CN110310209A - A kind of group's distribution detection method based on Chebyshev filter - Google Patents
A kind of group's distribution detection method based on Chebyshev filter Download PDFInfo
- Publication number
- CN110310209A CN110310209A CN201810455982.1A CN201810455982A CN110310209A CN 110310209 A CN110310209 A CN 110310209A CN 201810455982 A CN201810455982 A CN 201810455982A CN 110310209 A CN110310209 A CN 110310209A
- Authority
- CN
- China
- Prior art keywords
- node
- group
- label
- graph model
- chebyshev filter
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 24
- 238000000034 method Methods 0.000 claims abstract description 15
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 238000013507 mapping Methods 0.000 claims abstract description 10
- 230000013016 learning Effects 0.000 claims abstract description 6
- 239000000284 extract Substances 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 25
- 230000008569 process Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 2
- 208000031968 Cadaver Diseases 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000004850 protein–protein interaction Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000000547 structure data Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Abstract
The present invention provides a kind of group's distribution detection method based on Chebyshev filter, mainly comprises the steps that and reads complex network data, complex network is abstracted as graph model；Label initialization, as one unique label of part of nodes label, for group where indicating the node, other nodes are not marked；The node of initialization is input to Chebyshev filter, extracts graph model hidden layer feature；The Feature Mapping of extraction is handled to flag node space, and output valve is input in classifier, the node with same label is same group.The present invention realizes the detection of group's distribution using the method for Chebyshev filter semi-supervised learning, and less parameter has been used to find the group structure of network, and improves the efficiency of complex network group distribution detection.
Description
Technical field
The present invention relates to deep learnings and complex network community to find field, is filtered more particularly to one kind based on Chebyshev
The group of wave device is distributed detection method.
Background technique
In recent years, deep learning is galore applied in fields such as speech recognition, image recognition, natural language processings, and right
In complex network such as social networks, telecommunication network, protein-protein interaction network and road network etc. application almost without.
Nearly 2 years picture scrolls product nerual network technique has new breakthrough, which can handle a large amount of network data, can be to one
Network carries out convolution operation, and does coarse, to realize that multilayer signal is handled.
The Combo discovering method of traditional complex network needs to be arranged compared with multi-parameter, and computational efficiency is low.This method uses picture scroll
Product Processing with Neural Network complex network, the detection of group's distribution is realized with less parameter, to improve group's distribution detection
Efficiency.
Summary of the invention
Technical problem to be solved by the invention is to provide a kind of, and the group based on Chebyshev filter is distributed detection
Method realizes that group is detected using less parameter, to improve the efficiency of group's detection.
The technical scheme adopted by the invention is that the group based on Chebyshev filter is distributed detection method, comprising:
Step 1: reading complex network data, complex network is abstracted as graph model, complex network data refer to all have
Self-organizing, self similarity, attractor, worldlet, the part or all of network of property in uncalibrated visual servo.Such as: human relation network, society
Handing over relational network, correspondence network is all complex network；
Step 2: the node in graph model is subjected to label initialization, as one unique label of part of nodes label,
For indicating the group where the node, other nodes are not marked, and graph model is made to can be used for semi-supervised learning；
Step 3: setting Chebyshev filter is configured with the selection that Chebyshev polynomials parameterize；
Step 4: the node of initialization being input to Chebyshev filter, coarse and pondization operation is carried out, extracts figure
The hidden layer feature of model；
Step 5: the Feature Mapping of extraction being handled to flag node space, and output valve is input to classifier
In, the node with same label is same group.
The beneficial effects of the present invention are: complex network is input to Chebyshev filter with the structure of graph model, use
Semi-supervised learning method realizes the associated data processing to graph structure, coarse and pond by Chebyshev filter
Operation effectively extracts the hidden layer feature of network, and by handling the Feature Mapping of extraction to flag node space, real
Existing network weight is shared, and the detection of complex network group is realized using less parameter, improves the efficiency of group's detection.
Based on the above technical solution, the present invention can also be improved as follows.
It further, include node and side, adjacent weight matrix, specific symbol in the graph model in the step 1
It indicates are as follows: G=(V, E, W), wherein V is the set of node, and E represents the set on node connection side, for each node vi∈ V,
Side (vi,vj) ∈ E, W ∈ RN×NFor the adjoining weight matrix of two nodes.
Using the beneficial effect of above-mentioned further scheme is: complex network is indicated with the graph structure of node, side and weight
The relationship of data indicates the incidence relation of data in the matrix form, to reduce the setting of parameter.
Further, in the step 3, Chebyshev polynomials parameterize specifically being expressed as follows for selection:
Wherein, x is a signal, and θ is parameter, TkIt is Chebyshev polynomials, formula are as follows:
Tk(x)=2xTk-1(x)-Tk-2(x) (2)
Wherein, T0(x)=1, T1(x)=x.
Beneficial effect using above-mentioned further scheme is: being parameterized and is selected by Chebyshev polynomials, makes filter
Parameter it is optimal；
Further, realize that specific step is as follows for step 4:
Step 4-1: the node of input initialization to Chebyshev filter, using ReLU activation primitive, then the biography of node
It is as follows to broadcast rule:
Wherein,It is the adjacency matrix of graph model G, INIt is eigenmatrix,Degree of a representation matrix,
W(l)It is the 1st layer of weight matrix, ReLU () is activation primitive, H(l)∈RN×DIt is the activated matrix at the 1st layer, and H(0)=X, X
For the feature of node collection V；
Step 4-2: coarse operation is carried out to the node in graph model, by unlabelled node i no mark adjacent thereto
It signs node j and carries out local normalization, then two adjacent nodes are marked, using the weight of coarse posterior nodal point as this
The sum of the weight of two nodes repeats this process until all nodes are all by coarse；
Step 4-3: pondization operation is carried out to the node after coarse, rearranges node, similar node is then assembled
Together；
Step 4-4: the hidden layer feature of graph model structure is extracted.
Beneficial effect using above-mentioned further scheme is: using Chebyshev filter to graph structure data processing, leading to
It crosses to operate the coarse and pondization of node and comes out the hidden layer feature extraction of data, to improve computational efficiency.
Further, realize that specific step is as follows for step 5:
Step 5-1: the Feature Mapping of extraction is handled to flag node space；
Step 5-2: the result of processing is input to classifier, model is as follows:
Wherein, W(0)∈RN×NIt is the weight matrix of input layer, W(1)∈RN×NIt is the weight matrix of output layer, softmax swashs
Function living is defined as follows:
Wherein, Z=∑iexp(xi)。
Step 5-3: output label node, the node with same label are the same group.
Beneficial effect using the above scheme is: inputting after the Feature Mapping of extraction is handled to flag node space
To classifier, share weight by connection, to reduce the setting of parameter, the efficiency that diplomatic corps's physical examination is surveyed is improved.
Detailed description of the invention
Fig. 1 is that the present invention is based on the groups of Chebyshev filter to be distributed detection method implementation flow chart；
Fig. 2 is coarse and the pond operating process of Chebyshev filter of the present invention.
Specific embodiment
Group's detection method the present invention is based on Chebyshev filter is further described with reference to the accompanying drawing.
As shown in Figure 1, the present invention provides a kind of group's distribution detection method based on Chebyshev filter.
The complicated community network formed for college football league matches, 115 university student delegations of competition
It is divided into 12 alliances, has carried out 616 matches altogether, included following with the detailed process that method of the invention carries out group's detection
Step:
Step 1: read complex network data, complex network be abstracted as graph model, include in the graph model node and
Side, adjacent weight matrix, specific symbol are expressed as: G=(V, E, W), wherein V is the set of node, indicates team, and E is represented
Node connects the set on side, indicates that Liang Zhi team carried out match, for each node vi∈ V, side (vi, vj) ∈ E, W ∈ RN×N
For the adjoining weight matrix of two nodes, the number that Liang Zhi team plays is indicated；
Step 2: the node in graph model being subjected to label initialization, is one unique label of part of nodes label, uses
Group where indicating the node, other nodes do not mark, and graph model is made to can be used for semi-supervised learning, i.e. selector bulb separation
Badge remembers alliance's label where them, and remaining team does not mark；The purpose of part label is that graph model is made to can be used for half prison
Educational inspector practises, and unlabelled node can generate one's own mark by deep learning according to the feature and relationship of flag node
Label.
Step 3: setting Chebyshev filter is configured, Qie Bixue with the selection that Chebyshev polynomials parameterize
The selection of husband's polynomial parametersization is specifically expressed as follows:
Wherein, x is a signal, and θ is parameter, TkIt is Chebyshev polynomials, formula are as follows:
Tk(x)=2xTk-1(x)-Tk-2(x) (2)
Wherein, T0(x)=1, T1(x)=x.
Step 4: the node of initialization being input to Chebyshev filter, coarse and pondization operation is carried out, extracts figure
The hidden layer feature of model, detailed process is as follows:
Step 4-1: the node of input initialization to Chebyshev filter, using ReLU activation primitive, then the biography of node
It is as follows to broadcast rule:
Wherein,It is the adjacency matrix of graph model G, INIt is eigenmatrix,Degree of a representation matrix,
W(l)It is l layers of weight matrix, ReLU () is activation primitive, H(l)∈RN×DIt is the activated matrix at l layers, and H(0)=X, X
For the feature of node collection V；
Step 4-2: coarse operation is carried out to the node in graph model, by unlabelled node i no mark adjacent thereto
It signs node j and carries out local normalization, then two adjacent nodes are marked, using the weight of coarse posterior nodal point as this
The sum of the weight of two nodes repeats this process until all nodes are all by coarse；
Step 4-3: pondization operation is carried out to the node after coarse, rearranges node, similar node is then assembled
Together；
Step 4-4: the hidden layer feature of graph model structure is extracted.
Step 5: the Feature Mapping of extraction being handled to flag node space, and output valve is input to classifier
In, the node with same label is same group, and the as same alliance, detailed process is as follows:
Step 5-1: the Feature Mapping of extraction is handled to flag node space；
Step 5-2: the result of processing is input to classifier, model is as follows:
Wherein, W(0)∈RN×NIt is the weight matrix of input layer, W(1)∈RN×NIt is the weight matrix of output layer, softmax swashs
Function living is defined as follows:
Wherein, Z=∑iexp(xi)。
Step 5-3: output label node, the node with same label are the same group, i.e., the same alliance.
In conclusion the group provided by the invention based on Chebyshev filter is distributed detection method, have following excellent
Point: it finds that the group of complex network is distributed by the semi-supervised learning method of Chebyshev filter, is filtered by Chebyshev
Device carries out the extraction to network hidden layer feature of coarse and pondization operation, and by the Feature Mapping of extraction to flag node space into
Row processing keeps the weight in network shared, to reduce the setting of parameter, improves the efficiency of group's detection.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (5)
1. a kind of group based on Chebyshev filter is distributed detection method, it is characterised in that the following steps are included:
Step 1: reading complex network data, complex network is abstracted as graph model；
Step 2: label initialization, as one unique label of part of nodes label being carried out to the node in graph model, are used for
Indicate the group where the node, other nodes do not mark, and graph model is made to can be used for semi-supervised learning；
Step 3: setting Chebyshev filter is configured with the selection that Chebyshev polynomials parameterize；
Step 4: the node of initialization being input to Chebyshev filter, coarse and pondization operation is carried out, extracts graph model
Hidden layer feature；
Step 5: the Feature Mapping of extraction being handled to flag node space, and output valve is input in classifier, is had
The node for having same label is same group.
2. a kind of group based on Chebyshev filter according to claim 1 is distributed detection method, which is characterized in that
In the step 1, includes node and side in the graph model, abut weight matrix, specific symbol is expressed as: G=(V, E,
W), wherein V be node set, E represent node connection side set, for each node vi∈ V, side (vi,vj) ∈ E, W ∈
RN×NFor the adjoining weight matrix of two nodes.
3. a kind of group based on Chebyshev filter according to claim 1 is distributed detection method, which is characterized in that
In the step 3, Chebyshev polynomials parameterize specifically being expressed as follows for selection:
Wherein, x is a signal, and θ is parameter, TkIt is Chebyshev polynomials, formula are as follows:
Tk(x)=2xTk-1(x)-Tk-2(x) (2)
Wherein, T0(x)=1, T1(x)=x.
4. a kind of group based on Chebyshev filter according to claim 1 is distributed detection method, which is characterized in that
In the step 4, the specific steps are as follows:
Step 4-1: the node of input initialization to Chebyshev filter, using ReLU activation primitive, then the propagation of node is advised
It is then as follows:
Wherein,It is the adjacency matrix of graph model G, INIt is eigenmatrix,Degree of a representation matrix, W(l)
It is l layers of weight matrix, ReLU () is activation primitive, H(l)∈RN×DIt is the activated matrix at l layers, and H(0)=X, X are
The feature of node collection V；
Step 4-2: coarse operation is carried out to the node in graph model, by unlabelled node i no label section adjacent thereto
Point j carries out local normalization, then two adjacent nodes is marked, using the weight of coarse posterior nodal point as the two
The sum of weight of node repeats this process until all nodes are all by coarse；
Step 4-3: pondization operation is carried out to the node after coarse, rearranges node, similar node is then gathered in one
It rises；
Step 4-4: the feature of graph model structure is extracted.
5. a kind of group based on Chebyshev filter according to claim 1 is distributed detection method, which is characterized in that
The specific steps of the step 5 are as follows:
Step 5-1: the Feature Mapping of extraction is handled to flag node space；
Step 5-2: the result of processing is input to classifier, model is as follows:
Wherein, W(0)∈RN×NIt is the weight matrix of input layer, W(1)∈RN×NIt is the weight matrix of output layer, softmax activates letter
Number is defined as follows:
Wherein, Z=∑iexp(xi)；
Step 5-3: output label node, the node with same label are the same group.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810455982.1A CN110310209A (en) | 2018-05-14 | 2018-05-14 | A kind of group's distribution detection method based on Chebyshev filter |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810455982.1A CN110310209A (en) | 2018-05-14 | 2018-05-14 | A kind of group's distribution detection method based on Chebyshev filter |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110310209A true CN110310209A (en) | 2019-10-08 |
Family
ID=68073886
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810455982.1A Pending CN110310209A (en) | 2018-05-14 | 2018-05-14 | A kind of group's distribution detection method based on Chebyshev filter |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110310209A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989508A (en) * | 2021-02-01 | 2021-06-18 | 复旦大学 | Filter optimization design method based on deep learning algorithm |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030172043A1 (en) * | 1998-05-01 | 2003-09-11 | Isabelle Guyon | Methods of identifying patterns in biological systems and uses thereof |
| CN106547876A (en) * | 2016-10-26 | 2017-03-29 | 桂林电子科技大学 | A kind of community discovery processing method propagated based on degree of membership label and system |
| CN106934722A (en) * | 2017-02-24 | 2017-07-07 | 西安电子科技大学 | Multi-objective community detection method based on k node updates Yu similarity matrix |
| CN107392784A (en) * | 2017-08-02 | 2017-11-24 | 西安电子科技大学 | The adjustable overlapping Combo discovering method propagated based on label |
-
2018
- 2018-05-14 CN CN201810455982.1A patent/CN110310209A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030172043A1 (en) * | 1998-05-01 | 2003-09-11 | Isabelle Guyon | Methods of identifying patterns in biological systems and uses thereof |
| CN106547876A (en) * | 2016-10-26 | 2017-03-29 | 桂林电子科技大学 | A kind of community discovery processing method propagated based on degree of membership label and system |
| CN106934722A (en) * | 2017-02-24 | 2017-07-07 | 西安电子科技大学 | Multi-objective community detection method based on k node updates Yu similarity matrix |
| CN107392784A (en) * | 2017-08-02 | 2017-11-24 | 西安电子科技大学 | The adjustable overlapping Combo discovering method propagated based on label |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989508A (en) * | 2021-02-01 | 2021-06-18 | 复旦大学 | Filter optimization design method based on deep learning algorithm |
| CN112989508B (en) * | 2021-02-01 | 2022-05-20 | 复旦大学 | Filter optimization design method based on deep learning algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Che et al. | Distilling knowledge from deep networks with applications to healthcare domain | |
| Chen et al. | Mfanet: A multi-level feature aggregation network for semantic segmentation of land cover | |
| Estrada | Topological structural classes of complex networks | |
| Lee et al. | Automated classification of the tympanic membrane using a convolutional neural network | |
| CN110427990A (en) | A kind of art pattern classification method based on convolutional neural networks | |
| CN105824904B (en) | Chinese herbal medicine picture crawling method based on tcm field profession term vector | |
| CN110297931A (en) | A kind of image search method | |
| CN112035661A (en) | Text emotion analysis method and system based on graph convolution network and electronic device | |
| CN112115265A (en) | Small sample learning method in text classification | |
| CN108053035A (en) | Based on the network representation learning method under the complete non-equilibrium label constrained in approximate class between class | |
| Yin et al. | Maize small leaf spot classification based on improved deep convolutional neural networks with a multi-scale attention mechanism | |
| CN112765370B (en) | Entity alignment method and device of knowledge graph, computer equipment and storage medium | |
| CN111724401A (en) | Image segmentation method and system based on boundary constraint cascade U-Net | |
| CN110993037A (en) | Protein activity prediction device based on multi-view classification model | |
| Xu et al. | Attention-based pyramid network for segmentation and classification of high-resolution and hyperspectral remote sensing images | |
| CN113111657A (en) | Cross-language knowledge graph alignment and fusion method, device and storage medium | |
| Chugh et al. | An image retrieval framework design analysis using saliency structure and color difference histogram | |
| Wang et al. | A deformable convolutional neural network with spatial-channel attention for remote sensing scene classification | |
| Guo et al. | Graph neural networks with diverse spectral filtering | |
| CN110310209A (en) | A kind of group's distribution detection method based on Chebyshev filter | |
| Sundararaman et al. | Transformative role of artificial intelligence in advancing sustainable tomato (Solanum lycopersicum) disease management for global food security: A comprehensive review | |
| Tan et al. | Deep contrastive self-supervised hashing for remote sensing image retrieval | |
| CN112905906B (en) | Recommendation method and system fusing local collaboration and feature intersection | |
| Béthune et al. | Hierarchical and unsupervised graph representation learning with loukas’s coarsening | |
| CN111241326A (en) | Image visual relation referring and positioning method based on attention pyramid network |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191008 |