-grained mutations, which effectively balances the algorithm's global exploration and local search capabilities, and breaks through the dilemma of existing community information hiding methods that are easily trapped in local optimality due to being limited to coarse-grained mutations. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] 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.

[0035] Figure 1 It is a framework diagram of the present invention.

[0036] Figure 2 A schematic diagram of the present invention for finding the optimal network reconnection strategy using a genetic algorithm based on feature similarity-guided bit mutation operators.

[0037] Figure 3 Schematic diagram of the multi-granularity mutation operation designed for individuals in the present invention.

[0038] Figure 4 Schematic diagram of the fine-grained variation designed for the present invention.

[0039] Figure 5 It is a schematic diagram of the device structure of the present invention. DETAILED DESCRIPTION

[0040] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as limiting the present invention, but should be understood as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0041] It should be understood that the terms described in the present invention are only for describing special embodiments and are not intended to limit the present invention. In addition, for the numerical range in the present invention, it should be understood that each intermediate value between the upper and lower limits of the scope is also specifically disclosed. Each smaller range between the intermediate value in any stated value or stated range and any other stated value or intermediate value in the described range is also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded in the scope.

[0042] Unless otherwise indicated, all technical and scientific terms used herein have the same meanings as those generally understood by those skilled in the art. Although the present invention describes only preferred methods and materials, any methods and materials similar or equivalent to those described herein may also be used in the implementation or testing of the present invention. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods and / or materials associated with the documents. In the event of a conflict with any incorporated document, the content of this specification shall prevail.

[0043] It will be apparent to those skilled in the art that various modifications and variations may be made to the specific embodiments of the present invention description without departing from the scope or spirit of the present invention. Other embodiments derived from the present invention description will be apparent to those skilled in the art. The present application description and examples are exemplary only.

[0044] The words “include,” “including,” “have,” “contain,” etc. used in this document are open-ended terms, meaning including but not limited to.

[0045] Unless otherwise specified, the "parts" described in the present invention are all based on mass parts.

[0046] Example 1

[0047] This embodiment provides a method for hiding community information on social networks guided by structural information. First, a new type of individual genotype is designed based on the local structural information of the network. Then, a feature similarity-guided bit mutation operator is introduced, and a multi-granularity mutation strategy coupled with coarse-grained and fine-grained mutations is proposed. The optimal reconnection strategy is obtained by combining the improved genetic algorithm search, and it is deployed on the original network to construct an adversarial network, and finally the community hiding on the social network is realized. Figure 1 As shown, here we take the social network based on the karate club and the Louvain community detection algorithm as an example, including the following implementation steps:

[0048] Collect network data and build node-embedding dictionary:

[0049] Step S1: Collect social network data and obtain user node topology information.

[0050] Get the Karate Club social network, load the network in Represents the network The set of all user nodes in , Indicates the total number of nodes, represents the set of connections between any two users, e ij Indicates the connection between user i and user j. If there is a connection, the value is 1, otherwise it is 0.

[0051] Step S2: Construct a node-embedding dictionary NR.

[0052] where h i represents the i-th node v i The embedding vector of the node is obtained through the DeepWalk algorithm, and the values ​​of each dimension in the embedding vector of all nodes are counted to construct the distribution model F(h). For node v i , whose embedding vector can be expressed as h i ={h i [1], h i [2], ..., h i [m]}, where m is the dimension of the embedding vector, h i [m] represents the value of the embedding vector in the mth feature dimension. When a specific dimension of an embedding vector needs to be mutated later, a new random value can be generated through F(h) to ensure that the mutated embedding vector conforms to a reasonable distribution and the rationality of the newly generated embedding vector.

[0053] Finding the optimal network reconnection strategy:

[0054] Step S3: Use a genetic algorithm based on feature similarity-guided mutation operators to find the optimal network reconnection strategy.

[0055] S3.1: Population initialization. The population obtained by initialization is pop 0 is a collection of individual phenotypes, i.e., pop 0 ={ind_phe1, ind_phe2,..., ind_phe g}. Among them, g represents the population size, ind_phe i Indicates the phenotype of the i-th individual in the population. i ={ind_phe i [1], ind_phe i [2], ..., ind_phe i [T]}, where T represents the number of edge reconnection operations. Indicates individual ind_phe i The lth gene of , a gene represents a set of reconnection operations. i，l is the target node, represents the non-neighboring nodes of the target node, Represents the neighbor nodes of the target node. That is, in v i，l and Add an edge between them and delete v i，l and Set the population size g = 100 and the number of reconnection operations T = 6.

[0056] S3.2: Calculate individual fitness. Individual ind_phe i The fitness value calculation formula is as follows:

[0057]

[0058] in, and Respectively represent individual ind_phe i Edges to be added and edges to be deleted. Represents the adversarial network generated by this set of adding and deleting edge operations. This is the community result obtained by dividing the adversarial network using the community detection algorithm Louvain based on modularity optimization. The value of node v m The affiliation with community n. This element is 1 if the node belongs to the community, otherwise it is 0. To counter the modularity of the network, the smaller the modularity, the lower the quality of community division and the higher the corresponding individual fitness. Modularity does not rely on prior community label information and can intuitively measure the quality of community division.

[0059] S3.3: Selection operation. Use the roulette wheel selection operator to select individuals from the population for subsequent crossover and mutation operations, so that individuals with larger fitness values ​​have a greater probability of being selected. i The probabilities of being selected are as follows:

[0060]

[0061] Where f(·) represents the individual fitness function.

[0062] S3.4: Crossover operation. First, a fixed-point crossover operator is used to generate a crossover site for the individual that needs to perform the crossover operation, and then the crossover rate p is used to generate a crossover site. c =0.8 Exchange the gene fragments of the two individuals before and after the crossover site.

[0063] S3.5: Multi-granularity mutation operation. Traverse the individuals in the population generated by the crossover operation, refer to Figure 3 , for each individual, determine whether to perform the mutation operation according to the mutation rate pm = 0.1. For individuals that need to be mutated, according to the bit mutation rate pm bit= 0.2 to determine whether to perform coarse-grained mutation or feature similarity-oriented position mutation. If it is a coarse-grained mutation, the genes at each gene position of the individual phenotype are traversed, and one of the three operations of adding edge mutation, deleting edge mutation, and reconnecting edge mutation is randomly selected. Specifically, the lth gene of the phenotype of the i-th individual in the population can be expressed as When edge-increasing variation occurs, Will randomly change to the same as v i，l Any one of the non-adjacent nodes; edge deletion mutation makes Randomly change to v i，l Any of the adjacent nodes; reconnection mutation will regenerate the entire gene, and the newly generated target node is not v i，l . If it is a fine-grained bit mutation guided by feature similarity, the individual phenotype is first encoded based on the mapping operation to obtain the genotype. The embedding vector of the gene corresponding to the individual phenotype gene can be obtained by querying the node-embedding dictionary NR. Traverse each gene of the individual genotype, randomly select one of the three operations of adding edge bit mutation, deleting edge bit mutation and reconnecting edge bit mutation, and then calculate the variable range of each dimension of the gene according to different mutation types. Specifically, the lth gene of the genotype of the i-th individual in the population can be expressed as where h i,l , Respectively represent v i，l , The corresponding embedding vector is Respectively represent the embedded vectors obtained after mutation. If edge-adding mutation occurs, Each dimension in is quantized according to the dimensional mutation rate pm e Changes occur with equal probability and need to be satisfied With h i,l The feature similarity of is as high as possible. Among them, the Distance(·,·) function can be used to measure the feature similarity between two embedding vectors. The smaller the value, the more similar the two are. The specific calculation formula is as follows.

[0064]

[0065] Specifically, for the tth dimension Its variable range must satisfy the requirement of Increase. If edge deletion occurs, Each dimension in is quantized according to the dimensional mutation rate pm e changes with equal probability, where for the tth dimension Its variable range must satisfy the requirement of If the reconnection edge position variation occurs, h i,l Each dimension in is quantized according to the dimensional mutation rate pm eChanges occur with equal probability. Specifically, for the tth dimension h i,l [t], its variable interval must satisfy the requirement of Increase and Reduce. In each mutation interval, a new value of the mutation bit is randomly generated according to F(h) to obtain a new embedding vector, and then the newly generated embedding vector is decoded based on similarity, that is, the node that meets the edge condition and has the highest feature similarity is found as the new node in the individual phenotype after the mutation operation, and finally the corresponding new gene is obtained.

[0066] S3.6: Elite retention. First, traverse the individuals in the population generated by the mutation operation, calculate the fitness according to their phenotype, then sort the individual fitness from high to low, and finally replace the worst 10% of individuals in the new population after the mutation operation with the best 10% of individuals in the parent generation to accelerate the convergence of the algorithm.

[0067] S3.7: Termination condition. Determine whether the number of iterations reaches 1500. If not, the newly generated child population becomes the parent population and steps S3.2 to S3.6 are repeated. Otherwise, the algorithm terminates.

[0068] Adversarial network generation to achieve anonymity:

[0069] Step S4: Generate adversarial network Achieve community concealment.

[0070] Get the best individual from the last generation population obtained in step S3 and deploy its reconnection operation to the network Adversarial Network The original network and adversarial networks As the input of Louvain algorithm, community division is performed. The community cluster P obtained by community division shows a clear community structure. However, the adversarial network Community clusters obtained by community division The original clarity is lost and the community structure is blurred. This change shows that adversarial networks This effectively interferes with the Louvain algorithm, causing the community division results output by the algorithm to become invalid, thus achieving community hiding. This result also verifies the effectiveness of the structural information-guided social network community information hiding method proposed in this invention.

[0071] The algorithm proposed in the present invention is compared with other models, and the community information hiding algorithm used for comparison is the latest baseline algorithm Q-Attack. The present invention uses modularity (Q) and normalized mutual information (NMI) as evaluation functions. The lower the Q value, the more blurred the community structure, and the lower the NMI value, the greater the difference between the community detection result and the real division. Under the same configuration, the modularity value of the Q-Attack algorithm is reduced by 17.29% relative to the original network value, and the normalized mutual information is 0.7856. The modularity value of the algorithm proposed in the present invention is reduced by 22.34% relative to the original network value, and the normalized mutual information is 0.7620, which indicates that the community structure of the adversarial network obtained by the algorithm is fuzzy, the degree of information hiding is high, and the performance is excellent.

[0072] In summary, the present invention utilizes a structural information-oriented social network community information hiding method, designs a new individual genotype based on feature embedding by strengthening the focus on local structural information, and realizes the integration of topological information and algorithm; furthermore, the present invention proposes a feature similarity-oriented fine-grained bit mutation operator, which can effectively utilize additional information, guide the algorithm search direction, improve the algorithm search ability, and accelerate the algorithm convergence. The present invention can be applied to the social network community information hiding task in real scenarios to enhance the privacy protection of community members.

[0073] Example 2

[0074] like Figure 5 This embodiment relates to a structure information-guided social network community information hiding device, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the structure information-guided social network community information hiding method of embodiment 1.

[0075] Example 3

[0076] This embodiment relates to a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the structure information-oriented social network community information hiding method of the present invention is implemented.

[0077] The embodiments described above are only descriptions of the preferred modes 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 structural information-oriented social network community information hiding method, characterized in that: The following steps are involved: S1: Collect social network data and obtain user node topology information; S2: build node-embedding dictionary NR; S3: Use a genetic algorithm based on feature similarity-guided mutation operators to find the optimal network reconnection strategy; S4: Generative Adversarial Networks Achieve community concealment.

2. According to claim 1, a structure information-oriented social network community information hiding method is characterized by: The step S1 requires loading the network in Represents the social network The set of all user nodes in , Indicates the total number of nodes, represents the set of connections between any two users, e ij Indicates the connection between user i and user j. If there is a connection, the value is 1, otherwise it is 0.

3. The structural information-oriented social network community information hiding method according to claim 1 is characterized by: In the step S2, where h i represents the i-th node v i The embedding vector of the node is obtained through the network embedding algorithm, and the values ​​of each dimension in the embedding vector of all nodes are counted to construct the distribution model F(h); for node v i , whose embedding vector is represented as h i ={h i [1],h i [2],...,h i [m]}, where m is the dimension of the embedding vector, h i [m] represents the value of the embedding vector in the mth feature dimension. When a specific dimension of an embedding vector needs to be mutated later, a new random value is generated through F(h) to ensure that the mutated embedding vector conforms to a reasonable distribution and the rationality of the newly generated embedding vector.

4. The structural information-oriented social network community information hiding method according to claim 1, characterized in that: The step S3 uses a genetic algorithm based on feature similarity-guided position mutation operators to find the optimal network reconnection strategy, which specifically includes the following steps: S3.1: Population initialization; the population obtained by initialization is pop 0 is a collection of individual phenotypes, i.e., pop 0 ={ind_phe1,ind_phe2,…,ind_phe g }; where g represents the population size, ind_phe i Indicates the phenotype of the i-th individual in the population; ind_phe i ={ind_phe i [1],ind_phe i [2],…,ind_phe i [T]}, T represents the number of edge reconnection operations; Indicates individual ind_phe i The lth gene of , a gene represents a set of reconnection operations; where v i,l is the target node, represents the non-neighboring nodes of the target node, Represents the neighbor nodes of the target node; That is, in v i,l and Add an edge between them and delete v i,l and The connecting edges between S3.2: Calculate individual fitness; individual ind_phe i The fitness value of is calculated by the modularity Q of the adversarial network as follows: f(ind_phe i )=and -Q (1) S3.3: Selection operation; use the roulette wheel selection operator to select individuals from the population for subsequent crossover and mutation operations, so that individuals with larger fitness values ​​have a greater probability of being selected; individual ind_phe i The probabilities of being selected are as follows: Where f(·) represents the individual fitness function; S3.4: Crossover operation: First, a fixed-point crossover operator is used to generate a crossover site for the individual that needs to perform the crossover operation, and then the crossover probability p is calculated. c Exchange the gene fragments of the two individuals before and after the crossover site; S3.5: Multi-granularity mutation operation; traverse the individuals in the population generated by the crossover operation, and determine whether to perform the mutation operation based on the set mutation rate pm; for individuals that need to be mutated, according to the bit mutation rate pm bit Determine whether to perform coarse-grained mutation or fine-grained mutation guided by feature similarity; if it is coarse-grained mutation, traverse the genes at each gene position of the individual phenotype and randomly select one of the three operations of adding edge mutation, deleting edge mutation and reconnecting edge mutation; specifically, the lth gene of the phenotype of the i-th individual in the population is represented by When edge-increasing variation occurs, Will randomly change to the same as v i,l Any one of the non-adjacent nodes; edge deletion mutation makes Randomly change to v i,l Any of the adjacent nodes; reconnection mutation will cause the entire gene to be regenerated, and the newly generated target node is not v i,l ; If it is a fine-grained bit mutation guided by feature similarity, the individual phenotype is first encoded based on the mapping operation to obtain the genotype, and the embedding vector of its gene corresponding to the individual phenotype gene is obtained by querying the node-embedding dictionary NR; traverse each gene of the individual genotype, randomly select one of the three operations of adding edge bit mutation, deleting edge bit mutation and reconnecting edge bit mutation, and then calculate the variable range of each dimension of the gene according to different mutation types; Specifically, the lth gene of the genotype of the i-th individual in the population is expressed as where h i,l , Respectively represent v i,l , The corresponding embedding vector is Respectively represent the embedded vectors obtained after mutation; if edge mutation occurs, Each dimension in is quantized according to the dimensional mutation rate pm e Changes occur with equal probability and need to be satisfied With h i,l The feature similarity of is as high as possible; the Distance(·,·) function is used to measure the feature similarity between two embedded vectors. The smaller the value, the more similar the two are. The calculation formula is as follows; Specifically, for the tth dimension Its variable range must satisfy the requirement of Increase; if edge deletion occurs, Each dimension in the dimension is quantized according to the dimensional mutation rate pm e changes with equal probability, where for the tth dimension Its variable range must satisfy the requirement of decreases; if reconnection edge variation occurs, h i,l Each dimension in the dimension is quantized according to the dimensional mutation rate pm e Changes occur with equal probability. Specifically, for the tth dimension h i,l [t], its variable interval must satisfy the requirement of Increase and Reduce; randomly generate new values ​​of the mutation bit according to F(h) in each mutation interval to obtain a new embedding vector, and then perform similarity-based decoding on the newly generated embedding vector, that is, find the node that meets the edge condition and has the highest feature similarity as the new node in the individual phenotype after the mutation operation, and finally obtain the corresponding new gene; S3.6: Elite retention; first traverse the individuals in the population generated by the mutation operation, calculate the fitness according to their phenotype, then sort the individual fitness from high to low, and finally select the individuals with the highest fitness in the parent generation according to the elite retention rate elite_rate, and replace the individuals with the worst fitness in the new population after the mutation operation to accelerate the convergence of the algorithm; S3.7: Termination condition; determine whether the number of iterations reaches the preset value N. If not, the newly generated child population becomes the parent population, and steps S3.2 to S3.6 are repeated. Otherwise, the algorithm terminates.

5. The structural information-oriented social network community information hiding method according to claim 1, characterized in that: In step S4, the best individual is obtained from the last generation population obtained in step S3, and its reconnection operation is deployed to the network. Adversarial Network by Replace the original network Confuse the community detection algorithm, making its division invalid, thereby achieving community hiding.

6. A structure information-oriented social network community information hiding device, characterized in that: It comprises a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement the structural information-guided social network community information hiding method described in any one of claims 1-5.

7. A computer-readable storage medium, characterized in that: A program is stored thereon, and when the program is executed by a processor, the structural information-guided social network community information hiding method described in any one of claims 1-5 is implemented.

Images