A method and medium for maximizing the influence of discrete Jaya based on adaptive terrain perception.

Abstract

This invention provides a discrete Jaya influence maximization method and medium based on fitness terrain awareness, relating to the field of virtual network social networking. It designs a discrete Jaya algorithm update mechanism, enabling individuals to perform directional searches based on optimal and worst candidate sets without requiring additional parameters. A dominant terrain vector is introduced to dynamically evaluate the population state and identify local stagnation. A parallel elite local search strategy is integrated to enhance global exploration capabilities. Experiments show that this scheme performs superiorly in various real-world social networks, with influence propagation approaching that of traditional greedy algorithms, significantly improving operational efficiency, and exhibiting good robustness and scalability. It is applicable to networks with different structures, providing an efficient and stable solution for the influence maximization problem.

Description

Technical Field

[0001] This invention relates to the field of virtual network social networking, and in particular to a method and medium for maximizing discrete Jaya influence based on adaptive terrain perception. Background Technology

[0002] In recent years, various metaheuristic optimization algorithms have been proposed for solving influence maximization problems, and researchers have improved optimization performance by designing specialized search mechanisms. Representative methods include discrete particle swarm optimization, local influence descent differential evolution, discrete Harris-Eagle optimization, staged evolutionary algorithms, and co-evolutionary metaheuristic algorithms. Compared to traditional greedy or heuristic strategies, these methods exhibit good solution performance through customized fitness functions and evolutionary strategies. However, existing metaheuristic methods fail to fully exploit the intrinsic characteristics of network structures to guide the evolutionary direction, especially in large-scale network environments, where they are prone to getting trapped in local optima and exhibiting premature convergence. Furthermore, the performance and stability of these methods are usually highly dependent on the algorithm parameter settings, and parameter sensitivity limits their robustness in practical applications.

[0003] Therefore, to address the shortcomings of existing technologies in maximizing large-scale influence problems, such as premature convergence and high parameter sensitivity, a discrete Jaya influence maximization method and medium based on adaptive terrain perception is proposed to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a method and medium for maximizing the influence of discrete Jaya based on fitness terrain perception. It draws on the process in nature where individuals approach the optimal solution through cooperation and competition. By incorporating fitness terrain information, it guides the population to evolve in a more promising direction, thereby effectively improving the quality of the solution and the robustness of the algorithm.

[0005] To achieve the above objectives, this invention provides a method for maximizing discrete Jaya influence based on fitness terrain perception, comprising the following steps: S1: Collect social network data and model a social network graph through user relationships; S2: Design a discrete Jaya update mechanism, construct the best candidate set and the worst candidate set based on the current individual, the best individual and the worst individual respectively, and perform discrete update operations according to the state of the candidate sets; S3: Construct the dominant terrain vector to calculate the fitness terrain index. When the fitness terrain index is 0, trigger the parallel elite local search mechanism. S4: Integrates the fitness terrain index with the discrete Jaya update mechanism to dynamically regulate the evolution process and finally output the optimal seed node set.

[0006] Preferably, in S1, the social network graph is as follows: ,in, Represents a set of nodes. Represents the set of edges.

[0007] Preferably, the optimal candidate set in S2 includes nodes that exist in the optimal individual but not in the current individual, defined as: The optimal candidate set provides a clear direction for improvement in the "addition" operation, guiding the current individual to evolve toward the optimal solution. The worst candidate set includes nodes that exist in both the current individual and the worst individual, and is defined as: The worst candidate set is used to identify low-quality nodes that reduce the quality of the solution, providing clear guidance for the "removal" operation and avoiding inferior nodes.

[0008] Preferably, the discrete Jaya update mechanism in S2 specifically includes: New individuals are generated based on whether the best and worst candidate sets are empty; the specific update rules are as follows: ; in, express A node is randomly selected from the data. Indicates from A node is randomly selected from the data. Indicates from the current individual A node is randomly selected from the data. Indicates from A node randomly selected from the data; V Indicates the number of nodes. This represents the i-th individual in the population.

[0009] When both the best and worst candidate sets are empty, a local influence value is introduced. LFV To quantify the influence of each node, local influence value LFV The specific definitions are as follows: ; In the formula, and Representing nodes respectively and A set of neighbors that jumps over time. and They represent the nodes respectively and and from the node arrive Activation probability; Deleted node Defined as the current individual The smallest LFV The value of the node is calculated using the following formula: ; Nodes to be added From the candidate set The selection process is as follows: ①Calculation Each node in LFV Values, and stored in a vector middle: ; ② Normalization is performed to obtain the probability vector. : ; ③ Based on candidate node set With probability vector Select nodes to be added through categorical sampling. : ; Better individuals are selected and retained for the next generation of the population through fitness comparison.

[0010] Preferably, in S3, the dominant terrain vector is constructed based on the difference in fitness values ​​between two consecutive generations of individuals, and is used to quantify the population evolution trend; the fitness terrain index is the sum of the elements of the dominant terrain vector. The parallel elite local search mechanism performs one-hop neighbor replacement operations in parallel on each node in the current seed set and updates the seed set by selecting the best one.

[0011] Preferably, the fitness vector of the parent generation is set as follows: The fitness vector of the offspring is Then the dominant terrain vector The i Each element is represented as: ; In the process of population evolution, the dominant terrain vector is calculated based on the fitness values ​​of two consecutive generations. The fitness terrain index R is defined as the sum of all elements of the dominant terrain vector, which is used to quantify the overall evolutionary trend of the population. .

[0012] Preferably, if the fitness terrain index is not zero, the discrete Jaya update step in S4 is executed until the termination condition is met; specifically, the following steps are covered: S41: Generate a candidate node set based on two-hop coverage, and randomly select nodes from it to construct the initial population; S42: For a given current individual Based on the current best and worst individuals, generate the best candidate set and the worst candidate set; apply the discrete update rule to... To generate new individuals ',exist 'and Individuals with higher fitness are selected to enter the next generation; S43: Construct a dominant terrain vector based on the fitness value of each individual in the population over two consecutive generations; calculate the fitness terrain index R based on the dominant terrain vector to assess whether the population is trapped in a local optimum. like This indicates that population evolution has stalled, triggering a parallel elite local search mechanism; like Return to the discrete Jaya update mechanism in S2 and iterate until the optimal solution is obtained; S44: When the maximum number of iterations is reached, the loop terminates, and the individual with the highest fitness is output from the final population as the optimal seed set.

[0013] A computer-readable medium storing a computer program that, when executed, implements steps of a method for maximizing discrete Jaya influence based on fitness-based terrain perception.

[0014] Therefore, the present invention employs the aforementioned method and medium for maximizing discrete Jaya influence based on adaptive terrain perception, and the technical effects are as follows: (1) This invention discretizes and reconstructs the update mechanism of the Jaya algorithm and proposes a three-stage update strategy, which effectively enhances the Jaya algorithm's directional search capability in the discrete solution space.

[0015] (2) This invention proposes for the first time to use the dominant terrain vector to quantify the ruggedness of the solution space during the evolution process, and applies it to the problem of maximizing influence to determine whether the population is trapped in a local optimum.

[0016] (3) This invention proposes a discrete Jaya optimization method based on adaptive terrain perception, which can effectively identify the most influential seed node set, thereby maximizing the spread of social network influence.

[0017] (4) The present invention has good generalization performance and can be applied to the problem of maximizing influence in social networks with various topologies. Attached Figure Description

[0018] Figure 1 This is a flowchart of the discrete Jaya influence maximization method based on adaptive terrain perception according to the present invention; Figure 2 This is a social network graph in an embodiment of the present invention; Figure 3 The above are topological diagrams of the datasets in this embodiment of the invention; where (a), (b), (c), (d), (e), and (f) are complete topological diagrams of six real social network datasets; Figure 4 This is a comparison diagram of the influence propagation of different algorithms in the comparative example of this invention; Figure 5 This is a comparison chart showing the running time of different algorithms in the comparative example of this invention. Detailed Implementation

[0019] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0020] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0021] Example 1 This invention provides a method for maximizing discrete Jaya influence based on fitness terrain perception, comprising the following steps: S1: Collect social network data and model a social network graph using user relationships; such as... Figure 2 As shown, in S1, the social network graph is... ,in, Represents a set of nodes. ; Denotes the set of edges. .

[0022] S2: Design a discrete Jaya update mechanism, based on the current individual Optimal Individual With the worst individual Construct the best candidate set and the worst candidate set, and perform discrete update operations based on the state of the candidate sets; like Figure 1 As shown, the update rule of the Jaya algorithm is redesigned for the discrete feature space of the influence maximization problem. The proposed discrete update mechanism mainly includes three steps: (1) constructing a guiding candidate set; (2) generating new individuals based on the candidate set; and (3) selecting individuals with higher fitness values ​​as offspring.

[0023] The optimal candidate set in S2 includes nodes that exist in the optimal individual but not in the current individual, defined as: The optimal candidate set provides a clear direction for improvement in the "addition" operation, guiding the current individual to evolve toward the optimal solution. The worst candidate set includes nodes that exist in both the current individual and the worst individual, and is defined as: The worst candidate set is used to identify low-quality nodes that reduce the quality of the solution, providing clear guidance for the "removal" operation and avoiding inferior nodes.

[0024] The discrete Jaya update mechanism in S2 specifically includes: New individuals are generated based on whether the best and worst candidate sets are empty; the specific update rules are as follows: ; in, express A node is randomly selected from the data. Indicates from A node is randomly selected from the data. Indicates from the current individual A node is randomly selected from the data. Indicates from A node randomly selected from the data; V Indicates the number of nodes. This represents the i-th individual in the population.

[0025] When both the best and worst candidate sets are empty, a local influence value is introduced. LFV To quantify the influence of each node, local influence value LFV The specific definitions are as follows: ; In the formula, and Representing nodes respectively and A set of neighbors that jumps over time. and They represent the nodes respectively and and from the node arrive Activation probability; Deleted node Defined as the current individual The smallest LFV The value of the node is calculated using the following formula: ; Nodes to be added From the candidate set The selection process is as follows: ①Calculation Each node in LFV Values, and stored in a vector middle: ; ② Normalization is performed to obtain the probability vector. : ; ③ Based on candidate node set With probability vector Select nodes to be added through categorical sampling. : ; Better individuals are selected and retained for the next generation of the population through fitness comparison.

[0026] During the selection phase, the Expected Diffusion Value (EDV) is used as the fitness function to evaluate the quality of individuals. This EDV is calculated based on the one-hop neighbor coverage index and aims to efficiently approximate the expected influence of the seed set in an independent cascade model. After generating offspring individuals through the update mechanism, the fitness values ​​of the parent and offspring individuals are compared, and individuals with higher fitness are selected to enter the next generation to ensure continuous improvement in the evolutionary process. The specific selection rules are defined as follows: ; In the formula, Represents the set of candidate seeds Number of nodes for A set of neighbors that jumps over time. For activation probability, Represents nodes The number of connected candidate seed nodes.

[0027] S3: Construct the dominant terrain vector to calculate the fitness terrain index. When the fitness terrain index is 0, trigger the parallel elite local search mechanism. In S3, the dominant terrain vector is constructed based on the difference in fitness values ​​between two consecutive generations of individuals and is used to quantify the population's evolutionary trend; the fitness terrain index is the sum of the elements of the dominant terrain vector. The parallel elite local search mechanism performs one-hop neighbor replacement operations in parallel on each node in the current seed set and updates the seed set by selecting the best one.

[0028] Set the parent's fitness vector as The fitness vector of the offspring is Then the dominant terrain vector The i Each element is represented as: ; In the process of population evolution, the dominant terrain vector is calculated based on the fitness values ​​of two consecutive generations. The fitness terrain index R is defined as the sum of all elements of the dominant terrain vector, which is used to quantify the overall evolutionary trend of the population. .

[0029] In the population evolution process of the Discrete Jaya algorithm, its core update mechanism ensures that a parent is replaced only when its fitness is not lower than that of its parent; otherwise, the parent is retained. This mechanism guarantees that the overall quality of the population will not degrade over time, thus maintaining or improving the fitness level throughout the evolutionary process. Therefore, the elements in the dominant terrain vector can only take values ​​of 0 or 1, and negative values ​​are not possible. This indicates that the value range of the fitness terrain index R is... .when When this occurs, it indicates that the population evolution has stagnated. At this point, a parallel local elite search strategy will be triggered to help the population escape the local optimum.

[0030] The detailed steps of the parallel elite local search mechanism are as follows: Input social network graph as and the initial candidate seed set The current optimal seed set Initialize to ; For each seed node u in the current candidate seed set S, perform the following local search optimization process in parallel: 1. Set the initial optimal state: Initialize the current node to be replaced (Node) to u; simultaneously, calculate the current optimal state. Influence dissemination value And set it as the current optimal propagation value. BestEDV ; 2. Traverse neighbor nodes: Examine all neighbor nodes v of the current node u, i.e., v∈Neighbors(u); 3. For each neighbor node v A new candidate seed set S' is formed by selecting from the current seed set. Remove node u And add neighbor nodes. v get; Right now ; 4. Calculate the influence propagation value of the new candidate seed set S'. (S'), if (S') is greater than the best value recorded so far. BestEDV Then update the best replacement node to v, and BestEDV Updated to (S'); 5. After traversing all neighbors of node u, set the current... The original node in u Replace with the best node found in the search; After all parallel local search tasks are completed, the algorithm will compare the initial seed set S with the final optimized set S. The influence propagation values ​​are compared, and the set with the larger EDV value is selected as the final output. Finally, return this optimized seed set. .

[0031] Since these replacement operations are independent of each other, they can be parallelized to accelerate computation.

[0032] S4: Integrates the fitness terrain index with the discrete Jaya update mechanism to dynamically regulate the evolution process and finally output the optimal seed node set.

[0033] If the fitness terrain index is not zero, execute the discrete Jaya update step in S4 until the termination condition is met; specifically, this includes the following steps: S41: Generate a candidate node set based on two-hop coverage, and randomly select nodes from it to construct the initial population; S42: For a given current individual Based on the current best and worst individuals, generate the best candidate set and the worst candidate set; apply the discrete update rule to... To generate new individuals ',exist 'and Individuals with higher fitness are selected to enter the next generation; S43: Construct a dominant terrain vector based on the fitness value of each individual in the population over two consecutive generations; calculate the fitness terrain index R based on the dominant terrain vector to assess whether the population is trapped in a local optimum. like This indicates that population evolution has stalled, triggering a parallel elite local search mechanism; like Return to the discrete Jaya update mechanism in S2 and iterate until the optimal solution is obtained; S44: When the maximum number of iterations is reached, the loop terminates, and the individual with the highest fitness is output from the final population as the optimal seed set.

[0034] A computer-readable medium storing a computer program that, when executed, implements steps of a method for maximizing discrete Jaya influence based on fitness-based terrain perception.

[0035] Comparative Example Figure 2This is a schematic diagram of a social network graph constructed based on real user relationship data in an embodiment of the present invention. The graph uses nodes (circles) to represent individual users on the social platform, and edges (lines) to represent following, friendship, or interaction relationships between users. In the actual data collection process, we crawled publicly available social platform user profiles and interaction logs to extract directed or undirected connections between users, and then modeled it as an undirected, unweighted graph G=(V,E), where | V ∣= n Let be the total number of nodes, and |E| = m be the total number of edges. Figure 2 A typical subnetwork from the experimental dataset was selected for local visualization, clearly demonstrating the network's scale-free and small-world characteristics: a few nodes have high degree (hub nodes), while most nodes are sparsely connected; simultaneously, the average path between nodes is short, and the clustering coefficient is high. This network graph not only serves as the solution space for the influence maximization problem but also provides an intuitive topological basis for subsequent operations such as candidate node selection and local search neighbor replacement in the discrete Jaya algorithm.

[0036] Figure 3 It showcases the different topologies of various networks, visualizing the structures of different networks, some densely distributed and others sparsely distributed. Figure 3 (a) to Figure 3 (f) shows the complete topological structures of the six real-world social network datasets used in this experiment. All subgraphs were drawn using the force-directed layout algorithm, with node positions automatically determined by structural relationships. Color intensity indicates node degree, and line density visually reflects the network connectivity. The six datasets differ significantly in size, density, and clustering patterns, making them highly representative.

[0037] The Blog network has approximately 1,200 nodes but nearly 19,000 edges, exhibiting a high average degree. Visually, it displays a highly dense cluster structure with tightly interconnected nodes and low resistance to information propagation, typical of a dense small-world network. The Deezer network is even larger, with over 50,000 nodes and over 1.5 million edges, boasting an average degree exceeding 50. It represents a super-large-scale dense network, with its visualization showing an extremely concentrated center and gradually sparser edges, suitable for testing the scalability of algorithms in large-scale, high-density environments. In stark contrast are the CA-HepTh and NetHEPT networks, with approximately 9,800 and 15,000 nodes respectively, but only 20,000 to 30,000 edges and an average degree less than 5. Visually, their nodes are widely scattered with sparse connections and significant local community segmentation, representing an extremely sparse structure. In these networks, propagation paths are singular and fragile, highly sensitive to the initial position of seed nodes, effectively testing the adaptability of algorithms in environments where information propagation is difficult. The CA-GrQc and P2P networks fall between these two categories. CA-GrQc has approximately 5,200 nodes and 14,000 edges. Although it is sparse overall, it exhibits high local clustering coefficients, and multiple tightly connected sub-clusters are visible in the visualization, reflecting the clustering characteristics of researchers in collaborative networks. The P2P network has approximately 6,300 nodes and 20,000 edges, but its clustering coefficient is very low. The visualization shows a loosely interwoven mesh with almost no obvious clustering centers, representing a network type with a uniform structure but weak connectivity. By selecting datasets covering various topological characteristics such as dense and sparse, small and very large, and high and low clustering, this experiment comprehensively examines the robustness, search efficiency, and final influence propagation effect of the proposed method in different network environments, providing a rich and reliable experimental foundation for subsequent analysis of the correlation between network structure characteristics and algorithm performance.

[0038] Figure 4 Experimental results in (a) to (f) show that, in six network environments, LA-DJaya achieves the closest solution performance to CELF compared to other typical methods, with its average influence range reaching approximately 97% of CELF. Figure 4 (d) and Figure 4 (e) The performance in a network environment is even optimal, indicating that the algorithm proposed in this invention can obtain a near-optimal solution to the influence maximization problem. Although CELF achieves the maximum propagation range in most cases due to its use of numerous Monte Carlo simulations to accurately estimate influence, the computational cost of this method is extremely high, making it difficult to apply to large-scale networks. Combined with... Figure 5The algorithm's performance efficiency, as shown, further highlights the superior performance of the LA-DJaya algorithm. This excellent performance is attributed to the effective integration of the terrain-aware evolutionary mechanism and the discrete update strategy. Compared to metaheuristic algorithms such as DCGM++, LIDDE, TS-VA-MODE, and PHEE, LA-DJaya demonstrates the best performance, with its advantages being particularly pronounced when the network size or seed set size is large. This is due to the synergistic effect of the terrain dominance vector evaluation mechanism and the parallel elite local search strategy.

[0039] The algorithm exhibits differentiated performance characteristics across different network architectures. Figure 4 (e) In the corresponding network, the performance differences between the algorithms are relatively small. This is because the network topology is relatively uniform and lacks dense local cluster structures, which is also confirmed by its extremely low average clustering coefficient. In this homogeneous structure, the influence of nodes is relatively small, which makes it possible for even simple heuristic methods to achieve good results, thereby weakening the relative advantage of complex optimization algorithms such as LA-DJaya. Figure 4 In the network shown in (b), the algorithm performance exhibits a clear hierarchy. This network has a significant "rich club" characteristic, meaning that highly interconnected nodes are tightly linked. This structure reduces the effectiveness of heuristic algorithms based on node degree centrality (such as DCGM++). In contrast, LA-DJaya's update mechanism can utilize information from the global optimum and worst solution to identify and select influential nodes outside the core of the "rich club," thus achieving better results. LA-DJaya also demonstrates stable superiority on other networks such as CA-HepTh, Deezer, and NetHEHT. These networks have more complex topologies, containing intricate community structures and heterogeneous node connections, resulting in multimodal fitness terrain with numerous local optima. In such network spaces, evolutionary algorithms such as LIDDE and PHEE are prone to premature convergence. LA-DJaya, however, can detect evolutionary stagnation in a timely manner through terrain dominance vectors and trigger parallel elite local searches, thus effectively escaping local optima and continuously exploring more promising regions in the solution space.

[0040] Figure 5Experimental results show that the CELF algorithm has a significantly longer runtime than other methods, due to the high computational cost of relying on extensive Monte Carlo simulations. In contrast, LA-DJaya runs approximately two to three orders of magnitude faster than CELF. DCGM++ has the shortest runtime because it can perform seed selection solely based on node index sorting, without iterative optimization. Compared to metaheuristic algorithms, LA-DJaya demonstrates superior computational efficiency compared to LIDDE, mainly because LIDDE has a higher computational cost per iteration, thus prolonging the overall search process. LA-DJaya's runtime is roughly equivalent to PHEE and TS-VA-MODE, but it is shorter than both of these algorithms on large-scale networks. This advantage is attributed to the adaptive terrain-aware evolutionary mechanism incorporated in LA-DJaya, which can more effectively guide the search in a more complex solution space, thereby accelerating its convergence process.

[0041] Therefore, the discrete Jaya influence maximization method based on adaptive terrain perception, as described above, can be widely applied to various social network analysis and information dissemination optimization scenarios. In the business field, this algorithm can provide core technical support for viral marketing: enterprises can accurately locate the most influential initial user groups in complex user relationship networks, initiating information cascade reactions with minimal promotion costs, thereby achieving efficient dissemination of new products, services, or brand content. Similarly, in the field of public administration, this algorithm demonstrates significant value. Public opinion monitoring departments can use it to identify key opinion leaders and network hub nodes, thereby effectively guiding the healthy development of public opinion and suppressing the spread of rumors and misinformation. Furthermore, in social welfare projects such as emergency management and public health promotion, this algorithm can optimize seed user selection strategies for activities such as emergency information dissemination and health behavior advocacy, significantly improving the coverage and dissemination efficiency of key information.

[0042] The application scope of this method can be further expanded to more fields that require optimization of information dissemination paths. In the financial industry, it can be used to identify key nodes in the promotion of financial products; in e-commerce platforms, it helps optimize the delivery strategies of promotional information; and in policy promotion, it can find the most effective dissemination paths for government propaganda. Its core advantage lies in its ability to adapt to network environments of different sizes and topologies. Whether in large social platforms with tens of thousands of nodes or in professional community networks in specific fields, it can reliably identify high-quality sets of influential seeds. This strong adaptability and robustness make this method highly practical and have broad prospects for promotion in real-world applications.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A discrete Jaya influence maximization method based on adaptive terrain perception, characterized in that, Includes the following steps: S1: Collect social network data and model a social network graph through user relationships; S2: Design a discrete Jaya update mechanism, construct the best candidate set and the worst candidate set based on the current individual, the best individual and the worst individual respectively, and perform discrete update operations according to the state of the candidate sets; S3: Construct the dominant terrain vector to calculate the fitness terrain index. When the fitness terrain index is 0, trigger the parallel elite local search mechanism. S4: Integrates the fitness terrain index with the discrete Jaya update mechanism to dynamically regulate the evolution process and finally output the optimal seed node set.

2. The discrete Jaya influence maximization method based on adaptive terrain perception according to claim 1, characterized in that, In S1, the social network graph is as follows: ,in, Represents a set of nodes. Represents the set of edges.

3. The discrete Jaya influence maximization method based on adaptive terrain perception according to claim 1, characterized in that, The optimal candidate set in S2 includes nodes that exist in the optimal individual but not in the current individual, defined as: The optimal candidate set provides a clear direction for improvement in the "addition" operation, guiding the current individual to evolve toward the optimal solution. The worst candidate set includes nodes that exist in both the current individual and the worst individual, and is defined as: The worst candidate set is used to identify low-quality nodes that reduce the quality of the solution, providing clear guidance for the "removal" operation and avoiding inferior nodes.

4. The discrete Jaya influence maximization method based on adaptive terrain perception according to claim 3, characterized in that, The discrete Jaya update mechanism in S2 specifically includes: New individuals are generated based on whether the best and worst candidate sets are empty; the specific update rules are as follows: ； in, express A node is randomly selected from the data. Indicates from A node is randomly selected from the data. Indicates from the current individual A node is randomly selected from the data. Indicates from A node randomly selected from the data; V Indicates the number of nodes. This represents the i-th individual in the population; When both the best and worst candidate sets are empty, a local influence value is introduced. LFV To quantify the influence of each node, local influence value LFV The specific definitions are as follows: ； In the formula, and Representing nodes respectively and A set of neighbors that jumps over time. and They represent the nodes respectively and and from the node arrive Activation probability; Deleted node Defined as the current individual The smallest LFV The value of the node is calculated using the following formula: ； Nodes to be added From the candidate set The selection process is as follows: ①Calculation Each node in LFV Values, and stored in a vector middle: ； ② Normalization is performed to obtain the probability vector. : ； ③ Based on candidate node set With probability vector Select nodes to be added through categorical sampling. : ； Better individuals are selected and retained for the next generation of the population through fitness comparison.

5. The discrete Jaya influence maximization method based on adaptive terrain perception according to claim 1, characterized in that, In S3, the dominant terrain vector is constructed based on the difference in fitness values ​​between two consecutive generations of individuals and is used to quantify the population's evolutionary trend; the fitness terrain index is the sum of the elements of the dominant terrain vector. The parallel elite local search mechanism performs one-hop neighbor replacement operations in parallel on each node in the current seed set and updates the seed set by selecting the best one.

6. The discrete Jaya influence maximization method based on adaptive terrain perception according to claim 5, characterized in that, Set the parent's fitness vector as The fitness vector of the offspring is Then the dominant terrain vector The i Each element is represented as: ； In the process of population evolution, the dominant terrain vector is calculated based on the fitness values ​​of two consecutive generations. The fitness terrain index R is defined as the sum of all elements of the dominant terrain vector, which is used to quantify the overall evolutionary trend of the population. .

7. The discrete Jaya influence maximization method based on adaptive terrain perception according to claim 4, characterized in that, If the fitness terrain index is not zero, execute the discrete Jaya update step in S4 until the termination condition is met; specifically, this includes the following steps: S41: Generate a candidate node set based on two-hop coverage, and randomly select nodes from it to construct the initial population; S42: For a given current individual Based on the current best and worst individuals, generate the best candidate set and the worst candidate set; apply the discrete update rule to... To generate new individuals ',exist 'and Individuals with higher fitness are selected to enter the next generation; S43: Construct a dominant terrain vector based on the fitness value of each individual in the population over two consecutive generations; calculate the fitness terrain index R based on the dominant terrain vector to assess whether the population is trapped in a local optimum. like This indicates that population evolution has stalled, triggering a parallel elite local search mechanism; like Return to the discrete Jaya update mechanism in S2 and iterate until the optimal solution is obtained; S44: When the maximum number of iterations is reached, the loop terminates, and the individual with the highest fitness is output from the final population as the optimal seed set.

8. A computer-readable medium, characterized in that, The computer program is stored therein, and when executed, it implements the steps of the discrete Jaya influence maximization method based on adaptive terrain perception as described in any one of claims 1-7.

Images