CN109039721A - Node importance evaluation method based on error reconstruction - Google Patents

Node importance evaluation method based on error reconstruction Download PDF

Info

Publication number
CN109039721A
CN109039721A CN201810802825.3A CN201810802825A CN109039721A CN 109039721 A CN109039721 A CN 109039721A CN 201810802825 A CN201810802825 A CN 201810802825A CN 109039721 A CN109039721 A CN 109039721A
Authority
CN
China
Prior art keywords
node
error
network
conspicuousness
sparse
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201810802825.3A
Other languages
Chinese (zh)
Other versions
CN109039721B (en
Inventor
朱先强
郭园园
朱承
周鋆
黄金才
林福良
丁兆云
闫晶晶
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
National University of Defense Technology
Original Assignee
National University of Defense Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by National University of Defense Technology filed Critical National University of Defense Technology
Priority to CN201810802825.3A priority Critical patent/CN109039721B/en
Publication of CN109039721A publication Critical patent/CN109039721A/en
Application granted granted Critical
Publication of CN109039721B publication Critical patent/CN109039721B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0893—Assignment of logical groups to network elements
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/50—Network service management, e.g. ensuring proper service fulfilment according to agreements

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a node importance evaluation method based on error reconstruction, which specifically comprises the following steps: taking a sparse matrix of network connection as input, and calculating a network Node characteristic representation matrix X through a Node2Vec algorithm of network representation learning; constructing a multi-scale network; calculating two reconstruction error significances of network nodes according to the reconstruction error model for the networks under different scales; integrating two reconstruction error significances under different scales; and for the two reconstruction error significances, calculating the reconstruction error significance after fusion according to a weighted fusion algorithm and taking the reconstruction error significance as an index for finally measuring the node significance degree. The method has universality for networks with different structures and types, so that a relatively accurate conclusion can be obtained under the condition that the property of the network is not fully known in advance, and a large amount of analysis work on the network in advance in the traditional method and result errors caused by the wrong ranking mode are avoided. The invention is applied to the technical field of complex networks.

Description

Node importance appraisal procedure based on error reconstruct
Technical field
The present invention relates to complex network technical field more particularly to a kind of node importance assessment sides based on error reconstruct Method.
Background technique
Effectively identification key node is a basic problem in complex network, and is had a wide range of applications.Network key section Point refers to some special sections compared to the structure or function that can influence network for other nodes in network to a greater extent Point.For example, key node to network and critical link, which carry out redundancy backup, can increase the fault-tolerant survivability and effectively of network Improve the robustness of network.In addition, people can promote the propagation of information by optimizing using limited resource, identification is social Influential user and disseminator in network or community have the iterative algorithm of AP Rank a kind of in the prior art, with them it Between multiple association measure the influence of author and publication, to effectively distinguish popularity and pouplarity.
So far, it has been proposed that the method for a large amount of assessment key nodes such as spends centrality, and betweenness center approaches Spend centrality, eigenvector centrality etc..Wherein, degree centrality is a relatively straightforward index, but under normal conditions less Important or less correct, similar index includes Local Rank, the degree of approach and H index etc..Prior art degree of having studied, core Mathematical relationship between degree and three kinds of H index simple but important network center's property Measure Indexes.Centrality based on path, such as Degree of approach centrality and betweenness center are a kind of Measure Indexes of overall importance, can more effectively identify the crucial section in network Point, but their computation complexity is higher.Eigenvector centrality and PageRank are also widely used for the important of network node Property.Recently, the partial approach of Cluster Rank also has good performance in some cases.In addition, half local center purport Ignoring topological relation between neighbours and is only considering that the nearest-neighbors of node and the quantity of time neighbours in correlation and calculate multiple Weighed between miscellaneous degree, but the position of node in a network there may be prior effect compared to global properties such as degree.Kitsak Show that the position of node in a network is the key factor for influencing node conspicuousness Deng by Kshell value.In this measurement plan Under slightly, the node with larger Kshell value usually has wider array of extended capability, if but them is selected to pass as source node Sowing time, but propagating than degree has worse spread speed.In addition, other arrangement methods such as degree of approach, PageRank, Leader- Rank, ClusterRank etc. also have similar limitation.
Summary of the invention
The object of the present invention is to provide a kind of node importance appraisal procedures based on error reconstruct, realize from inessential section Point angle come reversely finds out important node so that this method the network of different structure and type is all had it is pervasive Property, use scope is greatly improved, so that in advance for the property of network without fully realizing, based on this Method can be derived that accurate conclusion, avoid in conventional method a large amount of analysis work that network is carried out in advance and Mistake has used result mistake caused by ranking mode.
The technical solution adopted by the present invention is that:
A kind of node importance appraisal procedure based on error reconstruct, specifically includes the following steps:
S1, using the sparse matrix of network connection as input, pass through network representation learn Node2Vec algorithm calculate net Network node diagnostic representing matrix X；
S2, multiple dimensioned network is constructed according to network node character representation matrix X；
S3, to the network under the different scale constructed in step S2, it is off line that each scale is calculated according to reconstructed error model Two kinds of reconstructed error conspicuousnesses of network node, two kinds of reconstructed error conspicuousnesses are that sparse reconstructed error conspicuousness and dense reconstruct miss Poor conspicuousness；
S4, two kinds of reconstructed error conspicuousnesses under different scale are integrated；
S5, fused reconstructed error conspicuousness is calculated simultaneously according to Weighted Fusion algorithm to two kinds of reconstructed error conspicuousnesses As the final index for measuring node conspicuousness degree.
As a further improvement of the above technical scheme, in step S1, the form of network node character representation matrix X are as follows:
X=[x1,x2,...,xN],X∈RD×N
In formula, D is intrinsic dimensionality, and N is the node number in network
As a further improvement of the above technical scheme, step S2 is specifically included:
S21, the scale for initializing network are N；
S22,0.95N, 0.9N, 0.85N, 0.8N are carried out to network node character representation matrix X by Kmeans clustering algorithm Four kinds of scales cluster, calculate each network node said module region；
Network node in S23, each module region of statistics, by network node mark sheet in said module region in network Character representation of the mean value shown as this module region；
Network characterization matrix under S24, construction different scale.
As a further improvement of the above technical scheme, step S3 is specifically included:
Unessential node difference structure is corresponding background module B in S31, each scale lower network node of extraction；
S32, the network under each scale is reconstructed by sparse reconstruct and two kinds of models of dense reconstruct, calculates each ruler Spend two kinds of reconstructed error conspicuousnesses of lower network；
S33, propagation reconstructed error conspicuousness for measuring propagation effect between adjacent node is calculated.
As a further improvement of the above technical scheme, in step S31, unessential node position in the network architecture In the node of network edge, network is decomposed using Kshell decomposition method, the node that selection is located at network edge is background Node simultaneously constitutes background module B.
As a further improvement of the above technical scheme, in step S32, two kinds of reconstructed error conspicuousnesses is calculated and are specifically wrapped It includes:
S321, the sparse reconstruction model of construction, seek sparse reconstruction coefficients α and sparse reconstructed error conspicuousness εs:
In formula, xiIt is the character representation of node i, B is the eigenmatrix that corresponding scale network context node is constituted, αiIt is section The sparse reconstruction coefficients of point i, λ are L1 regularization coefficients,It is the sparse reconstructed error conspicuousness of node i；
S322, the dense reconstruction model of construction, seek dense reconstruction coefficients β and dense reconstructed error conspicuousness εd:
In formula, xiIt is the character representation of node i,It is the characteristics of mean of all nodes, UB=[u1,u2,...,uD'], uiIt is I-th of principal component, D' are the principal component number extracted, the transposition of T representing matrix, βiIt is the dense reconstruction coefficients of node i,It is The dense reconstructed error conspicuousness of node i.
As a further improvement of the above technical scheme, it in step S33, calculates propagation reconstructed error conspicuousness and specifically wraps It includes:
S331, N number of node is clustered by K mean cluster algorithm；
S332, by its affiliated class with the similitude of remaining node construct likeness coefficient, to node i carry out error repair Just；
S333, estimation is weighted to the reconstructed error of node.
As a further improvement of the above technical scheme, in step S332:
The likeness coefficient is defined as:
In formula, { k1,k2,...,kNcIndicate the Nc node label in cluster block k,Refer to label j in cluster block k Node and node i similarity standard weight,It is the variance and x of each intrinsic dimensionality of xiIt is the mark sheet of node i Show, kjIt is the peripheral adjacent node of the i.e. node i of node of the label j in cluster block k,It is the node of label j in cluster block k Character representation, δ (kj- i) it is kjThe indicator function of-i；
Correct error conspicuousness are as follows:
In formula, τ is weight parameter,It is the amendment sparse error conspicuousness or the dense error conspicuousness of amendment of node i, Be the peripheral adjacent node of the i.e. node i of node of the label j in cluster block k sparse error conspicuousness or dense error it is significant Property, εiIt is the sparse error conspicuousness or dense error conspicuousness of node i.
As a further improvement of the above technical scheme, in step S4, the reconstructed error conspicuousness under different scale is integrated Expression formula are as follows:
In formula, z indicates that nodes, Ns indicate the scale number of multiscale analysis,Indicate that scale is under s and includes The amendment sparse error conspicuousness or the dense error conspicuousness of amendment of the module of node z,Indicate node z and mould where it The characteristic similarity of block, as the weight under current scale；
Expression formula are as follows:
In formula, fzIndicate the corresponding node diagnostic of node z,Indicate the mean value of node z said module interior joint feature, σs 2Be each intrinsic dimensionality of s variance and.
As a further improvement of the above technical scheme, in step S5, the calculation formula of the Weighted Fusion are as follows:
S (z)=α E1(z)+(1-α)E2(z)
In formula, E1(z) be node z sparse reconstructed error conspicuousness, E2(z) be node z dense reconstructed error it is significant Property, α is Weighted Fusion coefficient, α ∈ R, 0≤α≤1, and S (z) is to calculate fused reconstructed error conspicuousness as to measure node The index of conspicuousness degree.
Advantageous effects of the invention:
The present invention is based on the node importance appraisal procedures of error reconstruct, pass through the spy for the network that Node2Vec algorithm generates Representing matrix is levied, and constructs multiple dimensioned network, solves the dense and sparse reconstructed error conspicuousness under the network under different scale Value, and sparse reconstruct and two kinds of models of dense reconstruct are merged by Weighted Fusion technology, obtained fusion significant result The index for as differentiating node conspicuousness realizes that the angle of never important node reversely finds out important node, so that this Method all has universality for the network of different structure and type, greatly improves use scope so that in advance for The property of network can be derived that accurate conclusion based on this method, avoid tradition without in the case where fully realizing In method in advance to network carry out a large amount of analysis work and mistake used result mistake caused by ranking mode.
Detailed description of the invention
Fig. 1 is the flow diagram of the embodiment of the present invention；
Fig. 2 is the collaborative network CA-GrQc schematic diagram in embodiment one；
Fig. 3 is that 200 nodes are average to 50 times of four kinds of control methods and the present embodiment method before the ranking in embodiment two SIR propagates schematic diagram.
Specific embodiment
For the ease of implementation of the invention, it is further described below with reference to specific example.
A kind of node importance appraisal procedure based on error reconstruct as shown in Figure 1, specifically includes the following steps:
S1, using the sparse matrix of network connection as input, pass through network representation learn Node2Vec algorithm calculate net Network node diagnostic representing matrix X
Node2vec algorithm on the basis of on-line study (the Deep Walk) of social presence, define one have it is inclined Random walk strategy formation sequence, so that in breadth first search (Breadth First Search) and depth-first search Certain balance is taken between (Depth First Search), and uses the neural network model side Skip-Gram of propagated forward Method trains node diagnostic, and the feature for extracting nodes, which is converted to, optimizes " possibility " objective function problem, mesh Scalar functions are as follows:
In formula, Zu=∑v∈Vexp(f(v)·f(u)).f:V→RD, f is node collection V to D dimension real number space RD×NReflect It penetrates, u is a certain node in network, and Ns (u) is the neighbor node set of node u, f (ni) it is node niIn D dimension real number space Character representation, f (u) are character representation of the node u in D dimension real number space, and f (v) is that the neighbor node v of node u is empty in D dimension real number Between character representation.
By the form for the optimal network node diagnostic representing matrix X that objective function obtains are as follows:
X=[x1,x2,...,xN],X∈RD×N
In formula, D is intrinsic dimensionality, and N is the node number in network.
S2, multiple dimensioned network is constructed according to network node character representation matrix X
Include:
S21, the scale for initializing network are N；
S22,0.95N, 0.9N, 0.85N, 0.8N are carried out to network node character representation matrix X by Kmeans clustering algorithm Four kinds of scales cluster, calculate each network node said module region；
Network node in S23, each module region of statistics, by network node mark sheet in said module region in network Character representation of the mean value shown as this module region；
Network characterization matrix under S24, construction different scale.
S3, to the network under the different scale constructed in step S2, it is off line that each scale is calculated according to reconstructed error model Two kinds of reconstructed error conspicuousnesses of network node, two kinds of reconstructed error conspicuousnesses are that sparse reconstructed error conspicuousness and dense reconstruct miss Poor conspicuousness
Include:
Unessential node difference structure is corresponding background module B in S31, each scale lower network node of extraction:
The unessential node is located at the node of network edge in the network architecture, using Kshell decomposition method to network It is decomposed, choose the node for being located at network edge as background node and constitutes background module B:
B=[b1,b2,...,bM],B∈RD×M
In formula, M indicates the background number of nodes chosen.
S32, the network under each scale is reconstructed by sparse reconstruct and two kinds of models of dense reconstruct, calculates each ruler Two kinds of reconstructed error conspicuousnesses for spending lower network, specifically include:
S321, the sparse reconstruction model of construction, seek sparse reconstruction coefficients α and sparse reconstructed error conspicuousness εs, by background Node does multiple linear regression as one group of base vector, to the node in network, to reduce over-fitting and quickly calculating back Return coefficient, (the L1 regularization of linear regression) is returned using currently used Lasso, i.e., when carrying out linear regression to high dimensional data Addition L1 regularization (λ | | αi||1).And its regression coefficient is calculated using Lars algorithm (minimum angle regression algorithm), as dilute Dredge reconstruction coefficients α:
In formula, xiIt is the character representation of node i, B is the eigenmatrix that corresponding scale network context node is constituted, αiIt is section The sparse reconstruction coefficients of point i,It is the sparse reconstructed error conspicuousness of node i, λ is L1 regularization coefficient, and λ is arranged in test It is 0.01；
S322, the dense reconstruction model of construction, dense reconstruction model is the background section using principal component analysis (PCA) to selection Point B extracts its principal component UB, then dense reconstruction coefficients β is constructed by the residual error of node diagnostic and mean value and seeks dense reconstruct and is missed Poor conspicuousness εd:
In formula, xiIt is the character representation of node i,It is the characteristics of mean of all nodes, UB=[u1,u2,...,uD'], uiIt is I-th of principal component, D' are the principal component number extracted, the transposition of T representing matrix, βiIt is the dense reconstruction coefficients of node i,It is The dense reconstructed error conspicuousness of node i.
S33, consideration are calculated by the propagation effect of adjacent node and propagate reconstructed error conspicuousness, specifically included:
S331, N number of node is clustered by K mean cluster algorithm；
S332, by its affiliated class with the similitude of remaining node construct likeness coefficient, to node i carry out error repair Just, in which:
The likeness coefficient is defined as:
In formula, { k1,k2,...,kNcIndicate the Nc node label in cluster block k,Refer to label j in cluster block k Node and node i similarity standard weight,Be each intrinsic dimensionality of x variance and, δ () is indicator function, xiIt is the character representation of node i, kjIt is the peripheral adjacent node of the i.e. node i of node of the label j in cluster block k,It is cluster The character representation of the node of label j in block k, δ (kj- i) it is kjThe indicator function of-i；
Correct error conspicuousness are as follows:
In formula, τ is weight parameter,It is the amendment sparse error conspicuousness or the dense error conspicuousness of amendment of node i, Be the peripheral adjacent node of the i.e. node i of node of the label j in cluster block k sparse error conspicuousness or dense error it is significant Property, εiIt is the sparse error conspicuousness or dense error conspicuousness of node i, sparse error conspicuousness is substituted into above formula i.e. acquisition and is repaired Dense error conspicuousness is substituted into above formula and obtains the dense error conspicuousness of amendment by positive sparse error conspicuousness.
S333, estimation is weighted to the reconstructed error of node.
S4, two kinds of reconstructed error conspicuousnesses under different scale are integrated
Integrate the expression formula of the reconstructed error conspicuousness under different scale are as follows:
In formula, z indicates that nodes, Ns indicate the scale number of multiscale analysis,Indicate that scale is under s and includes The amendment sparse error conspicuousness or the dense error conspicuousness of amendment of node z,Indicate the spy of node z and module where it Similarity is levied, as the weight under current scale；
Expression formula are as follows:
In formula, fzIndicate the corresponding node diagnostic of node z,Indicate the mean value of node z said module interior joint feature, σs 2Be each intrinsic dimensionality of s variance and.
S5, fused reconstructed error conspicuousness is calculated simultaneously according to Weighted Fusion algorithm to two kinds of reconstructed error conspicuousnesses As the final index for measuring node conspicuousness degree.
The calculation formula of the Weighted Fusion are as follows:
S (z)=α E1(z)+(1-α)E2(z)
In formula, E1(z) be node z sparse reconstructed error conspicuousness, E2(z) be node z dense reconstructed error it is significant Property, α is Weighted Fusion coefficient, α ∈ R, 0≤α≤1, α=0.5 in the present embodiment, and S (z) is to calculate fused reconstructed error to show Work property is to measure the index of node conspicuousness degree.
Embodiment one
Public data is used to integrate collaborative network CA-GrQc as embodiment, as shown in Fig. 2, it is collaborative network CA-GrQc's Schematic network structure observes data set it is found that the indirected net that the collaborative network is made of 5242 nodes and 28980 sides Network, wherein node table shows author, and side indicates that two authors at least cooperate an article, and node ID is up to 26196,5242 sections The non-ordered arrangement of point, possible cause are part of nodes there is no even side, as isolated point, and expression does not cooperate with other people.
Embodiment two
In order to verify the technical effect of the present embodiment method, using four kinds of degree, the degree of approach, betweenness, feature vector sequence sides Method is as a comparison.200 nodes are averaged to 50 times of four kinds of control methods and the present embodiment method before Fig. 3 illustrates selection ranking SIR propagates schematic diagram, as shown in figure 3, the present embodiment method is compared with feature vector ranking method and degree of approach ranking method by higher biography Speed is broadcast, though spread speed is not as good as degree and betweenness ranking method, when reaching propagation stable state, the present embodiment method in communication process Propagation number of nodes have bigger propagation scale compared to four kinds of control methods.
As shown above, eigenvector centrality, degree centrality, four kinds of betweenness center, K- nuclear decomposition method control methods Reaching number of nodes when propagating stable state is 4158, and it is 4299 that the present invention, which reaches and propagates the number of nodes of stable state, compared with four kinds of control methods It can be propagated more on a large scale；Though the propagation times when present invention reaches stable state are higher than spent and two methods of betweenness, Relatively it is lower than the two kinds of control methods of feature vector and the degree of approach；To spread speed, spread speed of the invention be effectively higher than four kinds it is right Ratio method.In summary, the present invention is based on the node conspicuousness methods of error reconstruct to have certain propagation advantage.The present invention can It is applied to complex network field as a kind of new node importance evaluation index, finds phase from the unessential node side in part To the node of core key, its recognition accuracy is effectively improved and being capable of the more rapidly broadly node in communication network.
Contain the explanation of the preferred embodiment of the present invention above, this be for the technical characteristic that the present invention will be described in detail, and Be not intended to for summary of the invention being limited in concrete form described in embodiment, according to the present invention content purport carry out other Modifications and variations are also protected by this patent.The purport of the content of present invention is to be defined by the claims, rather than by embodiment Specific descriptions are defined.

Claims (10)

1. a kind of node importance appraisal procedure based on error reconstruct, which is characterized in that specifically includes the following steps:
S1, using the sparse matrix of network connection as input, pass through network representation learn Node2Vec algorithm calculate network section Point feature representing matrix X；
S2, multiple dimensioned network is constructed according to network node character representation matrix X；
S3, to the network under the different scale constructed in step S2, each scale lower network section is calculated according to reconstructed error model Two kinds of reconstructed error conspicuousnesses of point, two kinds of reconstructed error conspicuousnesses are that sparse reconstructed error conspicuousness and dense reconstructed error are aobvious Work property；
S4, two kinds of reconstructed error conspicuousnesses under different scale are integrated；
S5, fused reconstructed error conspicuousness and conduct are calculated according to Weighted Fusion algorithm to two kinds of reconstructed error conspicuousnesses The final index for measuring node conspicuousness degree.
2. the node importance appraisal procedure according to claim 1 based on error reconstruct, which is characterized in that in step S1, The form of network node character representation matrix X are as follows:
X=[x1,x2,...,xN],X∈RD×N
In formula, D is intrinsic dimensionality, and N is the node number in network.
3. the node importance appraisal procedure according to claim 1 based on error reconstruct, which is characterized in that step S2 is specific Include:
S21, the scale for initializing network are N；
S22, the four of 0.95N, 0.9N, 0.85N, 0.8N is carried out to network node character representation matrix X by Kmeans clustering algorithm The cluster of kind scale, calculates each network node said module region；
Network node in S23, each module region of statistics, by network node character representation in said module region in network Character representation of the mean value as this module region；
Network characterization matrix under S24, construction different scale.
4. the node importance appraisal procedure according to claim 1 based on error reconstruct, which is characterized in that step S3 is specific Include:
Unessential node difference structure is corresponding background module B in S31, each scale lower network node of extraction；
S32, the network under each scale is reconstructed by sparse reconstruct and two kinds of models of dense reconstruct, is calculated under each scale Two kinds of reconstructed error conspicuousnesses of network；
S33, propagation reconstructed error conspicuousness for measuring propagation effect between adjacent node is calculated.
5. the node importance appraisal procedure according to claim 4 based on error reconstruct, which is characterized in that in step S31, The unessential node is located at the node of network edge in the network architecture, is divided using Kshell decomposition method network Solution chooses the node for being located at network edge as background node and constitutes background module B.
6. the node importance appraisal procedure according to claim 4 based on error reconstruct, which is characterized in that in step S32, Two kinds of reconstructed error conspicuousnesses are calculated to specifically include:
S321, the sparse reconstruction model of construction, seek sparse reconstruction coefficients α and sparse reconstructed error conspicuousness εs:
In formula, xiIt is the character representation of node i, B is the eigenmatrix that corresponding scale network context node is constituted, αiIt is node i Sparse reconstruction coefficients, λ are L1 regularization coefficients,It is the sparse reconstructed error conspicuousness of node i；
S322, the dense reconstruction model of construction, seek dense reconstruction coefficients β and dense reconstructed error conspicuousness εd:
In formula, xiIt is the character representation of node i,It is the characteristics of mean of all nodes, UB=[u1,u2,...,uD'], uiIt is i-th A principal component, D' are the principal component number extracted, the transposition of T representing matrix, βiIt is the dense reconstruction coefficients of node i,It is section The dense reconstructed error conspicuousness of point i.
7. the node importance appraisal procedure according to claim 4 based on error reconstruct, which is characterized in that in step S33, Propagation reconstructed error conspicuousness is calculated to specifically include:
S331, N number of node is clustered by K mean cluster algorithm；
S332, by its affiliated class with the similitude of remaining node construct likeness coefficient, to node i carry out error correction；
S333, estimation is weighted to the reconstructed error of node.
8. the node importance appraisal procedure according to claim 7 based on error reconstruct, which is characterized in that step S332 In:
The likeness coefficient is defined as:
In formula, { k1,k2,...,kNcIndicate the Nc node label in cluster block k,Refer to the section of label j in cluster block k The similarity standard weight of point and node i,Be each intrinsic dimensionality of x variance and, δ () is indicator function, xiIt is The character representation of node i, kjIt is the peripheral adjacent node of the i.e. node i of node of the label j in cluster block k,It is in cluster block k The character representation of the node of label j, δ (kj- i) it is kjThe indicator function of-i；
Correct error conspicuousness are as follows:
In formula, τ is weight parameter,It is the amendment sparse error conspicuousness or the dense error conspicuousness of amendment of node i,It is poly- The sparse error conspicuousness or dense error conspicuousness of node, that is, node i peripheral adjacent node of label j in class block k, εi It is the sparse error conspicuousness or dense error conspicuousness of node i.
9. the node importance appraisal procedure according to claim 8 based on error reconstruct, which is characterized in that in step S4, Integrate the expression formula of the reconstructed error conspicuousness under different scale are as follows:
In formula, z indicates that nodes, Ns indicate the scale number of multiscale analysis,Indicating that scale is includes node under s The amendment sparse error conspicuousness or the dense error conspicuousness of amendment of the module of z,Indicate node z and module where it Characteristic similarity, as the weight under current scale；
Expression formula are as follows:
In formula, fzIndicate the corresponding node diagnostic of node z,Indicate the mean value of node z said module interior joint feature, σs 2It is The variance of each intrinsic dimensionality of s and.
10. the node importance appraisal procedure according to claim 9 based on error reconstruct, which is characterized in that in step S5, The calculation formula of the Weighted Fusion are as follows:
S (z)=α E1(z)+(1-α)E2(z)
In formula, E1(z) be node z sparse reconstructed error conspicuousness, E2(z) be node z dense reconstructed error conspicuousness, α is Weighted Fusion coefficient, α ∈ R, 0≤α≤1, S (z) are to calculate fused reconstructed error conspicuousness as to measure node conspicuousness The index of degree.
CN201810802825.3A 2018-07-20 2018-07-20 Node importance evaluation method based on error reconstruction Active CN109039721B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810802825.3A CN109039721B (en) 2018-07-20 2018-07-20 Node importance evaluation method based on error reconstruction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810802825.3A CN109039721B (en) 2018-07-20 2018-07-20 Node importance evaluation method based on error reconstruction

Publications (2)

Publication Number Publication Date
CN109039721A true CN109039721A (en) 2018-12-18
CN109039721B CN109039721B (en) 2021-06-18

Family

ID=64643807

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810802825.3A Active CN109039721B (en) 2018-07-20 2018-07-20 Node importance evaluation method based on error reconstruction

Country Status (1)

Country Link
CN (1) CN109039721B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109495328A (en) * 2018-12-30 2019-03-19 李保安 A kind of method of Logistics networks communication reliability
CN110826164A (en) * 2019-11-06 2020-02-21 中国人民解放军国防科技大学 Complex network node importance evaluation method based on local and global connectivity
CN112104515A (en) * 2020-11-19 2020-12-18 中国人民解放军国防科技大学 Network reconstruction method and device, computer equipment and storage medium
CN116436799A (en) * 2023-06-13 2023-07-14 中国人民解放军国防科技大学 Complex network node importance assessment method, device, equipment and storage medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101895957A (en) * 2010-07-23 2010-11-24 浙江大学 Random routing method facing matrix type wireless sensor network distribution weighing
CN102547903A (en) * 2012-01-18 2012-07-04 北京航空航天大学 Network backbone node selecting method for wireless sensor based on compressed sensing
CN106780053A (en) * 2015-11-20 2017-05-31 香港中文大学深圳研究院 A kind of overlap community discovery method and system based on node connection preference
CN106788885A (en) * 2016-12-07 2017-05-31 上海交通大学 A kind of compress ecg data rate Automatic adjusument wireless transmitting system and transmission method
CN106934398A (en) * 2017-03-09 2017-07-07 西安电子科技大学 Image de-noising method based on super-pixel cluster and rarefaction representation
CN107967440A (en) * 2017-09-19 2018-04-27 北京工业大学 A kind of monitor video method for detecting abnormality based on multizone mutative scale 3D-HOF

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101895957A (en) * 2010-07-23 2010-11-24 浙江大学 Random routing method facing matrix type wireless sensor network distribution weighing
CN102547903A (en) * 2012-01-18 2012-07-04 北京航空航天大学 Network backbone node selecting method for wireless sensor based on compressed sensing
CN106780053A (en) * 2015-11-20 2017-05-31 香港中文大学深圳研究院 A kind of overlap community discovery method and system based on node connection preference
CN106788885A (en) * 2016-12-07 2017-05-31 上海交通大学 A kind of compress ecg data rate Automatic adjusument wireless transmitting system and transmission method
CN106934398A (en) * 2017-03-09 2017-07-07 西安电子科技大学 Image de-noising method based on super-pixel cluster and rarefaction representation
CN107967440A (en) * 2017-09-19 2018-04-27 北京工业大学 A kind of monitor video method for detecting abnormality based on multizone mutative scale 3D-HOF

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
范明喆: "基于两重误差重构的显著性区域检测算法", 《红外技术》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109495328A (en) * 2018-12-30 2019-03-19 李保安 A kind of method of Logistics networks communication reliability
CN110826164A (en) * 2019-11-06 2020-02-21 中国人民解放军国防科技大学 Complex network node importance evaluation method based on local and global connectivity
CN110826164B (en) * 2019-11-06 2023-07-04 中国人民解放军国防科技大学 Complex network node importance evaluation method based on local and global connectivity
CN112104515A (en) * 2020-11-19 2020-12-18 中国人民解放军国防科技大学 Network reconstruction method and device, computer equipment and storage medium
CN116436799A (en) * 2023-06-13 2023-07-14 中国人民解放军国防科技大学 Complex network node importance assessment method, device, equipment and storage medium
CN116436799B (en) * 2023-06-13 2023-08-11 中国人民解放军国防科技大学 Complex network node importance assessment method, device, equipment and storage medium

Also Published As

Publication number Publication date
CN109039721B (en) 2021-06-18

Similar Documents

Publication Publication Date Title
Ou et al. Asymmetric transitivity preserving graph embedding
CN109039721A (en) Node importance evaluation method based on error reconstruction
CN110532436B (en) Cross-social network user identity recognition method based on community structure
CN112256981B (en) Rumor detection method based on linear and nonlinear propagation
CN113486190B (en) Multi-mode knowledge representation method integrating entity image information and entity category information
CN111815468B (en) Multi-source social network construction method based on user identity association
CN114004871B (en) Point cloud registration method and system based on point cloud completion
CN106971205A (en) A kind of embedded dynamic feature selection method based on k nearest neighbor Mutual Information Estimation
CN111966793B (en) Intelligent question-answering method and system based on knowledge graph and knowledge graph updating system
CN109857457B (en) Function level embedding representation method in source code learning in hyperbolic space
CN113255895A (en) Graph neural network representation learning-based structure graph alignment method and multi-graph joint data mining method
CN107577742A (en) A kind of more relation social network patterns method for digging based on bayes method
CN107943897A (en) A kind of user recommends method
CN105825430A (en) Heterogeneous social network-based detection method
CN113792110A (en) Equipment trust value evaluation method based on social networking services
CN114970736A (en) Network node depth anomaly detection method based on density estimation
CN110458466B (en) Patent estimation method and system based on data mining and heterogeneous knowledge association
CN110716998B (en) Fine scale population data spatialization method
CN113065321B (en) User behavior prediction method and system based on LSTM model and hypergraph
CN110084423A (en) A kind of link prediction method based on local similarity
CN114385953A (en) Link prediction method and system based on heterogeneous motif characteristics and storable medium
CN114943019A (en) Top k non-overlapping diversified community discovery method based on double-layer weight network random walk
CN111985086A (en) Community detection method integrating prior information and sparse constraint
Du et al. [Retracted] Application of Improved Interactive Multimodel Algorithm in Player Trajectory Feature Matching
CN111882124A (en) Homogeneous platform development effect prediction method based on generation confrontation simulation learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant