CN115134247A

CN115134247A - Node identification method and device, electronic equipment and computer readable storage medium

Info

Publication number: CN115134247A
Application number: CN202210379035.5A
Authority: CN
Inventors: 荣钰; 马凯丽; 李蓝青
Original assignee: Shenzhen Tencent Computer Systems Co Ltd
Current assignee: Shenzhen Tencent Computer Systems Co Ltd
Priority date: 2022-04-12
Filing date: 2022-04-12
Publication date: 2022-09-30
Anticipated expiration: 2042-04-12
Also published as: CN115134247B

Abstract

The embodiment of the application discloses a node identification method, a node identification device, electronic equipment and a computer readable storage medium, wherein the method comprises the following steps: acquiring a connection relation between nodes in a network to be tested; determining a node influence parameter of each node in the network to be tested according to the connection relation, wherein the node influence parameter is used for representing the influence degree of the removal node on the network structure of the network to be tested; and selecting a target node with the network structure influence degree reaching a preset degree from the network to be tested according to the node influence parameter of each node. The key nodes of the network can be screened out according to the influence degree of the nodes on the network structure after the nodes are removed, namely, the larger the influence degree is, the more the nodes are critical, and the analysis and maintenance of the network are facilitated by identifying the key nodes of the network.

Description

Node identification method and device, electronic equipment and computer readable storage medium

Technical Field

The embodiment of the application relates to the technical field of computers, in particular to a node identification method and device, electronic equipment and a computer readable storage medium.

Background

With the rapid development of computer technology, the complexity of the network is continuously improved due to the increase of the number of nodes or the connection relationship among the nodes in the network, and the identification of the key nodes in the network has important significance for the understanding, application and maintenance of the network.

At present, key nodes are mainly determined according to the connection distance between each node and the highest dimension father node in the network, but the highest dimension father node influences the network on one side, so that the condition of inaccurate key node selection exists in the process of selecting the key nodes according to the highest dimension father node.

Disclosure of Invention

The embodiment of the application provides a node identification method, a node identification device, electronic equipment and a computer-readable storage medium, which can realize selection of key nodes having a large influence on a network structure.

In a first aspect, an embodiment of the present application provides a node identification method, including:

acquiring a connection relation between nodes in a network to be tested;

determining a node influence parameter of each node in the network to be tested according to the connection relation, wherein the node influence parameter is used for representing and removing the influence degree of the node on the network structure of the network to be tested;

and selecting a target node with the network structure influence degree reaching a preset degree from the network to be tested according to the node influence parameter of each node.

In a second aspect, an embodiment of the present application further provides a node identification apparatus, including:

the acquisition module is used for acquiring the connection relation between nodes in the network to be tested;

the determining module is used for determining a node influence parameter of each node in the network to be tested according to the connection relation, wherein the node influence parameter is used for representing and removing the influence degree of the node on the network structure of the network to be tested;

and the selecting module is used for selecting a target node with the network structure influence degree reaching a preset degree from the network to be tested according to the node influence parameter of each node.

Wherein, in some embodiments of the present application, the determining module comprises:

the first determining unit is used for determining the connection relation characteristics of each node in the network to be tested according to the connection relation;

and the second determining unit is used for respectively inputting the connection relation characteristics of each node into the trained node parameter model to obtain the node influence parameters corresponding to each node.

Wherein, in some embodiments of the present application, the apparatus further comprises a training module, the training module comprising:

the system comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring a training sample group, the training sample group comprises a sample network, a node removal strategy, a network state variation and a reward value, the network state variation is the variation before and after the network state after the action corresponding to the node removal strategy is executed, and the reward value is generated after the action corresponding to the node removal strategy is executed;

and the training unit is used for training a preset node parameter model according to the sample network, the node removal strategy, the network state variation and the reward value in the training sample group until a preset stopping condition is met, so as to obtain the trained node parameter model.

Wherein, in some embodiments of the present application, the obtaining unit includes:

the acquisition subunit is used for acquiring a sample network with the same network type as that of the network to be detected;

the first determining subunit is used for determining the number of nodes corresponding to the sample network reduced when each node in the sample network is removed;

the second determining subunit is used for determining nodes to be removed from the sample network according to the reduced number of nodes corresponding to each node in the sample network;

the generation subunit is used for generating a node removal strategy according to the node to be removed;

a third determining subunit, configured to use, as a network state variation after the node removal policy is executed, a variation of the sample network after the node to be removed is removed;

a fourth determining subunit, configured to determine, according to the number of nodes decreased when the node to be removed is removed, an incentive value corresponding to the node removal policy;

and the generating subunit is used for obtaining a training sample group according to the sample network, the node removal strategy, the network state variation and the reward value.

In some embodiments of the present application, the nodes to be removed include two nodes, and the second determining subunit is specifically configured to:

selecting a node with the reduced number of nodes meeting a first preset condition from the sample network, and taking the node as a first reference node;

removing the first reference node from the sample network to obtain a sample sub-network corresponding to the sample network;

selecting a node with the reduced number of nodes meeting the first preset condition from the sample sub-network, and taking the node as a second reference node;

and respectively taking the first reference node and the second reference node as nodes to be removed.

selecting at least two nodes with the reduced number of nodes meeting a second preset condition from the sample network, and taking the at least two nodes as first reference nodes respectively;

for each first reference node, removing the first reference node from the sample network to obtain a sample sub-network corresponding to the sample network, and constructing a network tree branch group according to the sample sub-network and the first reference node;

for each sample sub-network, selecting at least two nodes of which the number of nodes is reduced and which meet the second preset condition from the sample sub-network, and taking the at least two nodes as second reference nodes;

respectively establishing node groups for a first reference node and each second reference node in each network tree branch group, and counting the sum of the node reduction number corresponding to the first reference node and the node reduction number corresponding to the second reference node in each node group to obtain the node reduction number sum corresponding to each node group;

screening out the node reduction quantity and the target node reduction quantity sum meeting a third preset requirement according to the at least two node reduction quantity sums respectively corresponding to all the network tree branch groups;

and respectively taking the reduced number of the target nodes and the first reference nodes and the second reference nodes in the corresponding node groups as nodes to be removed.

In some embodiments of the present application, the third determining subunit is specifically configured to:

taking the variable quantity of the sample network after the first reference node is removed as the network state variable quantity of the node removal strategy corresponding to the first reference node after execution, and taking the variable quantity of the sample subnetwork after the second reference node is removed as the network state variable quantity of the node removal strategy corresponding to the second reference node after execution;

the fourth determining subunit is specifically configured to:

calculating the sum of the node reduction number corresponding to the first reference node and the node reduction number corresponding to the second reference node to obtain the node accumulated reduction number;

taking the accumulated reduced number of the nodes as a reward value of a node removal strategy corresponding to the first reference node, and taking the reduced number of the nodes corresponding to the second reference node as a reward value of a node removal strategy corresponding to the second reference node;

the generating subunit is specifically configured to:

and obtaining a training sample group according to the sample network, the node removal strategy corresponding to the first reference node, the network state variation after the execution of the node removal strategy corresponding to the first reference node, the reward value after the execution of the node removal strategy corresponding to the first reference node, the sample subnetwork, the node removal strategy corresponding to the second reference node, the network state variation after the execution of the node removal strategy corresponding to the second reference node, and the reward value after the execution of the node removal strategy corresponding to the second reference node.

Wherein, in some embodiments of the present application, the training unit comprises:

the fifth determining subunit is configured to determine a first training sample group and a current training sample group of the preset node parameter model during training;

the first selection subunit is configured to select a training sample group to be executed at a next time from the training sample groups according to the characterization information of the first training sample group, the characterization information of the previous training sample group, and the characterization information of the node removal strategy corresponding to the current training sample group;

and the training subunit is used for training the preset node parameter model according to the sample network, the node removal strategy, the network state variation and the reward value in the training sample group to be executed until a preset stopping condition is met, so as to obtain the trained node parameter model.

Wherein, in some embodiments of the present application, the training sample set comprises a first sample set and a second sample set, the data quality of the second sample set is higher than the data quality of the first sample set, the training unit comprises:

training a preset node parameter model according to the sample network, the node removal strategy, the network state variation and the reward value in the first sample group;

training a preset node parameter model according to the sample networks, the node removal strategies, the network state variable quantities and the reward values in the first sample group and the second sample group;

and training the preset node parameter model according to the sample network, the node removal strategy, the network state variation and the reward value in the second sample group until a preset stopping condition is met, and obtaining the trained node parameter model.

In some embodiments of the present application, the apparatus further includes a second target node selecting unit, where the second target node selecting unit includes:

the removing subunit is used for removing the target node from the network to be tested to obtain a sub-network to be tested corresponding to the network to be tested;

and the second selecting subunit is used for selecting a second target node with the network structure influence degree reaching a preset degree from the sub-network to be detected according to the node influence parameter of each node in the sub-network to be detected.

In a third aspect, an embodiment of the present application further provides an electronic device, where the electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps in the node identification method when executing the computer program.

In a fourth aspect, the present application further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps in the node identification method are implemented.

According to the method and the device, the node influence parameters of each node in the network to be detected are determined through the connection relation among the nodes in the network to be detected, the influence degree of the node on the network structure is removed according to the node influence parameter representation, the target node with the large influence degree on the network structure to be detected is screened out, and the identification of the network key node is achieved. The connection relationship of the nodes reflects the number of nodes establishing connection with the node in the network, and the larger the number of nodes connected with the node is, the higher the proportion of the connection relationship formed by the node in the network structure is, that is, the larger the influence of the node on the network structure is, and when the nodes connected with other nodes in a larger number are removed from the network structure, the larger the influence on the network structure is, so in the embodiment of the present application, the key nodes can be screened by the influence degree on the network structure after the nodes are removed. And the key nodes are selected according to the overall influence degree on the network structure, so that the factors considered during the selection of the key nodes are more comprehensive, and the selection of the key nodes is more accurate.

Drawings

In order to more clearly illustrate the technical solutions in the present application, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

Fig. 1 is a schematic view of a scenario of a node identification method according to an embodiment of the present application;

fig. 2 is a schematic flowchart of a node identification method according to an embodiment of the present application;

fig. 3 is a schematic diagram of a network structure provided in an embodiment of the present application;

FIG. 4 is a diagram of a data structure of a training sample set provided in an embodiment of the present application;

FIG. 5 is a flowchart illustrating a method for identifying key objects in a social network according to an embodiment of the present disclosure;

FIG. 6 is a schematic diagram illustrating progress of model training in a method for identifying a key object in a social network according to an embodiment of the present disclosure;

fig. 7 is a schematic structural diagram of a node identification apparatus according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of an electronic device provided in an embodiment of the present application.

Detailed Description

The technical solutions in the present application will be described clearly and completely with reference to the accompanying drawings in the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The embodiment of the application provides a node identification method and device, electronic equipment and a computer readable storage medium. Specifically, the embodiment of the present application provides a node identification apparatus suitable for an electronic device, where the electronic device includes a terminal or a server, where the terminal may be a computer, a personal computer, or a mobile phone, the server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, a middleware service, a domain name service, a security service, a Content Delivery Network (CDN), and a big data and artificial intelligence platform, and the server may be directly or indirectly connected through a wired or wireless communication manner.

In the embodiment of the present application, a terminal device may execute a node identification method alone, or a server may execute a node identification method alone, or a terminal and a server execute a node identification method together, please refer to fig. 1, which takes the case where a terminal device and a server execute a node identification method together as an example, where a specific execution process of the node identification method is as follows:

the terminal device 10 acquires the connection relationship between the nodes in the network to be tested, sends the connection relationship between the nodes in the network to be tested to the server 11, and the server 11 performs operation processing and analysis on the connection relationship between the nodes.

After receiving the connection relationship between the nodes in the network to be tested sent by the terminal device 10, the server 11 determines the node influence parameter of each node in the network to be tested according to the connection relationship between the nodes, selects the target node according to the node influence parameter, and returns the node information of the target node to the terminal device 10.

After receiving the node information of the target node returned by the server 11, the terminal device 10 displays the node information of the target node in the visual interface of the terminal device 10, so as to facilitate viewing the node information of the target node through the visual interface of the terminal device 10.

In the embodiment of the application, the node influence parameter is used for representing the degree of influence of the removing node on the network structure of the network to be tested. The target node is selected according to the influence degree on the network structure after removal, so that the target node with the larger influence degree on the network structure after removal can be selected, namely, the key node in the network to be tested is selected, and the identification of the key node of the network to be tested is realized.

In the embodiment of the present application, the connection relationship between the nodes may be obtained according to the connection condition between the nodes.

In the embodiment of the application, the node influence parameters of each node in the network to be detected are determined according to the connection relation between the nodes in the network to be detected, the influence degree of the node influence parameters on the network structure is removed according to the node, the target node with large influence degree on the network structure to be detected is screened out, and the identification of the network key node is realized. The connection relationship of the nodes reflects the number of nodes establishing connection with the node in the network, and the larger the number of nodes connected with the node is, the higher the proportion of the connection relationship formed by the node in the network structure is, that is, the larger the influence of the node on the network structure is, and when the nodes connected with other nodes in a larger number are removed from the network structure, the larger the influence on the network structure is, so in the embodiment of the present application, the key nodes can be screened by the influence degree on the network structure after the nodes are removed. And the key nodes are selected according to the overall influence degree on the network structure, so that the factors considered during the selection of the key nodes are more comprehensive, and the selection of the key nodes is more accurate.

The following are detailed below. It should be noted that the description sequence of the following embodiments is not intended to limit the priority sequence of the embodiments.

Referring to fig. 2, fig. 2 is a schematic flowchart illustrating a node identification method according to an embodiment of the present disclosure. The specific flow of the node identification method may be as follows:

101. and acquiring the connection relation between the nodes in the network to be tested.

In the embodiment of the present application, a network is composed of a plurality of nodes and links connecting the nodes, and represents a plurality of objects and their mutual relations, and a plurality of network nodes are connected by communication lines to form a certain geometric relationship, which is a network topology. The network may include a social network, an identity relationship network, an object association relationship network, and the like.

For example, for a base station distribution network, each base station in the base station distribution network corresponds to a node, a communication channel between each base station corresponds to a link between nodes, and when a communication relationship exists between two base stations, a link exists between nodes corresponding to two base stations. For example, for a social network, each object or member in the social network corresponds to a node, and an association relationship between the objects or members corresponds to a link between the nodes, and when an association relationship exists between two objects, a link exists between the nodes corresponding to the two objects.

In the embodiment of the present application, the connection relationship between the nodes reflects the connection condition between the nodes, and the connection relationship is used to represent whether the nodes establish connection or not, and whether the nodes are connected or not, that is, whether links exist between the nodes. In the embodiment of the application, the connection relationship between the nodes in the network to be tested can be determined according to the connection condition between the nodes.

102. And determining a node influence parameter of each node in the network to be tested according to the connection relation, wherein the node influence parameter is used for representing and removing the degree of influence of the node on the network structure of the network to be tested.

In this embodiment, the node influence parameter is used to characterize the degree of influence of the node on the network structure of the network to be tested, where in this embodiment, the node influence parameter may include a specific numerical value or a scalar that reflects the degree of influence. Namely, the node influence parameters corresponding to the nodes are different, and the influence of different nodes on the network structure of the network to be tested after the different nodes are removed can be clearly distinguished according to the node influence parameters, so that the nodes with large influence on the network structure can be screened out according to the node influence parameters.

In this embodiment of the present application, a network structure is formed by a plurality of nodes and links between the nodes, when a node in a network is removed or a link between the nodes is disconnected, a network structure corresponding to the network changes, for example, when the number of nodes removed in the network is large or the number of links disconnected between the nodes is large, the structure of the network changes greatly, and similarly, when the number of links between a node and other nodes in the network is large, the removal of the node reduces the number of links in the network more, that is, the change of the network structure is large, so that an influence parameter of the node on the network structure can be determined according to the connection relationship of the node.

In the embodiment of the application, a sub-network corresponding to an original network can be obtained by simulating the removal of each node in the original network, the influence situation of the removal of each node on the network structure can be obtained according to the difference between the sub-network and the original network obtained after the removal, and the influence data of the removal of each node on the original network structure is determined according to the number of nodes in the original network, the number of links between the nodes, the number of nodes in the sub-network and the links between the nodes.

For example, referring to fig. 3, fig. 3 is a schematic diagram of a network structure provided in the embodiment of the present application, taking k-core as an example of a network structure for measuring key nodes, assuming that we perform importance evaluation on

nodes

4 and 11, when we use a total amount of removed nodes resulting from removing nodes to be evaluated as an index, we count the reasons for removing nodes and encounter the following situations: 1. other node removals resulting from a single node removal, such as the removal of

nodes

1, 3, 5, 6 resulting from the removal of node 4, and the removal of

nodes

14, 15 resulting from the removal of node 11; 2. the removal of certain nodes may be independently caused by the removal of one of the nodes, as node 2 may be caused by either

node

4 or 11 being removed; 3. the removal of certain nodes needs to be caused collectively by the removal of multiple nodes, as node 12 is caused by the removal of

nodes

4 and 11. From

cases

2 and 3, we find that the importance evaluation of a node is coupled to other nodes in the node set, such as the node set {11} with which the node set is composed needs to be considered when evaluating the importance of node 4, which results in that we ignore the common effects of the possible node sets or over-value some nodes that result in repeated removal of other nodes when evaluating the effect of removing a single node. Therefore, it is difficult to acquire the key nodes in the network by a statistical or fixed heuristic algorithm.

Therefore, in this embodiment of the present application, the connection relationship between nodes may be analyzed through a machine learning model, and a node influence parameter corresponding to each node is predicted and obtained through an auto-regression method, that is, optionally, in some embodiments of the present application, the step "determining the node influence parameter of each node in the network to be tested according to the connection relationship" includes:

determining the connection relation characteristics of each node in the network to be tested according to the connection relation;

and respectively inputting the connection relation characteristics of each node into the trained node parameter model to obtain the node influence parameters corresponding to each node.

In the embodiment of the present application, the connection relation characteristic of each node can be obtained through the connection relation between each node and other nodes. In the embodiment of the present application, the connection relationship features of the network nodes may be extracted through a feature extraction algorithm or model, for example, the dependency relationship in the Graph is captured through a Graph network or a Graph neural network (Graph neural networks) according to information transfer between nodes in the network Graph.

And identifying and analyzing the connection relation characteristics of the nodes through the trained node parameter model to obtain the node influence parameters corresponding to each node, so as to quickly acquire the node influence parameters corresponding to the nodes.

In this embodiment of the present application, the node parameter model is obtained after being trained according to sample data, and the accuracy of model prediction is improved through training of the model, that is, optionally, in some embodiments of the present application, before the step "inputting the connection relationship characteristics of each node into the trained node parameter model respectively to obtain the node influence parameter corresponding to each node", the method further includes:

acquiring a training sample group, wherein the training sample group comprises a sample network, a node removal strategy, a network state variation and an incentive value, the network state variation is the variation before and after the network state is subjected to the action corresponding to the node removal strategy, and the incentive value is generated after the action corresponding to the node removal strategy is performed;

and training a preset node parameter model according to the sample network, the node removal strategy, the network state variation and the reward value in the training sample group until a preset stopping condition is met, and obtaining the trained node parameter model.

In the embodiment of the present application, the model may be trained in a reinforcement learning manner, that is, the model is trained by extracting an action policy, a state change amount, and a reward value from sample data, where the action policy corresponds to a node removal policy from the sample data in the embodiment of the present application, the state change amount corresponds to a change amount of a network state after the node removal policy is executed, and the reward value is correspondingly generated when the node removal policy is executed. In the embodiment of the application, the model can obtain learning information and update the model parameters by receiving environment rewards (feedback) to actions through reinforcement learning.

In the embodiment of the present application, a cross entropy loss function (cross entropy loss) is calculated according to the result of model prediction and the sample data expression, and is further fed back to the adjustment of the model parameters.

In this embodiment of the present application, the data of the training sample set may be obtained by extracting sample data, that is, optionally, in some embodiments of the present application, the step "obtaining the training sample set" includes:

acquiring a sample network with the same network type as the network to be detected;

determining the number of nodes corresponding to the sample network reduced when each node in the sample network is removed;

determining nodes to be removed from the sample network according to the reduced number of nodes corresponding to each node in the sample network;

generating a node removal strategy according to the node to be removed;

taking the variable quantity of the sample network after the node to be removed is removed as the network state variable quantity of the node removal strategy after the node removal strategy is executed;

determining an incentive value corresponding to the node removal strategy according to the reduced number of the nodes corresponding to the nodes to be removed during the removal;

and obtaining a training sample group according to the sample network, the node removal strategy, the network state variation and the reward value.

The sample network is selected according to the type of the network to be tested, so that the selected sample network is consistent with the type of the network to be tested, and under the condition that the network types are consistent, the reliability and the referential performance of sample data in model training can be improved. In the embodiment of the application, a non-scale network can be randomly generated according to a scale-free network generation model (Barab si-Albert), and then a sample network with the same type as the network to be tested is selected.

In the embodiment of the present application, the node reduction number is a number of nodes reduced by the node in the sample network after the node is removed, and the node reduction number is obtained by counting the node reduction number in the network after the node to be removed is removed, for example, when the node only maintains a link relationship with the node to be removed and the node to be removed is removed, so that the node is in an isolated state in the network, at this time, the node can be removed from the network as the node to be removed, and the node reduction number in the sample network of the node to be removed is obtained by counting each reduction node.

In this embodiment of the present application, the reduced number of nodes corresponding to each node in the sample network may be obtained by simulating removal of each node in the sample network, for example, by simulating removal of each node in the sample network, obtaining a sub-network corresponding to the sample network after removal of each node, and obtaining the reduced number of nodes corresponding to the sample network after removal of each node according to a difference between the sample network and the sub-network.

In the embodiment of the present application, when a node is removed, the reduced number of nodes corresponding to the sample network reflects the important condition and the influence condition of the node on the network structure, so that a node having a large influence on the network structure can be selected as a node to be removed according to the reduced number of nodes corresponding to each node, that is, after a node to be removed having a large influence on the network structure is selected according to the reduced number of nodes, the node to be removed can be understood as a key node corresponding to the sample network.

In the embodiment of the present application, a node removal policy refers to an action policy for removing a node, where in the embodiment of the present application, for a node removal policy generated for a node to be removed, removal of the node to be removed can be achieved after execution, that is, by generating a node removal policy according to the node to be removed, after the generated node removal policy is executed, a removal action for the node to be removed can be performed. For the node removal strategy in each training sample group, the node removal strategy can be used as an action strategy for reinforcement learning to train the model.

The node removal strategy is executed according to the node number, wherein the node number is reduced, namely the node to be removed is greatly influenced on the network structure, so that the node removal strategy can be guaranteed to have a large reward value after being executed according to the reward value corresponding to the node to be removed obtained according to the node number. For example, in the embodiment of the present application, the number of nodes corresponding to each node may be reduced as the reward value corresponding to the node when the node is removed.

The method includes the steps that a node with the largest number of nodes reduced (namely, a key node of a sample network) is used as a node to be removed, and a model is subjected to reinforcement learning training according to an action strategy corresponding to the node to be removed and an incentive value of the action strategy during execution, so that the model can learn the action strategy, the incentive value and relationship information among the key nodes in sample data during training, for example, the more the node is critical, the greater the incentive value corresponding to the removed node is, so that the trained model can predict and identify the key node in the network to be tested, for example, when the network to be tested and a larger expected incentive value are input into a trained node parameter model, the key node in the network to be tested can be screened out according to an actual incentive value and an input expected incentive value of each node during removal. The expected reward value in model prediction may be a large prior value.

In this embodiment of the present application, the to-be-removed nodes corresponding to the sample network may include multiple to-be-removed nodes, and the model may be trained by using the multiple to-be-removed nodes, the network state variation amount corresponding to each to-be-removed node, the node removal strategy, and the reward value, so as to improve the accuracy of model prediction after training. That is, optionally, in some embodiments of the present application, the step "determining a node to be removed from the sample network according to the number of nodes corresponding to each node in the sample network" includes:

In the embodiment of the application, the first preset condition includes a condition that the number of nodes in the current network is reduced to the maximum, and the selected first reference node can meet the requirement of the most critical node in the sample network by taking the node with the maximum number of nodes in the sample network as the first reference node.

After the first reference node is removed from the sample network to obtain the sample sub-network, the selected second reference node can meet the requirement of the most critical node in the sample sub-network by taking the node with the largest reduced number of nodes in the sample sub-network as the second reference node.

In the embodiment of the present application, the selection manner is implemented by taking greedy algorithm (greedy algorithm) as an example, and a best selection corresponding to each network state is selected to implement a local optimal solution, so that the influence of the first reference node on the selection of the second reference node is overcome, and the requirement for sequentially selecting the key nodes is met, for example, the requirement is met with the expectation that the model outputs one most key node each time and selects the secondary key nodes from the remaining networks.

In the embodiment of the present application, in order to increase the selection diversity, when the node is selected in each network state, the selection method of the node may be further selected according to a random probability theory, for example, in the embodiment of the present application, the node with the largest number of nodes reduced may be selected according to the number of nodes corresponding to each node, and the node may also be selected in a random manner as the node to be removed. The randomness increases the variety of node selection, and enriches the sample data of model training to a certain extent.

In this embodiment, two or more key nodes may also be selected in each network state, and a key node in the next network state is selected according to the branch network corresponding to each key node, that is, optionally, in some embodiments of the present application, the step "the number of nodes to be removed is reduced according to the number of nodes corresponding to each node in the sample network, and the determining of the nodes to be removed from the sample network" includes:

In the embodiment of the present application, the method for selecting multiple nodes in each network state is exemplified by a beam search (beam search) algorithm. Because at least two nodes are selected in each network state, compared with the situation that one node is selected in each network state, the searching and selecting space is improved, the diversity of the selected nodes is improved, meanwhile, the situation that the wrong selection easily occurs when one node is selected in each network state is also avoided, and the accuracy is ensured to a certain extent.

And finally, screening out relatively better results by comparing the reduced number of the nodes in each group, and improving the accuracy of selection.

In the embodiment of the present application, for the requirement that the number of the nodes to be removed is three, the second reference node may be removed from the sample sub-network to obtain a sample sub-network corresponding to the sample sub-network, and the third reference node is selected from the sample sub-network according to the reduced number of nodes of each node. For the case that the third reference node is single, the third reference node may be directly used as the node to be removed, and for the case that the third reference node is multiple, the node group with the largest node reduction number and the largest node reduction number may be selected as the node to be removed according to the node reduction number sum corresponding to each node group.

In this embodiment of the present application, for the case of two nodes to be removed, the network state variation before and after removal of each node is determined according to the network state variation information before and after removal of the node, that is, optionally, in some embodiments of the present application, the step "taking the variation of the sample network after removal of the node to be removed as the network state variation after execution of the node removal policy" includes:

and taking the variation of the sample network after the first reference node is removed as the network state variation of the node removal strategy corresponding to the first reference node after the execution, and taking the variation of the sample sub-network after the second reference node is removed as the network state variation of the node removal strategy corresponding to the second reference node after the execution.

The network state variation of each node after the removal operation can be obtained by comparing the network state information of each node before and after the removal, that is, the network state variation of the node removal strategy corresponding to each node after the execution is obtained.

In the embodiment of the present application, since the nodes are selected according to the node reduction number corresponding to the node, the node reduction number of the node selected first is theoretically greater than the node reduction number of the node selected later, for example, after the nodes in a plurality of network states are selected, the node reduction number corresponding to the first reference node is greater than the node reduction number corresponding to the second reference node. That is, the node reduction number corresponding to each node may be used as a reward value for the node removing operation, so as to achieve obtaining of the reward value of each node. The model can conveniently learn the characteristics, and the node with the maximum reward value when the node is removed is taken as the most key node.

In this embodiment of the present application, in the training process of the model, the reward value of each node when being removed may be promoted in a manner of maximizing the cumulative reward value, that is, optionally, in some embodiments of the present application, the step "determining the reward value corresponding to the node removal policy according to the number of nodes that correspond to the node to be removed when being removed" includes:

calculating the sum of the node reduction number corresponding to the first reference node and the node reduction number corresponding to the second reference node to obtain the node cumulative reduction number;

and taking the accumulated reduced number of the nodes as the reward value of the node removal strategy corresponding to the first reference node, and taking the reduced number of the nodes corresponding to the second reference node as the reward value of the node removal strategy corresponding to the second reference node.

The method comprises the steps of adding a secondary key node corresponding to a current node and the reward value of a secondary key node corresponding to the secondary key node to the reward value of the current node, so that the node reward value is accumulated, and the reward value is maximized. For example, when the number of the key nodes is three, that is, the first reference node, the second reference node, and the third reference node, the sum of the reduced numbers of the nodes corresponding to the first reference node, the second reference node, and the third reference node may be used as the bonus value of the first reference node, the sum of the reduced numbers of the nodes corresponding to the second reference node and the third reference node may be used as the bonus value of the second reference node, and the sum of the reduced numbers of the nodes corresponding to the third reference node may be used as the bonus value of the third reference node.

The method comprises the steps that the rewarding values corresponding to the nodes with different key degrees are arranged in a layered mode, so that the model can learn the relation between the size of the rewarding value and the key degree of the nodes, the size of the rewarding value corresponding to the key nodes is improved through maximizing the accumulated rewarding value, the larger the rewarding value can be learned by the model, the more appropriate the action strategy is to be executed, namely, the larger the rewarding value is, the more accurate the key nodes are removed, selected or identified.

In the embodiment of the present application, for a case that a plurality of (more than two) nodes to be removed exist, networks, network state variation, node removal strategies, and reward values corresponding to the plurality of nodes may be combined, and the training sample group data as a whole is obtained. Taking two to-be-removed nodes as an example, a process of constructing a training sample group by two to-be-removed nodes in a single sample network is described, that is, optionally, in some embodiments of the present application, the step "obtaining a training sample group according to the sample network, the node removal policy, the network state variation, and the reward value" includes:

By learning the relationship among the plurality of nodes to be removed in each network, the corresponding plurality of network state change information, the node removal strategy and the reward value, the diversity of model training data is improved, and the learning of the relationship among the network state change, the node removal strategy and the reward value is facilitated.

For example, please refer to fig. 4, fig. 4 is a data structure diagram of training sample sets provided in the present embodiment, where each training sample set includes an original sample network and three pieces of key node information, and each piece of key node information includes a network state before removing the node, a node removal action policy corresponding to removing the node, and a reward value corresponding to the corresponding node removal action policy. The original sample network and the corresponding three key node information are used as a training sample group, so that the model can conveniently obtain learning information and update model parameters through reward (feedback) of the environment on the execution of the action strategy in the training process.

In this embodiment, when a plurality of training sample sets are included and a next training sample set is selected for model training, the training sample set to be executed at the next time may be determined according to the representation of the current model training result to improve the stability of the model training process, that is, optionally, in some embodiments of the present application, the step "training a preset node parameter model according to a sample network, a node removal strategy, a network state variation and a reward value in the training sample set until a preset stop condition is satisfied to obtain a trained node parameter model" includes:

determining a first training sample group and a current training sample group of the preset node parameter model during training;

selecting a training sample group to be executed at the next moment from the training sample group according to the characterization information of the first training sample group, the characterization information of the previous training sample group and the characterization information of the node removal strategy corresponding to the current training sample group;

and training the preset node parameter model according to the sample network, the node removal strategy, the network state variation and the reward value in the training sample group to be executed until a preset stop condition is met, and obtaining the trained node parameter model.

In the embodiment of the application, the representation information of the data is the overall or total representation information of the data, and reflects the overall characteristics of the data. The method comprises the steps that a training sample group to be executed at the next moment is selected through a first training sample group, a current training sample group and the characterization information of the node movement strategy corresponding to the current training sample group, so that the characterization information of the selected training sample group to be executed can be consistent with the characterization information of the first training sample group, the current training sample group and the node removal strategy corresponding to the current training sample group, the characteristics of consistent and stable training direction are presented, and the model training efficiency is improved.

In this embodiment, the node parameter model may include a decision transformer model (decision transformer), and a framework of encoding (encoder) and decoding (decoder) is adopted to solve the autoregressive process, specifically: in the encoding stage, a graph neural network is adopted to represent the state (network state) and the action (node removal strategy), and a multilayer perceptron (MLP) is used to represent the accumulated reward value corresponding to the node removal action strategy. In a decoding stage, firstly, a predicted result representation is obtained through a decision converter, then, for each state, the most initial network representation, the current network representation and the representation of the node removal strategy in the previous step are used as semantic representations, and the semantic representations and the action representations in the rest sub-networks are used for performing attention mechanism calculation to select the node removal strategy in the next step, namely, the next training data is selected.

In this embodiment of the present application, different training data corresponding to the model in different training stages may be set according to a source of a training sample group or a data overall quality condition, that is, optionally, in some embodiments of the present application, the training sample group includes a first sample group and a second sample group, data quality of the second sample group is higher than data quality of the first sample group, and then "training a preset node parameter model according to a sample network, a node removal strategy, a network state variation and a reward value in the training sample group until a preset stop condition is satisfied, to obtain a trained node parameter model" includes:

training a preset node parameter model according to the sample networks, the node removal strategies, the network state variation and the reward values in the first sample group and the second sample group;

In the embodiment of the present application, the difference in data source may be understood as a difference in a method for acquiring training data, that is, sample data for training is extracted from historical data by using different algorithms, where the difference in data quality may also be divided according to the difference in algorithms to which the sample data is acquired, for example, the sample data extracted by using a better algorithm is higher in data quality than the sample data extracted by using a common algorithm. For example, in the embodiment of the present application, for the requirement of sample data, a training sample group obtained by a bundle search algorithm may be used as the second sample group, and a training sample group obtained by extraction by a greedy algorithm may be used as the first sample group.

In the embodiment of the application, the reliability or quality of the model data is continuously improved in the model training process, the optimization effect of the model is improved, and the continuous optimization of the model is realized. And through the mixed training of the first sample group data and the second sample group data, the transition of model training data is realized, and the stability of the model training process is ensured.

In this embodiment of the present application, when at least two key nodes need to be extracted from a network to be tested, the first key node may be removed from the network after the first key node is identified, a second key node is selected based on the removed network, and so on, a plurality of key nodes meeting the quantity requirement are selected, that is, optionally, in some embodiments of the present application, after the step "selecting a target node whose network structure influence degree reaches a preset degree from the network to be tested according to a node influence parameter of each of the nodes", the method further includes:

removing the target node from the network to be tested to obtain a sub-network to be tested corresponding to the network to be tested;

and selecting a second target node with the network structure influence degree reaching a preset degree from the sub-network to be tested according to the node influence parameter of each node in the sub-network to be tested.

The method for selecting the secondary key nodes based on the residual network is also consistent with the acquisition process or method of the training data, and the influence of the first key node on the selection process can be avoided when the second key node is selected.

In the embodiment of the present application, based on the accumulation strategy of the reward value during model training (refer to the reward value accumulation part of sample data), when a second target node is selected from the sub-network to be tested, a difference between the expected reward value and the reduced number of nodes of the first target node may be used as a second expected reward value input when the second target node is selected, and according to the second expected reward value, a second target node, that is, a second key node, is selected from the sub-network to be tested.

In this embodiment, when three key nodes are needed, the third key node may be screened from the to-be-tested sub-network after the second target node is removed according to the to-be-tested sub-network, for example, a third reward value input by the third target node is obtained according to a difference between the second expected reward value of the second target node and a reduced number of nodes corresponding to the second target node, and the third key node in the to-be-tested sub-network is obtained according to the third reward value and the to-be-tested sub-network. The values of the reward values corresponding to the nodes with different key degrees are different, and the acquisition of the nodes with relative keys is realized according to the relative values of the reward values.

103. And selecting a target node with the network structure influence degree reaching a preset degree from the network to be tested according to the node influence parameter of each node.

The larger the influence of the node on the network structure after removal is, the higher the importance of the node in the network structure is, so in the embodiment of the application, the target node in the network can be screened according to the influence degree of the node on the network structure after removal, and the selection of the network key node is realized.

The network is convenient to analyze, understand, apply or maintain through identification of the key nodes of the network, sudden collapse or large-scale damage of the network structure is prevented, for example, trend of the network is predicted through checking the dynamic state of the key nodes in a social network, and the key nodes are protected to reduce the breakdown phenomenon of the infrastructure network.

With reference to fig. 5, the method for identifying a social network key object in the embodiment of the present application will be described below, where an embodiment of the method for identifying a social network key object in the embodiment of the present application includes:

201. and acquiring the association relation among the objects in the social network.

The network is composed of a plurality of nodes and links between the nodes, the social network is established by a plurality of objects and incidence relations between the objects, wherein the objects in the social network can comprise users, accounts or labels of the users, the accounts or the labels correspond to the nodes in the network, the incidence relations between the objects correspond to the links between the network nodes, and for example, when the incidence relations exist between two objects, the links exist between the nodes corresponding to the two objects. In the embodiment of the present application, the social network may be a network established by a game friend, for example, for the same game, the social network for the game is obtained according to an association relationship between players in the game.

202. And determining an object influence parameter of each object in the social network according to the association relationship, wherein the object influence parameter is used for representing the influence degree of the object on the network structure of the social network.

In the embodiment of the present application, the object influence parameter is used to characterize a degree of influence of separating the object on a network structure of the social network. In the implementation of the present application, the object influence parameters may include specific numerical values or scalars reflecting influence degrees, and the object influence parameters corresponding to each object are different, so that an object having a large influence on the social network may be screened out through the object influence parameters corresponding to each object, that is, a key object in the social network is screened out.

Because the association relationship of each object is different, and the objects with more association relationship reflect in the social network, that is, more connection links exist with the object node, that is, a plurality of other object nodes establish connection with the object node, that is, a plurality of other object nodes have connection relationships with the object node. Wherein, the more the association relationship of the object, the more the object which is associated with the object is indicated, i.e., the more objects that can be affected by the object, e.g., for high-end players, professional players or high-frequency players in the game, the friends of the player are more, the player has contact with a plurality of other players (such as a plurality of friends), when the player leaves the game, a plurality of players are easy to be offline (the friends are also offline), or when the player logs out from the game, it is easy to cause the game logging-out behavior of a plurality of other players (for example, the teammates of a team often log out accounts, and other teammates in the team are also easily affected to log out accounts), so that the more related objects, the higher the importance degree in the social network is, the greater the influence is, and therefore, the influence parameter or the influence degree of each object on the social network structure can be obtained according to the association relationship of the object.

When the total quantity of the other objects after the objects are separated from the network is taken as an object importance degree analysis index, the following problems exist in the statistics of the total quantity of the other objects: 1. the removal of a single object from the network structure may cause other objects to detach from the network, 2, the detachment of a single object may be caused by the detachment of other single objects or multiple objects, and 3, the removal of some objects may be caused by the removal of multiple objects in combination. Therefore, because the removal of the object is influenced in various situations, it is difficult to obtain the total detachment amount of other objects after detachment of each node in the network through a statistical or fixed heuristic algorithm, that is, it is difficult to determine the key object in the social network structure.

Therefore, in this embodiment of the present application, the association relationship of each object may be analyzed by using a machine learning model, and the degree of influence of each object on the social network is predicted in an auto-regression manner, that is, optionally, in some embodiments of the present application, the step "determining the object influence parameter of each object in the social network according to the association relationship" includes:

determining the incidence relation characteristics of each object in the social network according to the incidence relation;

and respectively inputting the incidence relation characteristics of each object into the trained influence parameter model to obtain the object influence parameters corresponding to each object.

The incidence relation characteristics of each object can be obtained according to a plurality of incidence relations existing with the object, and the incidence relation characteristics of the object corresponding to each object node can also be obtained by extracting the dependency relation between each object node in the social network graph according to the graph neural network.

Optionally, in some embodiments of the present application, before predicting the object influence parameters according to the influence parameter model, the model needs to be trained first to improve accuracy of model prediction, and before the step "inputting the association relationship features of each object into the trained influence parameter model respectively to obtain the object influence parameters corresponding to each object", the method further includes:

acquiring a training sample group, wherein the training sample group comprises a sample social network, an object separation strategy, a network state variation and an incentive value, the network state variation is the variation before and after the network state after the action corresponding to the object separation strategy is executed, and the incentive value is generated after the action corresponding to the object separation strategy is executed;

and training a preset object parameter model according to the sample social network, the object separation strategy, the network state variation and the reward value in the training sample group until a preset stop condition is met, and obtaining the trained object parameter model.

In the embodiment of the present application, the model may be trained in a reinforcement learning manner, that is, the model is trained by extracting an action policy, a state variation, and a reward value from sample data, where the action policy corresponds to an object separation policy in the sample data in the embodiment of the present application, the state variation corresponds to a variation of a network state after the object separation policy is executed, and the reward value is correspondingly generated when the object separation policy is executed. In the embodiment of the application, the model can obtain learning information and update the model parameters by receiving the reward (feedback) of the environment to the action corresponding to the object separation strategy through reinforcement learning.

In the model training process, a cross entropy loss function (cross entropy loss) can be calculated through the model prediction result and the sample data expression, and then the cross entropy loss function is fed back to the adjustment of the model parameters.

In some embodiments of the present application, optionally, the step "obtaining a training sample set" includes:

obtaining a sample social network of the same network type as the social network;

determining a reduced number of objects corresponding to the sample social network when each object in the sample social network is removed;

determining an object to be separated from the sample social network according to the reduced number of objects corresponding to each object in the sample social network;

generating an object separation strategy according to the object to be separated;

taking the variation of the sample social network after the object to be separated is separated as the variation of the network state of the object separation strategy after the object separation strategy is executed;

determining an award value corresponding to the object separation strategy according to the reduced number of the objects corresponding to the objects to be separated after the objects are separated;

and obtaining a training sample group according to the sample network, the object separation strategy, the network state variation and the reward value.

The sample social network is selected according to the type of the social network, so that the selected sample social network is consistent with the type of the social network, and under the condition that the network types are consistent, the reliability and the referential performance of sample data in model training can be improved.

In this embodiment of the present application, the social network and the sample social network may be defined as: g ═ V, E, the neighboring nodes of each node V are denoted as n (V) ═ { u | (u, V) ∈ E }, the degree of the node V is defined as deg (V) | (n (V) |, and the subgraph (k-core) of the network is defined as

Wherein, C _k (G) The degrees of all nodes are greater than or equal to k. We define the set of key nodes S as

Wherein, C _r (S) is defined as a reduction value of the number of k-core subgraph nodes before and after S is removed from the set V: c _k (V)-C _K (V \ S), wherein V \ S is a cohesive subgraph (such as a social son corresponding to a social network) with a key node set removedNetwork, sample social network corresponding to sample social network).

The objects to be separated are selected according to the object reduction number corresponding to each object during separation, namely, the object with the largest object reduction number is selected as the object to be separated, so that the object reduction number corresponding to the selected object to be separated is large, namely, the influence of the selected object to be separated on the network structure is large, and therefore when the reward value corresponding to the separated object is obtained according to the object reduction number, the object separation strategy can be guaranteed to have a large reward value after execution. For example, in the embodiment of the present application, the number of the object reduction corresponding to each object may be used as the reward value corresponding to the object when the object is separated from the social network.

The method comprises the steps of taking an object with the largest number of reduced objects (namely, a key object of a sample social network) as an object to be separated, carrying out reinforcement learning training on a model according to an action strategy corresponding to the object to be separated and an incentive value of the action strategy during execution, enabling the model to learn relationship information among the action strategy, the incentive value and the key object in sample data during training, wherein the more the object is critical, the greater the incentive value corresponding to the separated object is, so that the trained model can predict and identify key nodes in the social network to be tested, for example, when the social network to be tested and a larger expected incentive value are input into a trained influence parameter model, the key object in the social network to be tested can be screened out according to an actual incentive value and the input expected incentive value of each object during separation, and each object is simulated to be separated from the social network to be tested through the model, and obtaining the actual reward value corresponding to each object when the objects are separated, and screening out the key objects according to whether the actual reward value is large enough and is close to the expected reward value. Wherein, the expected reward value in model prediction can be a big prior value.

In the embodiment of the application, the objects to be separated in the sample social network may include a plurality of objects to be separated, and the accuracy of model prediction after training can be improved by training the model through the plurality of objects to be separated, the network state variation corresponding to each object to be separated, the object separation strategy and the reward value, and the method or process for acquiring the two objects to be separated is described below by taking the objects to be separated as two examples. That is, optionally, in some embodiments of the present application, the to-be-separated object includes two, and the step "determining the to-be-separated object from the sample social network according to the number of the objects corresponding to each object in the sample social network" includes:

selecting an object with a reduced number of objects meeting a first preset condition from the sample social network, and taking the object as a first reference object;

removing the first reference object from the sample social network to obtain a sample social sub-network corresponding to the sample social network;

selecting one object with the reduced number of objects meeting the first preset condition from the sample social subnetwork, and taking the object as a second reference object;

and respectively taking the first reference object and the second reference object as objects to be separated.

The process is exemplified by greedy algorithm (greedy algorithm), that is, an optimal result is selected in each network state, and the optimal result is used as an optimal result at each time step to ensure that each time is the optimal choice.

Correspondingly, in each network state, a plurality of objects may also be selected, and the most significant combination is selected from the plurality of network states as the finally selected key object, that is, optionally, in some embodiments of the present application, the step "to-be-separated object includes two, and the step of determining the to-be-separated object from the sample social network according to the number of the objects corresponding to each object in the sample social network" includes:

selecting at least two objects with the reduced number of objects meeting a second preset condition from the sample social network, and respectively taking the at least two objects as first reference objects;

for each first reference object, removing the first reference object from the sample social network to obtain a sample social sub-network corresponding to the sample social network, and constructing a network tree branch group according to the sample social sub-network and the first reference object;

for each sample social sub-network, selecting at least two objects with the number of reduced objects meeting the second preset condition from the sample social sub-network, and taking the at least two objects as second reference objects;

aiming at each network tree branch group, respectively establishing an object group for a first reference object and each second reference object in the network tree branch group, and counting the sum of the object reduction number corresponding to the first reference object and the object reduction number corresponding to the second reference object in each object group to obtain the object reduction number sum corresponding to each object group;

screening out the reduced number of the objects and the reduced number of the target objects meeting a third preset requirement according to the reduced number sum of at least two objects respectively corresponding to all the network tree branch groups;

and respectively taking the reduced number of the target objects and the first reference object and the second reference object in the corresponding object group as objects to be separated.

The object selection method is exemplified by a beam search (beam search) algorithm. Because at least two nodes are selected in each network state, compared with the situation that one node is selected in each network state, the searching and selecting space is improved, the diversity of the selected nodes is improved, meanwhile, the situation that the wrong selection easily occurs when one node is selected in each network state is also avoided, and the accuracy is ensured to a certain extent.

In the embodiment of the application, for the requirement that three objects to be separated are required, the second reference object can be separated from the sample social sub-network to obtain the sample social sub-network corresponding to the sample social sub-network, and the third reference object is selected from the sample social sub-network according to the number of the objects of each object reduced. For the case that the third reference object is single, the third reference object may be directly used as the object to be separated, and for the case that the third reference object is multiple, the object in the object group with the largest object reduction number and the largest object reduction number may be selected as the object to be separated according to the object reduction number sum corresponding to each object group.

In some embodiments of the present application, the step "taking the variation of the sample network after the object to be separated as the variation of the network state after the object separation policy is executed" may include:

and taking the variation of the sample network after the first reference object is separated as the variation of the network state after the object separation strategy corresponding to the first reference object is executed, and taking the variation of the sample subnetwork after the second reference object is separated as the variation of the network state after the object separation strategy corresponding to the second reference object is executed.

The network state variation of each object after the separation operation can be obtained by comparing the network state information of each object before and after the separation, that is, the network state variation of the object separation strategy corresponding to each object after the execution is obtained.

Optionally, in some embodiments of the present application, the step "determining the reward value corresponding to the object removal policy according to the number of the objects to be separated reduced when the object to be separated is separated" includes:

calculating the sum of the object reduction number corresponding to the first reference object and the object reduction number corresponding to the second reference object to obtain the accumulated reduction number of the objects;

and taking the accumulated reduction quantity of the objects as the reward value of the object removal strategy corresponding to the first reference object, and taking the reduction quantity of the objects corresponding to the second reference object as the reward value of the object removal strategy corresponding to the second reference object.

The method comprises the steps of obtaining a reward value of a current object, and obtaining a secondary key object corresponding to the current object and a reward value of a secondary key object corresponding to the secondary key object. For example, when the number of the key objects is three, that is, the first reference object, the second reference object, and the third reference object, the sum of the numbers of the objects corresponding to the first reference object, the second reference object, and the third reference object may be reduced as the bonus value of the first reference object, the sum of the numbers of the objects corresponding to the second reference object and the third reference object may be reduced as the bonus value of the second reference object, and the sum of the numbers of the objects corresponding to the third reference object may be reduced as the bonus value of the third reference object.

The model can learn the relationship between the size of the reward value and the key degree of the object through the layered arrangement of the sizes of the reward values corresponding to the objects with different key degrees, and the size of the reward value corresponding to the key object is improved through maximizing the accumulated reward value, so that the larger the reward value can be learned by the model, the more appropriate characteristic is executed by the action strategy, namely, the larger the reward value is, the more accurate the action of separating, selecting or identifying the key object is.

In the embodiment of the present application, for a case that there are a plurality of (two or more) objects to be separated, networks, network state variation, object separation strategies, and reward values corresponding to the plurality of objects may be combined, so as to obtain the whole training sample set data. Optionally, in some embodiments of the present application, the step "obtaining a training sample group according to the sample social network, the object separation policy, the network state variation and the reward value" includes:

obtaining a training sample group according to the sample social network, the object separation strategy corresponding to the first reference object, the network state variation after the execution of the object separation strategy corresponding to the first reference object, the reward value after the execution of the object separation strategy corresponding to the first reference object, the sample subnetwork, the object separation strategy corresponding to the second reference object, the network state variation after the execution of the object separation strategy corresponding to the second reference object, and the reward value after the execution of the object separation strategy corresponding to the second reference object.

By learning the relationship among the plurality of objects to be separated in each network, the corresponding plurality of network state change information, the object separation strategy and the reward value, the diversity of model training data is improved, and the learning of the relationship among the network state change, the object separation strategy and the reward value is facilitated.

In this embodiment of the present application, optionally, in some embodiments of the present application, the step "training a preset object parameter model according to a sample network, an object separation strategy, a network state variation and an incentive value in the training sample set until a preset stop condition is met to obtain a trained object parameter model" includes:

determining a first training sample set and a current training sample set of the preset object parameter model during training;

and training the preset object parameter model according to the sample network, the object separation strategy, the network state variation and the reward value in the training sample group to be executed until a preset stop condition is met, and obtaining the trained object parameter model.

In the embodiment of the application, the representation information of the data is the overall or overall representation information of the data, and reflects the overall characteristics of the data. The method comprises the steps that a training sample group to be executed at the next moment is selected through a first training sample group, a current training sample group and the characterization information of a node movement strategy corresponding to the current training sample group, so that the characterization information of the selected training sample group to be executed can be kept consistent with the characterization information of an object separation strategy corresponding to the first training sample group, the current training sample group and the current training sample group, namely, the characteristics of consistent and stable training direction are presented, and the model training efficiency is improved.

In this embodiment of the present application, sample data may be extracted through different algorithms, and for data obtained in different manners, different sample data may be selected at different training stages of the model to train the model, that is, optionally, in some embodiments of the present application, a training sample group includes a first sample group and a second sample group, data quality of the second sample group is higher than data quality of the first sample group, then "train a preset object parameter model according to a sample network, an object separation strategy, a network state variation and a reward value in the training sample group until a preset stop condition is met, to obtain a trained object parameter model", including:

training a preset object parameter model according to the sample network, the object separation strategy, the network state variation and the reward value in the first sample group;

training a preset object parameter model according to the sample networks, the object separation strategies, the network state variation and the reward values in the first sample group and the second sample group;

and training the preset object parameter model according to the sample network, the object separation strategy, the network state variation and the reward value in the second sample group until a preset stop condition is met, and obtaining the trained object parameter model.

In the embodiment of the present application, different data sources may be understood as different methods for acquiring training data, that is, sample data for training is extracted from historical data through different algorithms, where the difference in data quality may also be divided according to the difference in algorithms corresponding to the acquired sample data, for example, the sample data extracted through a better algorithm is higher in data quality than the sample data extracted through a common algorithm. For example, please refer to fig. 6, where fig. 6 is a schematic diagram illustrating a progress of model training in a social network key object recognition method provided in an embodiment of the present application, in the embodiment of the present application, a training sample group obtained by a bundle search algorithm may be used as a second sample group, a training sample group obtained by extraction of a greedy algorithm may be used as a first sample group, training is performed using data generated by the greedy algorithm at the beginning of training, training continues by mixing the data generated by the greedy algorithm and the data generated by the bundle search algorithm after the training reaches a certain stage, and finally, training is performed using the data generated by the bundle search algorithm. Wherein the data of the training process is increased.

Optionally, in some embodiments of the present application, after the step "selecting a target node with a network structure influence degree reaching a preset degree from the social network according to an object influence parameter of each object" is performed, the method further includes:

separating the target object from the social network to obtain a social sub-network corresponding to the social network;

and selecting a second target object with a network structure influence degree reaching a preset degree from the social sub-network according to the object influence parameters of each object in the social sub-network.

The method for selecting the secondary key objects based on the residual network also accords with the acquisition process or method of the training data, and the influence of the first key object on the selection process can be avoided when the second key object is selected.

In the embodiment of the present application, based on the model training-time reward value accumulation policy (refer to the sample data reward value accumulation section above), when a second target object is selected from the social subnetwork, a difference between the expected reward value and the object reduction number of the first target object may be used as a second expected reward value input when the second target object is selected, and according to the second expected reward value, a second target object, that is, a second key object, is selected from the social subnetwork.

In this embodiment, when three key objects are needed, the third key object may be selected from the social sub-network after the second target object is removed according to the social sub-network, for example, a third reward value input by the third target object is obtained according to a difference between a second expected reward value of the second target object and a reduced number of objects corresponding to the second target object, and the third key object in the social sub-network is obtained according to the third reward value and the social sub-network. The sizes of the reward values corresponding to the objects with different key degrees are different, and the acquisition of the objects with relative keys is realized according to the relative sizes of the reward values.

203. And selecting a target object with a network structure influence degree reaching a preset degree from the social network according to the object influence parameter of each object.

The larger the influence of the object on the network structure after separation is, the higher the importance of the object in the network structure is, so in the embodiment of the application, the target object in the network can be screened according to the degree of influence of the object on the network structure after separation, and the selection of the network key object is realized.

According to the method and the device for identifying the key objects in the social network, the object influence parameters of each object in the social network are determined through the association relation between the objects in the social network, and the target object with a large influence degree on the social network structure is screened out according to the influence degree of the object separation represented by the object influence parameters on the network structure, so that the key objects in the network are identified. The method facilitates analysis, understanding, application or maintenance of the social network through identification of key objects of the social network, prevents sudden collapse or extensive damage of a network structure, and predicts the trend of the network by checking the dynamics of the key objects in the social network.

In order to better implement the node identification method of the application, the application also provides a node identification device based on the node identification method. The meaning of the third target word is the same as that in the node identification method, and specific implementation details can refer to the description in the method embodiment.

Referring to fig. 7, fig. 7 is a schematic structural diagram of a node identification apparatus provided in the present application, where the node identification apparatus may include:

an obtaining module 301, configured to obtain a connection relationship between nodes in a network to be tested;

a determining module 302, configured to determine a node influence parameter of each node in the network to be tested according to the connection relationship, where the node influence parameter is used to characterize and remove a degree of influence of the node on a network structure of the network to be tested;

a selecting module 303, configured to select, according to the node influence parameter of each node, a target node from the network to be tested, where the network structure influence degree reaches a preset degree.

Among others, in some embodiments of the present application, the determining module 302 includes:

the acquisition subunit is used for acquiring a sample network with the same network type as the network to be detected;

taking the variation of the sample network after the first reference node is removed as the variation of the network state of the node removal strategy corresponding to the first reference node after the execution, and taking the variation of the sample sub-network after the second reference node is removed as the variation of the network state of the node removal strategy corresponding to the second reference node after the execution;

the fourth determining subunit is specifically configured to:

the generating subunit is specifically configured to:

In the embodiment of the application, the obtaining module 301 obtains the connection relation between nodes in the network to be tested, then the determining module 302 determines the node influence parameter of each node in the network to be tested according to the connection relation, the node influence parameter is used for representing and removing the node influence degree on the network structure of the network to be tested, and then the selecting module 303 selects the target node with the network structure influence degree reaching the preset degree in the network to be tested according to each node influence parameter of the node.

In addition, the present application also provides an electronic device, as shown in fig. 8, which shows a schematic structural diagram of the electronic device related to the present application, specifically:

the electronic device may include components such as a processor 401 of one or more processing cores, memory 402 of one or more computer-readable storage media, a power supply 403, and an input unit 404. Those skilled in the art will appreciate that the electronic device configuration shown in fig. 8 does not constitute a limitation of the electronic device and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components. Wherein:

the processor 401 is a control center of the electronic device, connects various parts of the entire electronic device using various interfaces and lines, and performs various functions of the electronic device and processes data by operating or executing software programs and/or modules stored in the memory 402 and calling data stored in the memory 402. Optionally, processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor, wherein the application processor mainly handles operating systems, object interfaces, application programs, and the like, and the modem processor mainly handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 401.

The memory 402 may be used to store software programs and modules, and the processor 401 executes various functional applications and data processing by operating the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data created according to use of the electronic device, and the like. Further, the memory 402 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 access to the memory 402.

The electronic device further comprises a power supply 403 for supplying power to the various components, and preferably, the power supply 403 is logically connected to the processor 401 through a power management system, so that functions of managing charging, discharging, and power consumption are realized through the power management system. The power supply 403 may also include any component of one or more dc or ac power sources, recharging systems, power failure detection circuitry, power converters or inverters, power status indicators, and the like.

The electronic device may further include an input unit 404, and the input unit 404 may be used to receive input numeric or character information and generate a keyboard, mouse, joystick, optical or trackball signal input related to object setting and function control.

Although not shown, the electronic device may further include a display unit and the like, which are not described in detail herein. Specifically, in this embodiment, the processor 401 in the electronic device loads an executable file corresponding to a process of one or more application programs into the memory 402 according to the following instructions, and the processor 401 runs the application programs stored in the memory 402, thereby implementing the steps in any node identification method provided by the present application.

According to the method and the device, the node influence parameters of each node in the network to be detected are determined through the connection relation between the nodes in the network to be detected, the influence degree of the node influence parameters on the network structure is removed according to the node, the target node with large influence degree on the network structure to be detected is screened out, and the identification of the network key node is achieved. The connection relationship of the nodes reflects the number of nodes establishing connection with the node in the network, and the larger the number of nodes connected with the node is, the higher the proportion of the connection relationship formed by the node in the network structure is, that is, the larger the influence of the node on the network structure is, and when the nodes connected with other nodes in a larger number are removed from the network structure, the larger the influence on the network structure is, so in the embodiment of the present application, the key nodes can be screened by the influence degree on the network structure after the nodes are removed. And the key nodes are selected according to the overall influence degree on the network structure, so that the factors considered during the selection of the key nodes are more comprehensive, and the selection of the key nodes is more accurate.

It will be understood by those skilled in the art that all or part of the steps of the methods of the above embodiments may be performed by instructions or by associated hardware controlled by the instructions, which may be stored in a computer readable storage medium and loaded and executed by a processor.

To this end, the present application provides a computer-readable storage medium having a computer program stored thereon, where the computer program can be loaded by a processor to execute the steps in any one of the node identification methods provided in the present application.

Wherein the computer-readable storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.

Since the instructions stored in the computer-readable storage medium can execute the steps in any node identification method provided by the present application, beneficial effects that can be achieved by any node identification method provided by the present application can be achieved, for details, see the foregoing embodiments, and are not described herein again.

The above detailed description is provided for a node identification method, a node identification apparatus, an electronic device, and a computer-readable storage medium, and a specific example is applied in this document to explain the principles and embodiments of the present invention, and the description of the above embodiment is only used to help understanding the method and the core idea of the present invention; meanwhile, for those skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.

It should be understood that, in the embodiments of the present application, related data such as user information, inter-object association, friend relationship, game behavior, etc. are involved, when the above embodiments of the present application are applied to specific products or technologies, user permission or consent needs to be obtained, and the collection, use and processing of the related data need to comply with the relevant laws and regulations and standards of the relevant countries and regions.

Claims

1. A node identification method, comprising:

acquiring a connection relation between nodes in a network to be tested;

2. The method according to claim 1, wherein the determining the node impact parameter of each node in the network under test according to the connection relationship comprises:

3. The method according to claim 2, wherein before the connection relationship characteristic of each node is input into the trained node parameter model to obtain the node influence parameter corresponding to each node, the method further comprises:

4. The method of claim 3, wherein the obtaining a training sample set comprises:

obtaining a sample network with the same network type as the network to be tested;

generating a node removal strategy according to the node to be removed;

5. The method according to claim 4, wherein the nodes to be removed include two nodes, and the determining the nodes to be removed from the sample network according to the reduced number of nodes corresponding to each node in the sample network comprises:

6. The method of claim 4, wherein the number of the nodes to be removed is two, and wherein the determining the nodes to be removed from the sample network according to the reduced number of the nodes corresponding to each node in the sample network comprises:

and respectively taking the reduced number of the target nodes and the first reference node and the second reference node in the corresponding node group as nodes to be removed.

7. The method according to any one of claims 5 or 6, wherein the taking a variation of the sample network after the node to be removed is removed as a variation of a network state of the node removal policy after the node removal policy is executed comprises:

the determining of the reward value corresponding to the node removal strategy according to the node reduction number corresponding to the node to be removed when the node to be removed is removed includes:

obtaining a training sample group according to the sample network, the node removal strategy, the network state variation and the reward value, including:

8. The method according to claim 3, wherein the training of the preset node parameter model according to the sample network, the node removal strategy, the network state variation and the reward value in the training sample group until a preset stop condition is met to obtain the trained node parameter model comprises:

9. The method according to claim 3, wherein the training sample group includes a first sample group and a second sample group, the data quality of the second sample group is higher than that of the first sample group, and the training of the preset node parameter model according to the sample network, the node removal strategy, the network state variation and the reward value in the training sample group is performed until a preset stop condition is met, so as to obtain the trained node parameter model, including:

and training the preset node parameter model according to the sample network, the node removal strategy, the network state variation and the reward value in the second sample group until a preset stop condition is met, and obtaining the trained node parameter model.

10. The method according to claim 1, wherein after selecting a target node with a network structure influence degree reaching a preset degree from the network to be tested according to the node influence parameter of each node, the method further comprises:

11. A node identifying apparatus, comprising:

a determining module, configured to determine a node influence parameter of each node in the network to be tested according to the connection relationship, where the node influence parameter is used to characterize and remove a degree of influence of the node on a network structure of the network to be tested;

12. An electronic device, characterized in that the electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the node identification method according to any one of claims 1-10 when executing the computer program.

13. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when being executed by a processor, carries out the steps of the node identification method according to any one of claims 1 to 10.