CN116756207A - Network key node mining method based on discount strategy and improved discrete crow search algorithm - Google Patents

Abstract

The invention relates to the technical field of computer complex network optimization, and discloses a network key node mining method based on a discount strategy and an improved discrete crow search algorithm, which includes: first preprocessing the citation network and converting it into an adjacency matrix, and inverting the network. to obtain the reverse network; then use the LRDiscount algorithm to initially screen the nodes of the reverse network to obtain the candidate node set C ; then optimize the candidate node set C according to the local optimization process of the improved discrete crow search algorithm; finally, from the optimized The nodes centrally select the optimal set and evaluate the node influence to obtain the final k key seed nodes. Compared with the existing technology, the present invention combines the influence discount strategy of network nodes with the improved discrete crow search algorithm, updates the node position to spread influence by imitating the crow search process, and generates influence through individual crows wandering in the citation network. Marginal gain to find key nodes.

Description

Mining of network key nodes based on discount strategy and improved discrete crow search algorithm method

Technical field

The invention belongs to the technical field of maximizing the influence of complex networks, and in particular relates to a network key node mining method based on a discount strategy and an improved discrete crow search algorithm.

Background technique

With the penetration of various mobile social services into human life and social interaction, social networks have played a role that cannot be ignored in information sharing, information dissemination and diffusion. The huge scale of social networks poses a great challenge to the traditional research on influence maximization, and also makes the research on this problem of greater practical significance. IM (InfluenceMaximization, IM) aims to solve K in social networks. Influential nodes (nodes represent social media users), and use the "word of mouth" effect to spread information to maximize the scope of influence of these nodes. Therefore, how to select K nodes in the network while ensuring time complexity and propagation effect is the main problem facing the problem of maximizing influence.

For the problem of influence maximization, the key lies in how to select key nodes in the network. This issue has been paid attention to by researchers since the 20th century. In the field of information science, with the deepening of research on information dissemination and diffusion issues in social media, these key nodes play an extremely important role in research on the dynamic evolution of network structure and communication control. Some researchers have applied the metaheuristic search algorithm to the mining of key nodes in the network. However, the disadvantage is that the algorithm is slow in execution and has high time complexity, making it unable to be applied in large-scale networks. Therefore, based on the current research field of this algorithm, Zhang et al. considered that neighbor nodes will play an important role in measuring node influence, so they proposed a heuristic algorithm PRDiscount combined with Pagerank. This algorithm has a clear discount The influence of all individuals with social relationships with the selected seeds. Although the metaheuristic algorithm improves the optimization efficiency of key node mining to a certain extent, this single-solution algorithm has only one solution during the iterative process. It has the advantages of simplicity and speed for solving the key node mining problem of small-scale networks. It is easy to fall into local optimality, resulting in network information redundancy. Then Gong et al. proposed a key node identification method based on the discrete particle swarm optimization algorithm. This method defines the position of the particle as the node number, and defines the speed of the particle as a mark to determine whether the node is updated. After multiple iterations, Find the global optimal particle Gbest under the constraints of the search conditions, and the position of this particle is the found optimal set of seed nodes. Therefore, in order to overcome the shortcomings of the algorithm, the current research on maximizing the influence of finding key nodes is combined with the swarm intelligence optimization algorithm, which simulates the cooperative behavior of biological populations or the evolutionary process of physical phenomena, because of its powerful heuristic search thinking and global search capabilities, which have been widely used in influence maximization optimization problems in recent years.

Therefore, how to design a method with high accuracy and low cost to obtain seed nodes in large-scale networks is extremely important.

Contents of the invention

Purpose of the invention: In view of the problems that traditional network key node mining has such problems as large network scale and large data volume, which results in low efficiency and high time complexity of directly mining key nodes, the present invention proposes a method based on discount strategy and improved discrete crow search. The network key node mining method of the algorithm combines the influence discount strategy of network nodes with the improved discrete crow search algorithm, and updates the node position to spread influence by imitating the crow search process. The influence is generated by individual crows wandering in the citation network. Marginal gain to find key nodes.

Technical solution: The present invention proposes a network key node mining method based on discount strategy and improved discrete crow search algorithm, which includes the following steps:

S1. Preprocess the citation network, convert the target network to obtain an adjacency matrix of the target network, and perform reverse operations on the citation network to obtain the reverse network;

S2. Discount the influence of each node pointing to the seed node in the reverse network in S1 according to the discount strategy algorithm LRDiscount to obtain the discounted node influence. Select the node with the largest value in turn and add it to the candidate node set C;

S3. Use the local optimization process of the improved discrete crow search algorithm to optimize the candidate node set C. The candidate node set is the candidate crow group. The improved discrete crow search algorithm adds parallel iterative processing on the basis of the traditional discrete crow search algorithm. That is, the discrete crow search only compares with the memory vector corresponding to the crow individual in each iteration, thereby updating the memory vector to meet the update of the crow individual position vector, and obtaining the optimized node set C ^* ;

S4. Select the optimal set from the optimized node set C ^* , and evaluate the node influence to obtain the final k key seed nodes.

Further, the specific steps for obtaining the candidate node set C in step S2 are:

S2.1. Add a background node bg to the reverse network and connect it to all nodes in the network to obtain a new network of N+1 nodes that are strongly connected;

S2.2. Assign an LR value of 1 unit to N nodes other than the background node bg. The LR value of the background node bg is 0;

S2.3. Distribute the LR value of this 1 unit evenly to its directly connected outgoing neighbor nodes, and continue iterating until a stable state is reached:

in, is the out-degree of node v _j , w _ji is the adjacency matrix element, indicating that there is an edge between node v _j and node v _i , then w _ji =1, otherwise it is 0;

S2.4. When the iteration ends, distribute the LR value LR _bg (t _c ) of the background node bg equally to all nodes in the network to obtain the final LR value of node _vi ;

S2.5. Discount strategy algorithm LRDiscount discounts the influence of neighbor nodes of each seed node, that is, in the reverse network, discounts the influence of each node pointing to the seed node, thereby obtaining the discounted node influence. force:

Among them, S is the seed node set, Represents the ratio of the number of seed nodes among the neighbor nodes of node v _i to the number of all neighbor nodes;

S2.6. Based on the final node influence, select the node with the largest value in INF and add it to the candidate node set C.

Further, in step S3, the specific steps of optimizing the candidate node set C using the local optimization process of the improved discrete crow search algorithm are:

S3.1. Initialize various data in the discrete crow search algorithm framework IDCSA, where the size of the crow group is N, the set of seed nodes to be solved is k, the maximum number of iterations t _max , the perception probability AP and the local search node Near neighbor range S initial parameters;

S3.2. Use the candidate node set C obtained in step S2 as the candidate crow group, and initialize the position vector x _i of the crow population = (node ₁ , node ₂ ,..., node _n ), the memory vector Memory _{t- 1} = [m ₁ , m ₂ ,..., m _n ] ^-1 ; and select the initial optimal solution position vector X* from the initialization population;

S3.3. Based on the node coding and the discretized representation of the crow group position vector and memory vector, the discretized search rules of the network space are constructed:

Among them, R (r _i , s) is a local search mechanism, and the symbol "∩" is defined as a logical crossover operation, whose purpose is to compare two position vectors to see whether there are duplicate nodes;

S3.4. Based on the node pool of the candidate node set C generated in step S2, define an objective function to calculate the function fitting values of N crow individuals, and use the local influence evaluation function LIE to perform the crow position vector evolution on the seed nodes. Approximate estimate of impact:

S3.5. Perform a local optimization search for the second-order neighbor nodes of each node in the candidate node set C according to the improved discrete crow search algorithm. If the marginal benefit value of a certain 2 neighbor node is larger than the marginal benefit of the node itself, use 2 The order neighbor node replaces the node in the current optimal solution, and the execution is repeated until the maximum number of iterations t _max reaches the upper limit.

Further, in step S3.5, the specific steps of performing local optimization search on the node set according to the improved crow search algorithm are:

1) Calculate the optimal position vector difference between the current crow individual i and the tracked crow individual j On this basis, perform a crossover operation to obtain the decision vector V _node and then determine whether to perform local search optimization;

2) Store the first- _order direct neighbor nodes of node After the node traversal is completed, the node set NodeSet is deduplicated to ensure that there are no duplicate nodes in the 2-hop neighbor nodes;

3) Calculate the LIE adaptation value of the seed set position vector x _i in sequence after the node at the corresponding position is replaced by its neighbor node, and select the node in these neighbor node sets NodeSet that can bring the maximum benefit in the x _i vector, and The nodes in the corresponding x _i vector are replaced.

Further, in S4, when the number of local search optimization iterations reaches t _mmax , the node set C ^* is obtained, and the nodes in the node set C ^* select k key node sets through the influence maximization algorithm.

Beneficial effects:

This invention is based on the discount strategy and the improved discrete crow search algorithm when mining key nodes, and solves the problems of slow convergence of the algorithm and low comprehensive influence of the optimal seed set. The specific performance is as follows:

(1) The LRDiscount algorithm is used to "discount" the interaction between nodes in the citation network, making full use of the network topology and node attribute information.

(2) In the local search optimization of the discrete crow search algorithm, the network node discount strategy is taken into consideration to screen out the seed set of the initial activation state, which avoids the shortcomings of the initialized seed node set on the convergence speed of the algorithm and the quality of the results.

(3) Adding a crossover operation to the discrete crow group search rule not only maintains the richness of the population during the search process and avoids falling into the local optimal solution, but also ensures that the updated best node has no duplicate nodes in the vector xi.

Aiming at the key node mining problem, this invention first converts the target problem into an optimization problem, then uses the proposed network node discount strategy to screen the initial seed set, and then improves the discrete crow search algorithm to optimize and solve the final seed set. This method optimizes the search for key nodes in complex networks in future research on influence maximization problems, and can achieve better results under the same conditions.

Description of the drawings

Figure 1 is an overall flow chart of the present invention;

Figure 2 is the sub-flow chart of the LRDiscount algorithm in Figure 1;

Figure 3 is a sub-flow chart of the local optimization process of the improved discrete crow search algorithm in Figure 1.

Detailed ways

The present invention will be further elucidated below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. After reading the present invention, those skilled in the art will make modifications to various equivalent forms of the present invention and all fall within the scope of the appended claims of this application. limited scope.

As shown in Figure 1, the specific steps of a network key node mining method based on discount strategy and improved discrete crow search algorithm disclosed by the present invention are as follows:

S1. Preprocess the citation network, convert the target network to obtain an adjacency matrix of the target network, and perform reverse operations on the citation network to obtain the reverse network.

S2. Discount the influence of each node pointing to the seed node in the reverse network in S1 according to the discount strategy algorithm LRDiscount to obtain the discounted node influence. Select the node with the largest value in turn and add it to the candidate node set C. The specific Proceed as follows:

S2.1. In the reverse network obtained in step S1, add a background node bg and connect it to all nodes in the network, thereby obtaining a new network of N+1 nodes that are strongly connected.

S2.2. Assign an LR value of 1 unit to N nodes other than the background node bg. The LR value of the background node bg is 0:

S2.3. Distribute the LR value of this 1 unit evenly to its directly connected outgoing neighbor nodes, and continue iterating until a stable state is reached:

in, is the out-degree of node v _j , w _ji is the adjacency matrix element, indicating that there is an edge between node v _j and node v _i , then w _ji =1, otherwise it is 0.

S2.4. When the iteration ends, distribute the LR value LR _bg (t _c ) of the background node bg equally to all nodes in the network to obtain the final LR value of node _vi :

S2.5. Discount strategy algorithm LRDiscount discounts the influence of neighbor nodes of each seed node, that is, in the reverse network, discounts the influence of each node pointing to the seed node, thereby obtaining the discounted node influence. force:

Among them, S is the seed node set, Indicates the ratio of the number of seed nodes among the neighbor nodes of node v _i to the number of all neighbor nodes.

S2.6. Based on the final node influence, select the node with the largest value in INF and add it to the candidate node set C.

S3. Use the local optimization process of the improved discrete crow search algorithm to optimize the candidate node set C. The candidate node set is the candidate crow group. The improved discrete crow search algorithm adds parallel iterative processing on the basis of the traditional discrete crow search algorithm. That is, the discrete crow search only compares with the memory vector corresponding to the crow individual in each iteration, thereby updating the memory vector to meet the update of the crow individual position vector, and obtains the optimized node set C ^* , as shown in Figure 2. The specific steps are as follows:

S3.1. Initialize various data in the discrete crow search algorithm framework IDCSA, where the size of the crow group is N, the set of seed nodes to be solved is k, the maximum number of iterations t _max , the perception probability AP and the local search node Initial parameters such as the nearest neighborhood range S.

S3.2. Initialize the position vector x _i of the crow population based on the candidate node set C obtained in step S2 = (node ₁ , node ₂ ,..., node _n ), and the memory vector Memory _t-1 = [m ₁ , m ₂ ,..., m _n ] ^-1 ; and select the initial optimal solution position vector X ^* from the initialization population.

S3.3. Based on the node coding and the discretized representation of the crow group position vector and memory vector, the discretized search rules of the network space are constructed:

Among them, R ( _ri , s) is a local search mechanism, and the symbol "∩" is defined as a logical crossover operation, whose purpose is to compare two position vectors to see whether there are duplicate nodes.

S3.4. Based on the node pool of the candidate node set C generated in step S2, define an objective function to calculate the function fitting values of N crow individuals, and use the local influence evaluation function LIE to perform the crow position vector evolution on the seed nodes. Approximate estimate of impact:

S3.5. Perform a local optimization search for the second-order neighbor nodes of each node in the candidate node set C according to the improved discrete crow search algorithm. If the marginal benefit value of a certain 2 neighbor node is larger than the marginal benefit of the node itself, use 2 The order neighbor node replaces the node in the current optimal solution, and the execution is repeated until the maximum number of iterations t _max reaches the upper limit. The specific process is as follows:

First, calculate the optimal position vector difference between the current crow individual i and the tracked crow individual j On this basis, a crossover operation is performed to obtain the decision vector V _node , and then it is judged whether to perform local search optimization.

Then store the _first -order direct neighbor nodes of node After the traversal is completed, the node set NodeSet is deduplicated to ensure that there are no duplicate nodes among the 2-hop neighboring nodes.

Finally, the LIE adaptation value of the seed set position vector x _i is calculated in turn after the node at the corresponding position is replaced by its neighbor node, and the node in these neighbor node sets NodeSet that can bring the maximum benefit in the x _i vector is selected, and The nodes in the corresponding x _i vector are replaced.

S4. Select the optimal set from the optimized node set C ^* , and evaluate the node influence to obtain the final k key seed nodes.

The invention can be combined with a computer system to complete the mining of seed nodes.

The invention discloses a network key node mining method based on discount strategy and improved discrete crow search algorithm, which can be used for mining key nodes in complex networks of different scales.

The above embodiments are only for illustrating the technical concepts and features of the present invention. Their purpose is to enable those familiar with this technology to understand the content of the present invention and implement it accordingly, and cannot limit the scope of protection of the present invention. All equivalent transformations or modifications made based on the spirit and essence of the present invention shall be included in the protection scope of the present invention.

Claims (5)

1. A network key node mining method based on discount strategy and improved discrete crow search algorithm is characterized by comprising the following steps:

s1, preprocessing a quotation network, converting a target network to obtain an adjacent matrix of the target network, and performing reverse operation on the quotation network to obtain a reverse network;

s2, in the reverse network described in S1, discount calculation is carried out on the node influence of the network according to an LRDiscount algorithm, so that the node influence after discount is obtained, and the node with the largest value is sequentially selected and added to the candidate node set C;

s3, optimizing a candidate node set C by utilizing a local optimization process of an improved discrete crow search algorithm, wherein the candidate node set is a candidate crow group, the improved discrete crow search algorithm adds parallelization iteration processing on the basis of a traditional discrete crow search algorithm, namely, the discrete crow search is only compared with a memory vector corresponding to a crow individual during each iteration, so that the memory vector is updated to meet the update of the position vector of the crow individual, and the optimized node set C is obtained ^* ；

S4, finally, from the optimized node set C ^* And selecting an optimal set, and evaluating the influence of the nodes, so as to obtain final k key seed nodes.

2. The network key node mining method based on the discount policy and the improved discrete crow search algorithm according to claim 1, wherein the specific steps of obtaining the candidate seed subset C in the step S2 are as follows:

s2.1, adding a background node bg in a reverse network, and connecting the background node bg with all nodes in the network, so as to obtain a new network of N+1 nodes which are strongly communicated;

s2.2, distributing LR values of 1 unit to N nodes except the background node bg, wherein the LR value of the background node bg is 0;

s2.3, equally distributing the LR value of 1 unit to directly connected neighbor nodes, and continuously iterating until reaching a stable state:

wherein ,for node v _j Degree of departure, w _ji For adjacency matrix elements, representing node v _j And node v _i Where there is an edge between, then w _ji =1, otherwise 0;

s2.4, after the iteration is finished, the LR value LR of the background node bg is calculated _bg (t _c ) Equally dividing to all nodes in the network to obtain node v _i Final LR values of (a);

s2.5, a discount policy algorithm LRDiscount discounts influence of neighbor nodes of each seed node, namely, in a reverse network, discounts influence of each node pointing to the seed node, so that discounted influence of the node is obtained:

wherein S is a seed node set,representing node v _i The number of seed nodes in the neighbor nodes accounts for the ratio of the number of all the neighbor nodes;

and S2.6, according to the finally obtained node influence, sequentially selecting the node with the largest value in the INF and adding the node into the candidate node set C.

3. The network key node mining method based on the discount policy and the improved discrete crow search algorithm according to claim 1, wherein the specific steps of optimizing the candidate node set C by using the local optimization process of the improved discrete crow search algorithm in the step S3 are as follows:

s3.1, initializing all data in an IDCSA (discrete crow search algorithm framework), wherein the crow group scale is N, the seed node set to be solved is k, and the maximum iteration number t is achieved _max Sensing probability AP and local search node neighbor domain range S initial parameters;

s3.2, using the candidate node set C obtained in the step S2 as a candidate crow group, thereby initializing a position vector x of the crow group _i ＝(node ₁ ,node ₂ ,…,node _n ) Memory vector Memory _t-1 ＝[m ₁ ,m ₂ ,…,m _n ] ^-1 The method comprises the steps of carrying out a first treatment on the surface of the And selecting an initial optimal solution position vector X from the initialized population ^* ；

S3.3, constructing a discretization search rule of a network space based on node coding and discretization representation of the crow group position vector and the memory vector:

wherein ,R(r_i S) is a local search mechanism, and the symbol "≡" is defined as a logical cross operation, which aims at comparing whether there are duplicate nodes in the two position vectors;

s3.4, defining an objective function to calculate function fitting values of N crow individuals based on the node pool of the candidate node set C generated in the step S2, and performing approximate evaluation on the influence of seed nodes in the crow position vector evolution by adopting a local influence evaluation function LIE:

s3.5, carrying out local optimization search on 2-order neighbor nodes of each node in the candidate node set C according to an improved discrete crow search algorithm, if the marginal benefit value of a certain 2-order neighbor node is large relative to the marginal benefit of the node, replacing the node in the current optimal solution by the 2-order neighbor node, and repeatedly executing until the maximum iteration number t is reached _max The upper limit is reached.

4. The network key node mining method based on the discount policy and the improved discrete crow search algorithm according to claim 3, wherein the specific steps of performing the local optimization search on the node set according to the improved crow search algorithm in step S3.5 are as follows:

1) Calculating the optimal position vector difference between the current crow individual i and the tracked crow individual jOn the basis, the cross operation is carried out to obtain a decision vector V _node Judging whether to perform local search optimization or not;

2) Node x _i Storing the first-order direct neighbor nodes of each node into a node set neighbor, traversing the nodes in sequence, finding out a 2-hop neighbor node set of the node, incorporating the 2-hop neighbor node set into the node set, and repeating the node set until the first-order neighbor nodes of each node are traversed, so as to ensure that the 2-hop neighbor nodes have no repeated nodes;

3) Sequentially calculating seed subset position vectors x _i LIE adaptation value after the node at the corresponding position is replaced by the neighboring node, and selecting the neighboring node set NodeSet to be able to x _i The node bringing the maximum benefit in the vector will correspond to x _i The nodes in the vector are replaced.

5. The network key node mining method based on discount policies and improved discrete crow search algorithm as recited in claim 1, wherein the number of local search optimization iterations in S4 reaches t _max The node set C obtained after ^* Node set C ^* And selecting k key node sets through an influence maximization algorithm.