A method and device for predicting links between network nodes

By acquiring network topology information and utilizing node importance calculation and connection prediction models, the problem of unknown network node topology relationships was solved, enabling accurate prediction of links between network nodes and restoration of topology relationships, thus improving the completeness and accuracy of network analysis.

CN118540263BActive Publication Date: 2024-11-26NO 15 INST OF CHINA ELECTRONICS TECH GRP
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410550620.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-06
Publication Date
2024-11-26
Estimated Expiration
2044-05-06

AI Technical Summary

Technical Problem

Existing technologies cannot effectively reconstruct and predict the topological relationships between undetected network nodes in computer or communication networks, leading to information asymmetry and incompleteness, which affects the accuracy of network action planning.

Method used

By acquiring a set of network topology information, using a node importance calculation model and a node connection prediction model, a node representation vector is constructed, cluster analysis and link prediction are performed, feature extraction and prediction sub-models are integrated, and the node connection prediction model is trained to predict the link probability between network nodes.

Benefits of technology

It enables the reconstruction of network node topology under incomplete network conditions, improves network connectivity, supports in-depth resource mining and analysis and the formation of network situation intelligence, and enhances the accuracy and completeness of network analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118540263B_ABST
    Figure CN118540263B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for predicting links between network nodes, the method comprising: obtaining a network topology information set; the network topology information set is used to characterize the connectivity of the network; the network comprises a number of nodes and links between the nodes; performing connection identification processing on the network topology information set to obtain a first link result information set and an updated network topology information set; performing a comparison and judgment operation on the first link result information set and the updated network topology information set to obtain a network unpredicted edge information set; the network unpredicted edge information set comprises network unpredicted edge information; using a trained node connection prediction model, processing the network unpredicted edge information set to obtain link prediction results between network nodes; the link prediction results between network nodes are used to characterize whether the links between network nodes are connected.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the fields of communications and artificial intelligence, and in particular to a method and device for predicting links between network nodes. Background Art

[0002] The actual computer or communication network environment is complex, and the situation information obtained through various technical means is often incomplete, with information missing and asymmetric. In order to improve network information and the accuracy of network action planning, and to solve the problem of inconsistent and incomplete information, it is necessary to solve the restoration and prediction of network missing information, especially how to analyze the possibility of a link between two nodes that are not detected in the network layer through known network nodes and network structure information, restore the topological relationship between nodes and other information, support physical topology restoration, logical link prediction, network role association analysis, resource deep mining analysis and other practical applications, and form the ability to conduct association mining and analysis of structured and unstructured data resources. Summary of the invention

[0003] The present invention mainly solves the problem of how to analyze the possibility of a link between two nodes whose connection is not detected in the network layer through known information such as network nodes and network structures, and restore information such as the topological relationship between nodes, and supports practical applications such as physical topology restoration, logical link prediction, network role association relationship analysis, and deep resource mining analysis.

[0004] In a first aspect of an embodiment of the present application, a method for predicting a link between network nodes is disclosed, comprising:

[0005] S1, obtaining a network topology information set; the network topology information set is used to characterize the connectivity of the network; the network includes a number of nodes and links between the nodes;

[0006] The network topology information set includes network topology information; the network topology information includes node information and direct link information of the node;

[0007] The node information includes the node number, the node number directly connected to the node, and the node importance value;

[0008] The direct link information of the node includes the number and sequence number of nodes passed through the direct link and the direct link score; the direct link of the node refers to the link between the current node and an adjacent node that can be connected through less than or equal to A1 transit nodes;

[0009] S2, performing connection identification processing on the network topology information set to obtain a first link result information set and an updated network topology information set;

[0010] S3, performing a comparison and judgment operation on the first link result information set and the updated network topology information set to obtain a network unpredicted edge information set; the network unpredicted edge information set includes network unpredicted edge information;

[0011] S4, using the trained node connection prediction model, processing the network unpredicted edge information set to obtain link prediction results between network nodes; the link prediction results between network nodes are used to characterize whether the links between network nodes are connected.

[0012] The node importance value is calculated using a node importance calculation model;

[0013] The node importance calculation model has the following calculation expression:

[0014] v i =arctan(s i / l i ) / (pi / 2),

[0015] Among them, v i is the node importance value of the i-th node, s i is the sum of the node numbers directly connected to the i-th node, l i is the sum of the direct link scores of all direct links of the node.

[0016] The performing connection identification processing on the network topology information set to obtain a first link result information set and an updated network topology information set includes:

[0017] S21, constructing a node representation vector based on the network topology information in the network topology information set; the elements of the node representation vector include the node number, the variance value of the node numbers directly connected to the node, the node importance value, the mean number of nodes of all direct links of the node, the median value of the node numbers of all direct links of the node, and the mean score of all direct links of the node;

[0018] S22, performing cluster analysis on all node representation vectors to obtain type information and node representation vector information contained in the type;

[0019] S23, sorting the type information from large to small according to the number of node representation vectors contained in the type to obtain a sorting sequence;

[0020] S24, extracting the first N1 type information in the sorted sequence;

[0021] S25, confirming that the direct link information of all nodes corresponding to the node representation vectors included in the first N1 types of information is the first link result information; using all the first link result information, construct a first link result information set;

[0022] S26, for each direct link information in the network topology information set, determining whether it appears in the first link result information set, and obtaining a first determination result;

[0023] If the first judgment result is yes, no operation is performed;

[0024] If the first judgment result is no, deleting the direct link information from the network topology information set;

[0025] S27, completing the determination process of all direct link information in the network topology information set to obtain an updated network topology information set.

[0026] The comparing and judging operation of the first link result information set and the updated network topology information set to obtain the network unpredicted edge information set includes:

[0027] S31, initializing an unpredicted edge information set of the network; constructing an edge information set using all two-to-two different nodes of the network; the edge information set includes edge information; the edge information includes two nodes corresponding to the edge and their node information;

[0028] S32, for each node in the edge information set, determine whether it appears in the nodes passed by the direct link in the direct link information in the first link result information set, and obtain a second determination result;

[0029] If the second judgment result is no, adding the side information to the network unpredicted side information set; if the second judgment result is yes, no operation is performed;

[0030] S33, after completing the judgment processing of each side information in the side information set, extracting the network topology information corresponding to the node included in each side information of the network unpredicted side information set;

[0031] S34, constructing an edge information vector by using the network topology information corresponding to the node included in each edge information; the edge information vector is obtained by concatenating the node information vectors of the two nodes corresponding to the edge; the node information vector includes the node number, the node number directly connected to the node, the node importance value, the number and number of nodes passed through on the direct link of the node, and the direct link score of the node;

[0032] S35, adding the side information vector of each edge to the corresponding side information in the network unpredicted side information set to obtain the network unpredicted side information; the network unpredicted side information includes the side information and the corresponding side information vector;

[0033] S36, using all the network unpredicted side information, construct a network unpredicted side information set.

[0034] The node connection prediction model includes: a first feature extraction sub-model, a second feature extraction sub-model and a fusion prediction sub-model;

[0035] The first feature extraction sub-model includes a first node feature extractor, a first encoder, a first decoder, a first feature reconstructor, a second node feature extractor, a second encoder, a second decoder, and a second feature reconstructor; the first encoder includes a first node encoder and a first link encoder; the second encoder includes a second node encoder and a second link encoder;

[0036] The output end of the first node feature extractor is connected to the input end of the first encoder; the output ends of the first node encoder and the second link encoder are connected to the input end of the first decoder; the output ends of the second node encoder and the first link encoder are connected to the input end of the second decoder; the output end of the first decoder is connected to the input end of the first feature reconstructor; the output end of the second decoder is connected to the input end of the second feature reconstructor;

[0037] The second feature extraction sub-model includes a third node feature extractor, a third encoder, a third decoder, a third feature reconstructor, a fourth node feature extractor, a fourth encoder, a fourth decoder, and a fourth feature reconstructor; the third encoder includes a third node encoder and a third link encoder; the fourth encoder includes a fourth node encoder and a fourth link encoder;

[0038] The output end of the third node feature extractor is connected to the input end of the third encoder; the output ends of the third node encoder and the fourth link encoder are connected to the input end of the third decoder; the output ends of the fourth node encoder and the third link encoder are connected to the input end of the fourth decoder; the output end of the third decoder is connected to the input end of the third feature reconstructor; the output end of the fourth decoder is connected to the input end of the fourth feature reconstructor;

[0039] The fusion prediction sub-model includes a first input module, a first convolution module, a second convolution module, a third convolution module, a first pooling module, a fourth convolution module and a first fully connected module;

[0040] The input end of the first input module of the fusion prediction sub-model is connected to the output ends of the first feature reconstructor, the second feature reconstructor, the third feature reconstructor, and the fourth feature reconstructor respectively; the output end of the first input module of the fusion prediction sub-model is connected to the input end of the first convolution module of the fusion prediction sub-model; the output end of the first convolution module of the fusion prediction sub-model is connected to the input end of the second convolution module of the fusion prediction sub-model; the output end of the second convolution module of the fusion prediction sub-model is connected to the input end of the third convolution module of the fusion prediction sub-model; the output end of the third convolution module of the fusion prediction sub-model is connected to the input end of the first pooling module of the fusion prediction sub-model; the output end of the first pooling module of the fusion prediction sub-model is connected to the input end of the fourth convolution module of the fusion prediction sub-model; the output end of the fourth convolution module of the fusion prediction sub-model is connected to the input end of the first fully connected module of the fusion prediction sub-model; the output end of the first fully connected module is used to output the link prediction result.

[0041] The training process of the node connection prediction model includes:

[0042] Initialize the number of training iterations;

[0043] Obtain network link data set;

[0044] Using the node information in the network link data set as input data to input into the node connection prediction model; using the node link information in the network link data set as label information;

[0045] Using the node connection prediction model, the input data is processed to obtain a link prediction result;

[0046] Performing difference calculation processing on the obtained link prediction result and label information to obtain a difference value;

[0047] Determine whether the difference value satisfies a convergence condition, and obtain a third determination result;

[0048] When the third judgment result is no, judging whether the training iteration number value is equal to the training number threshold, and obtaining a fourth judgment result;

[0049] When the fourth judgment result is no, determining that the model training state does not meet the training termination condition;

[0050] When the fourth judgment result is yes, determining that the model training state satisfies the training termination condition;

[0051] When the third judgment result is yes, determining that the model training state satisfies the training termination condition;

[0052] When the model training state does not meet the training termination condition, the parameter update model is used to update the parameters of the node connection prediction model, the training iteration number value is increased by 1, and the execution is triggered to input the node information in the network link data set as input data into the node connection prediction model;

[0053] When the model training state satisfies the training termination condition, the training process of the node connection prediction model is completed to obtain a trained node connection prediction model.

[0054] The parameter updating model is:

[0055]

[0056] θ←θ+v:

[0057] In the formula, x (i) is the label information of the i-th node information in the network link data set, is the loss function, v is the parameter update value of the node connection prediction model, θ is the parameter of the node connection prediction model, η is the initial parameter learning rate, α is the momentum angle parameter, 0≤α≤1, It means to find the partial derivative of variable θ, f(x (i) ; θ) represents the link prediction result calculated by the node connection prediction model for the i-th node information, f(·) is the calculation function corresponding to the node connection prediction model; exp represents the power operation of the constant e; η and α are preset values.

[0058] In a second aspect of the embodiment of the present application, a link prediction device between network nodes is disclosed, which is used to implement the link prediction method between network nodes, including: an information acquisition module, a first processing module, a second processing module, and a node connection prediction model;

[0059] The first processing module is connected to the information acquisition module and the second processing module;

[0060] The node connection prediction model is connected to the second processing module;

[0061] The information acquisition module is used to acquire a set of network topology information;

[0062] The first processing module is used to perform connection identification processing on the network topology information set to obtain a first link result information set and an updated network topology information set;

[0063] The second processing module is used to compare and determine the first link result information set and the updated network topology information set to obtain a network unpredicted edge information set;

[0064] The node connection prediction model is used to process the network unpredicted edge information set to obtain link prediction results between network nodes.

[0065] In a third aspect of the embodiments of the present application, a link prediction device between network nodes is disclosed, the device comprising:

[0066] A memory storing executable program code;

[0067] a processor coupled to the memory;

[0068] The processor calls the executable program code stored in the memory to execute the link prediction method between network nodes.

[0069] In a fourth aspect of an embodiment of the present application, a computer-storable medium is disclosed, wherein the computer-storable medium stores computer instructions, and when the computer instructions are called, they are used to execute the link prediction method between network nodes.

[0070] The beneficial effects of the present invention are:

[0071] When the network situation data is incomplete, the method of the present invention can integrate the advantages of the overall network factors and the edge clustering method, solve the problem of restoring and predicting missing information, predict the possibility of connections between network nodes, restore the topological relationship of the target network nodes, and improve the communication relationship of the network nodes. It can be directly applied to network applications such as filling in missing data information and forming network situation intelligence.

[0072] The method of the present invention can be used in network analysis to realize functions such as resource standardization processing, data mining analysis, and network target profiling. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a flow chart for implementing the method of the present invention. DETAILED DESCRIPTION

[0074] In order to better understand the content of the present invention, an embodiment is given here.

[0075] Figure 1 It is a flow chart for implementing the method of the present invention.

[0076] In a first aspect of an embodiment of the present application, a method for predicting a link between network nodes is disclosed, comprising:

[0077] S1, obtaining a network topology information set; the network topology information set is used to characterize the connectivity of the network; the network includes a number of nodes and links between the nodes;

[0078] The network topology information set includes network topology information; the network topology information includes node information and direct link information of the node;

[0079] The node information includes the node serial number, the serial number of the node directly connected to the node, and the node importance value; the node directly connected to the node refers to the node directly connected to the node without passing through a transit node;

[0080] The direct link information of the node includes the number and sequence number of nodes passed through the direct link and the direct link score; the direct link of the node refers to the link between the current node and an adjacent node that can be connected through less than or equal to A1 transit nodes;

[0081] The transit node refers to a node on a direct link; the adjacent node refers to a node in the network whose distance to the node is less than a preset value; A1 is a preset positive integer greater than or equal to 2;

[0082] The direct link score may be obtained by averaging historical data of direct link scores.

[0083] S2, performing connection identification processing on the network topology information set to obtain a first link result information set and an updated network topology information set;

[0084] S3, performing a comparison and judgment operation on the first link result information set and the updated network topology information set to obtain a network unpredicted edge information set; the network unpredicted edge information set includes network unpredicted edge information;

[0085] S4, using the trained node connection prediction model, processing the network unpredicted edge information set to obtain link prediction results between network nodes; the link prediction results between network nodes are used to characterize whether the links between network nodes are connected.

[0086] The node importance value is calculated using a node importance calculation model;

[0087] The node importance calculation model has the following calculation expression:

[0088] v i =arctan(s i / l i ) / (pi / 2),

[0089] Among them, v i is the node importance value of the i-th node, s i is the sum of the node numbers directly connected to the i-th node, l i is the sum of the direct link scores of all direct links of the node;

[0090] All direct links of a node refer to all direct links passing through the node;

[0091] The performing connection identification processing on the network topology information set to obtain a first link result information set and an updated network topology information set includes:

[0092] S21, constructing a node representation vector based on the network topology information in the network topology information set; the elements of the node representation vector include the node number, the variance value of the node numbers directly connected to the node, the node importance value, the mean number of nodes of all direct links of the node, the median value of the node numbers of all direct links of the node, and the mean score of all direct links of the node;

[0093] The average value of the number of nodes of all direct links of the node is obtained by averaging the number of nodes passed by the direct links in all direct link information of the node;

[0094] The median value of the node sequence numbers of all direct links of the node is obtained by calculating the median of the sequence numbers of the nodes passed through the direct links in all direct link information of the node;

[0095] The average score of all direct links of the node is obtained by averaging the direct link scores in all direct link information of the node.

[0096] S22, performing cluster analysis on all node representation vectors to obtain type information and type-included node representation vector information;

[0097] The cluster analysis can be implemented using the Kmeans algorithm;

[0098] S23, sorting the type information from large to small according to the number of node representation vectors contained in the type to obtain a sorting sequence;

[0099] S24, extracting the first N1 type information in the sorted sequence;

[0100] N1 is a preset positive integer greater than or equal to 2;

[0101] S25, confirming that the direct link information of all nodes corresponding to the node representation vectors included in the first N1 types of information is the first link result information; using all the first link result information, construct a first link result information set;

[0102] S26, for each direct link information in the network topology information set, determine whether it appears in the first link result information set, and obtain a first determination result; if the first determination result is no, delete the direct link information from the network topology information set; if the first determination result is yes, do not perform any operation;

[0103] S27, completing the determination process of all direct link information in the network topology information set to obtain an updated network topology information set.

[0104] The comparing and judging operation of the first link result information set and the updated network topology information set to obtain the network unpredicted edge information set includes:

[0105] S31, initializing an unpredicted edge information set of the network; constructing an edge information set using all two different nodes of the network; the edge information set includes edge information; the edge information includes two nodes corresponding to the edge and their node information; the edge is a link connecting two different nodes, and whether the link is connected needs to be predicted;

[0106] S32, for each node in the edge information set, determine whether it appears in the nodes passed by the direct link in the direct link information in the first link result information set, and obtain a second determination result;

[0107] If the second judgment result is no, adding the side information to the network unpredicted side information set; if the second judgment result is yes, no operation is performed;

[0108] S33, after completing the judgment processing of each side information in the side information set, extracting the network topology information corresponding to the node included in each side information of the network unpredicted side information set;

[0109] S34, constructing an edge information vector by using the network topology information corresponding to the node included in each edge information; the edge information vector is obtained by concatenating the node information vectors of the two nodes corresponding to the edge; the node information vector includes the node number, the node number directly connected to the node, the node importance value, the number and number of nodes passed through on the direct link of the node, and the direct link score of the node;

[0110] S35, adding the side information vector of each edge to the corresponding side information in the network unpredicted side information set to obtain the network unpredicted side information; the network unpredicted side information includes the side information and the corresponding side information vector;

[0111] S36, using all the network unpredicted side information, construct a network unpredicted side information set.

[0112] The splicing is to connect two vectors into one vector.

[0113] The node connection prediction model includes: a first feature extraction sub-model, a second feature extraction sub-model and a fusion prediction sub-model;

[0114] The first feature extraction sub-model includes a first node feature extractor, a first encoder, a first decoder, a first feature reconstructor, a second node feature extractor, a second encoder, a second decoder, and a second feature reconstructor; the first encoder includes a first node encoder and a first link encoder; the second encoder includes a second node encoder and a second link encoder;

[0115] The output end of the first node feature extractor is connected to the input end of the first encoder; the output ends of the first node encoder and the second link encoder are connected to the input end of the first decoder; the output ends of the second node encoder and the first link encoder are connected to the input end of the second decoder; the output end of the first decoder is connected to the input end of the first feature reconstructor; the output end of the second decoder is connected to the input end of the second feature reconstructor;

[0116] The second feature extraction sub-model includes a third node feature extractor, a third encoder, a third decoder, a third feature reconstructor, a fourth node feature extractor, a fourth encoder, a fourth decoder, and a fourth feature reconstructor; the third encoder includes a third node encoder and a third link encoder; the fourth encoder includes a fourth node encoder and a fourth link encoder;

[0117] The output end of the third node feature extractor is connected to the input end of the third encoder; the output ends of the third node encoder and the fourth link encoder are connected to the input end of the third decoder; the output ends of the fourth node encoder and the third link encoder are connected to the input end of the fourth decoder; the output end of the third decoder is connected to the input end of the third feature reconstructor; the output end of the fourth decoder is connected to the input end of the fourth feature reconstructor;

[0118] The node feature extractor can be implemented by a multi-channel convolution module; the link feature extractor can be implemented by a feature extraction module in a ResNet-18 network.

[0119] The node encoder can be implemented by using an autoencoder neural network or an automatic encoder;

[0120] The link encoder may be implemented using a DAE network;

[0121] The decoder may be implemented using a self-decoding neural network or an automatic decoder;

[0122] The feature reconstructor can be implemented using a graph convolutional neural network.

[0123] Specifically, the node encoder can be implemented by using the LSTM autoencoder in Tensflow; the link encoder can be implemented by using a graph convolutional network (GCN), a graph attention network (GAT), etc.

[0124] The fusion prediction sub-model includes a first input module, a first convolution module, a second convolution module, a third convolution module, a first pooling module, a fourth convolution module and a first fully connected module;

[0125] The input end of the first input module of the fusion prediction sub-model is connected to the output ends of the first feature reconstructor, the second feature reconstructor, the third feature reconstructor, and the fourth feature reconstructor respectively; the output end of the first input module of the fusion prediction sub-model is connected to the input end of the first convolution module of the fusion prediction sub-model; the output end of the first convolution module of the fusion prediction sub-model is connected to the input end of the second convolution module of the fusion prediction sub-model; the output end of the second convolution module of the fusion prediction sub-model is connected to the input end of the third convolution module of the fusion prediction sub-model; the output end of the third convolution module of the fusion prediction sub-model is connected to the input end of the first pooling module of the fusion prediction sub-model; the output end of the first pooling module of the fusion prediction sub-model is connected to the input end of the fourth convolution module of the fusion prediction sub-model; the output end of the fourth convolution module of the fusion prediction sub-model is connected to the input end of the first fully connected module of the fusion prediction sub-model; the output end of the first fully connected module is used to output the link prediction result;

[0126] The output link prediction result is a probability value of link connectivity; when the probability value is greater than a preset threshold, the link between the network nodes is judged to be connected, otherwise, it is judged to be disconnected.

[0127] The first node feature extractor, the second node feature extractor, the third node feature extractor and the fourth node feature extractor are used to extract features from the received edge information vector to obtain a node feature signal;

[0128] The first node encoder, the second node encoder, the third node encoder, and the fourth node encoder are used to perform node encoding on the received node characteristic signal to obtain a node encoding signal;

[0129] The first link encoder, the second link encoder, the third link encoder, and the fourth link encoder are used to perform link encoding on the received node characteristic signal to obtain a link encoding signal;

[0130] The link encoding may be implemented by using a variational autoencoder; the node encoding may be implemented by using a contraction autoencoder;

[0131] The first decoder, the second decoder, the third decoder and the fourth decoder are used to decode the received encoded signal to obtain a decoded signal;

[0132] The first feature reconstructor, the second feature reconstructor, the third feature reconstructor and the fourth feature reconstructor are used to reconstruct the decoded signal to obtain a corresponding reconstructed node signal.

[0133] The training process of the node connection prediction model includes:

[0134] Initialize the number of training iterations;

[0135] Obtain network link data set;

[0136] The node information in the network link data set is used as input data and input into the node connection prediction model; the node link information in the network link data set is used as label information; the node link information is used to indicate whether two nodes can be linked;

[0137] Using the node connection prediction model, the input data is processed to obtain a link prediction result;

[0138] Performing difference calculation processing on the obtained link prediction result and label information to obtain a difference value;

[0139] Determine whether the difference value satisfies a convergence condition, and obtain a first determination result;

[0140] When the first judgment result is no, judging whether the training iteration number value is equal to the training number threshold, and obtaining a second judgment result;

[0141] When the second judgment result is no, determining that the model training state does not meet the training termination condition;

[0142] When the second judgment result is yes, determining that the model training state satisfies the training termination condition;

[0143] When the first judgment result is yes, determining that the model training state satisfies a training termination condition;

[0144] When the model training state does not meet the training termination condition, the parameter update model is used to update the parameters of the node connection prediction model, the training iteration number value is increased by 1, and the execution is triggered to input the node information in the network link data set as input data into the node connection prediction model;

[0145] When the model training state satisfies the training termination condition, the training process of the node connection prediction model is completed to obtain a trained node connection prediction model.

[0146] The difference value satisfies the convergence condition, which means that the difference value is less than a preset convergence threshold; the difference value does not satisfy the convergence condition, which means that the difference value is not less than the preset convergence threshold.

[0147] The network link dataset may be the USA airports (USAir) data of the directional flight network between airports or the Facebook data set of the social network;

[0148] The parameter updating model is:

[0149]

[0150] θ←θ+v;

[0151] In the formula, x (i) is the label information of the i-th node information in the network link data set, is the loss function, v is the parameter update value, θ is the parameter of the node connection prediction model, η is the initial parameter learning rate, α is the momentum angle parameter, 0≤α≤1, It means to find the partial derivative of variable θ, f(x (i) ; θ) represents the link prediction result calculated by the node connection prediction model for the i-th node information, f(·) is the calculation function corresponding to the node connection prediction model; exp represents the power operation of the constant e; η and α are preset values.

[0152] When the trained node connection prediction model is used to process the network unpredicted edge information set, the number of direct links of the node in the node information vector is set to 2 by default. When the number of direct links of the node is greater than 2, the two direct links with the largest scores are taken, thereby ensuring the consistency of the dimension of the vector input into the node connection prediction model.

[0153] In a second aspect of the embodiment of the present application, a link prediction device between network nodes is disclosed, which is used to implement the link prediction method between network nodes, including: an information acquisition module, a first processing module, a second processing module, and a node connection prediction model;

[0154] The first processing module is connected to the information acquisition module and the second processing module;

[0155] The node connection prediction model is connected to the second processing module;

[0156] The information acquisition module is used to acquire a set of network topology information;

[0157] The first processing module is used to perform connection identification processing on the network topology information set to obtain first link result information and an updated network topology information set;

[0158] The second processing module is used to extract a network unpredicted edge information set from the first link result information and the updated network topology information set;

[0159] The node connection prediction model processes the network unpredicted edge information set to obtain link prediction results between network nodes.

[0160] In a third aspect of the embodiments of the present application, a link prediction device between network nodes is disclosed, the device comprising:

[0161] A memory storing executable program code;

[0162] a processor coupled to the memory;

[0163] The processor calls the executable program code stored in the memory to execute the link prediction method between network nodes.

[0164] In a fourth aspect of an embodiment of the present application, a computer-storable medium is disclosed, wherein the computer-storable medium stores computer instructions, and when the computer instructions are called, they are used to execute the link prediction method between network nodes.

[0165] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for predicting links between network nodes, characterized in that: include: S1, obtaining a network topology information set; the network topology information set is used to characterize the connectivity of the network; the network includes a number of nodes and links between the nodes; The network topology information set includes network topology information; the network topology information includes node information and direct link information of the node; The node information includes the node number, the node number directly connected to the node, and the node importance value; The direct link information of the node includes the number and sequence number of nodes passed through the direct link and the direct link score; the direct link of the node refers to the link between the current node and an adjacent node that can be connected through less than or equal to A1 transit nodes; S2, performing connection identification processing on the network topology information set to obtain a first link result information set and an updated network topology information set; S3, performing a comparison and judgment operation on the first link result information set and the updated network topology information set to obtain a network unpredicted edge information set; the network unpredicted edge information set includes network unpredicted edge information; S4, using the trained node connection prediction model, processing the network unpredicted edge information set to obtain link prediction results between network nodes; the link prediction results between network nodes are used to indicate whether the links between network nodes are connected; The performing connection identification processing on the network topology information set to obtain a first link result information set and an updated network topology information set includes: S21, constructing a node representation vector based on the network topology information in the network topology information set; the elements of the node representation vector include the node number, the variance value of the node numbers directly connected to the node, the node importance value, the mean number of nodes of all direct links of the node, the median value of the node numbers of all direct links of the node, and the mean score of all direct links of the node; S22, performing cluster analysis on all node representation vectors to obtain type information and node representation vector information contained in the type; S23, sorting the type information from large to small according to the number of node representation vectors contained in the type to obtain a sorting sequence; S24, extracting the first N1 type information in the sorted sequence; S25, confirming that the direct link information of all nodes corresponding to the node representation vectors included in the first N1 types of information is the first link result information; using all the first link result information, construct a first link result information set; S26, for each direct link information in the network topology information set, determining whether it appears in the first link result information set, and obtaining a first determination result; If the first judgment result is yes, no operation is performed; If the first judgment result is no, deleting the direct link information from the network topology information set; S27, completing the determination process of all direct link information in the network topology information set to obtain an updated network topology information set; The training process of the node connection prediction model includes: Initialize the number of training iterations; Obtain network link data set; Using the node information in the network link data set as input data to input into the node connection prediction model; using the node link information in the network link data set as label information; Using the node connection prediction model, the input data is processed to obtain a link prediction result; Performing difference calculation processing on the obtained link prediction result and label information to obtain a difference value; Determine whether the difference value satisfies a convergence condition, and obtain a third determination result; When the third judgment result is no, judging whether the training iteration number value is equal to the training number threshold, and obtaining a fourth judgment result; When the fourth judgment result is no, determining that the model training state does not meet the training termination condition; When the fourth judgment result is yes, determining that the model training state satisfies the training termination condition; When the third judgment result is yes, determining that the model training state satisfies the training termination condition; When the model training state does not meet the training termination condition, the parameter update model is used to update the parameters of the node connection prediction model, the training iteration number value is increased by 1, and the execution is triggered to input the node information in the network link data set as input data into the node connection prediction model; When the model training state satisfies the training termination condition, the training process of the node connection prediction model is completed to obtain a trained node connection prediction model.

2. The link prediction method between network nodes according to claim 1, characterized in that: The node importance value is calculated using a node importance calculation model; The node importance calculation model has the following calculation expression: v i arctan(s) i / l i ) / (pi / 2), Among them, v i is the node importance value of the i-th node, s i is the sum of the node numbers directly connected to the i-th node, l i is the sum of the direct link scores of all direct links of the node.

3. The link prediction method between network nodes according to claim 1, characterized in that: The comparing and judging operation of the first link result information set and the updated network topology information set to obtain the network unpredicted edge information set includes: S31, initializing an unpredicted edge information set of the network; constructing an edge information set using all two-to-two different nodes of the network; the edge information set includes edge information; the edge information includes two nodes corresponding to the edge and their node information; S32, for each node in the edge information set, determine whether it appears in the nodes passed by the direct link in the direct link information in the first link result information set, and obtain a second determination result; If the second judgment result is no, adding the side information to the network unpredicted side information set; if the second judgment result is yes, no operation is performed; S33, after completing the judgment processing of each side information in the side information set, extracting the network topology information corresponding to the node included in each side information of the network unpredicted side information set; S34, constructing an edge information vector by using the network topology information corresponding to the node included in each edge information; the edge information vector is obtained by concatenating the node information vectors of the two nodes corresponding to the edge; the node information vector includes the node number, the node number directly connected to the node, the node importance value, the number and number of nodes passed through on the direct link of the node, and the direct link score of the node; S35, adding the side information vector of each edge to the corresponding side information in the network unpredicted side information set to obtain the network unpredicted side information; the network unpredicted side information includes the side information and the corresponding side information vector; S36, using all the network unpredicted side information, construct a network unpredicted side information set.

4. The link prediction method between network nodes according to claim 1, characterized in that: The node connection prediction model includes: a first feature extraction sub-model, a second feature extraction sub-model and a fusion prediction sub-model; The first feature extraction sub-model includes a first node feature extractor, a first encoder, a first decoder, a first feature reconstructor, a second node feature extractor, a second encoder, a second decoder, and a second feature reconstructor; the first encoder includes a first node encoder and a first link encoder; the second encoder includes a second node encoder and a second link encoder; The output end of the first node feature extractor is connected to the input end of the first encoder; the output ends of the first node encoder and the second link encoder are connected to the input end of the first decoder; the output ends of the second node encoder and the first link encoder are connected to the input end of the second decoder; the output end of the first decoder is connected to the input end of the first feature reconstructor; the output end of the second decoder is connected to the input end of the second feature reconstructor; The second feature extraction sub-model includes a third node feature extractor, a third encoder, a third decoder, a third feature reconstructor, a fourth node feature extractor, a fourth encoder, a fourth decoder, and a fourth feature reconstructor; the third encoder includes a third node encoder and a third link encoder; the fourth encoder includes a fourth node encoder and a fourth link encoder; The output end of the third node feature extractor is connected to the input end of the third encoder; the output ends of the third node encoder and the fourth link encoder are connected to the input end of the third decoder; the output ends of the fourth node encoder and the third link encoder are connected to the input end of the fourth decoder; the output end of the third decoder is connected to the input end of the third feature reconstructor; the output end of the fourth decoder is connected to the input end of the fourth feature reconstructor; The fusion prediction sub-model includes a first input module, a first convolution module, a second convolution module, a third convolution module, a first pooling module, a fourth convolution module and a first fully connected module; The input end of the first input module of the fusion prediction sub-model is connected to the output ends of the first feature reconstructor, the second feature reconstructor, the third feature reconstructor, and the fourth feature reconstructor respectively; the output end of the first input module of the fusion prediction sub-model is connected to the input end of the first convolution module of the fusion prediction sub-model; the output end of the first convolution module of the fusion prediction sub-model is connected to the input end of the second convolution module of the fusion prediction sub-model; the output end of the second convolution module of the fusion prediction sub-model is connected to the input end of the third convolution module of the fusion prediction sub-model; the output end of the third convolution module of the fusion prediction sub-model is connected to the input end of the first pooling module of the fusion prediction sub-model; the output end of the first pooling module of the fusion prediction sub-model is connected to the input end of the fourth convolution module of the fusion prediction sub-model; the output end of the fourth convolution module of the fusion prediction sub-model is connected to the input end of the first fully connected module of the fusion prediction sub-model; the output end of the first fully connected module is used to output the link prediction result.

5. The link prediction method between network nodes according to claim 1, characterized in that: The parameter updating model is: θ←θ+v; In the formula, x (i) is the label information of the i-th node information in the network link data set, is the loss function, v is the parameter update value of the node connection prediction model, θ is the parameter of the node connection prediction model, η is the initial parameter learning rate, α is the momentum angle parameter, 0≤α≤, It means to find the partial derivative of variable θ, f(x (i) ; θ) represents the link prediction result calculated by the node connection prediction model for the i-th node information, f(·) is the calculation function corresponding to the node connection prediction model; exp represents the power operation of the constant e; η and α are preset values.

6. A link prediction device between network nodes, used to implement the link prediction method between network nodes according to any one of claims 1 to 5, comprising: Information acquisition module, first processing module, second processing module, node connection prediction model; The first processing module is connected to the information acquisition module and the second processing module; The node connection prediction model is connected to the second processing module; The information acquisition module is used to acquire a set of network topology information; The first processing module is used to perform connection identification processing on the network topology information set to obtain a first link result information set and an updated network topology information set; The second processing module is used to compare and determine the first link result information set and the updated network topology information set to obtain a network unpredicted edge information set; The node connection prediction model is used to process the network unpredicted edge information set to obtain link prediction results between network nodes.

7. A link prediction device between network nodes, characterized in that: The device comprises: A memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the link prediction method between network nodes according to any one of claims 1 to 5.

8. A computer storable medium, characterized in that: The computer storable medium stores computer instructions, and when the computer instructions are called, they are used to execute the link prediction method between network nodes according to any one of claims 1 to 5.