CN109960755A - A kind of privacy of user guard method based on Dynamic iterations Fast Field - Google Patents
A kind of privacy of user guard method based on Dynamic iterations Fast Field Download PDFInfo
- Publication number
- CN109960755A CN109960755A CN201910125911.XA CN201910125911A CN109960755A CN 109960755 A CN109960755 A CN 109960755A CN 201910125911 A CN201910125911 A CN 201910125911A CN 109960755 A CN109960755 A CN 109960755A
- Authority
- CN
- China
- Prior art keywords
- network
- privacy
- node
- matrix
- objective function
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 239000011159 matrix material Substances 0.000 claims abstract description 53
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 16
- 238000013528 artificial neural network Methods 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000009795 derivation Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 description 18
- 230000008859 change Effects 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000003012 network analysis Methods 0.000 description 2
- 240000000233 Melia azedarach Species 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000010348 incorporation Methods 0.000 description 1
- 238000003780 insertion Methods 0.000 description 1
- 230000037431 insertion Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000001681 protective effect Effects 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 238000004088 simulation Methods 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
- 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—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computing Systems (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Tourism & Hospitality (AREA)
- Primary Health Care (AREA)
- Marketing (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Strategic Management (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of privacy of user guard methods based on Dynamic iterations Fast Field, comprising: (1) for primitive network, using primitive network and corresponding true category training figure convolutional neural networks, and determines picture scroll product neural network parameter;(2) according to picture scroll product neural network parameter, cross entropy using in primitive network before and after destination node reconnection side is as hiding objective function, and it calculates and hides objective function to the gradient information of adjacency matrix, comprehensive last gradient information calculates momentum information again, is updated according to the momentum information of acquisition to the corresponding adjacency matrix of primitive network.The privacy of user guard method can fast and effeciently hide the information of user, realize the protection to privacy of user.
Description
Technical field
The invention belongs to the network privacys to protect field, and in particular to a kind of privacy of user based on Dynamic iterations Fast Field
Guard method.
Background technique
There are many network structures in real life, social networks is one of.Each section in social networks
Point indicates people, and even side indicates interpersonal exchange of information or friends.Third party can be according to the company side between node
Relationship can extract the feature of each node, and then carry out various analyses to them, such as they are partitioned into different clusters,
Or predict that they may interested people or thing.Instantly it is suggested there are many figure incorporation model, for the knot for analyzing social networks
Structure and feature provide new method.
The purpose of figure embedded mobile GIS is mapped to network structure in one lower dimensional space, thus by traditional network analysis
Problem is converted into mathematical problem and is solved.It is inspired by word2vec, skip-gram model is widely used in figure insertion neck
Domain produces a large amount of figure embedded mobile GIS, such as Deepwalk, LINE and node2vec algorithm.They are usually by random walk
Applied in sequence node, and these sequence nodes are regarded as the sentence in word2vec model.Be based on skip-gram mould
The figure embedded mobile GIS of type is different, and figure convolutional network (GCN) is a kind of picture scroll integration method based on deep learning.It learns Local map
Structure and node diagnostic simultaneously map that in hidden layer.Figure convolutional network it is only necessary to a small amount of node labels can be effectively
Nodes all in network are mapped in lower dimensional space, this is that other algorithms cannot compare.
Although figure embedded mobile GIS obtains huge success in network analysis field, it also results in many information and lets out
Dew problem, secret protection also become to have been to be concerned by more and more people.It is expected that the more information of acquisition are different from third party, it is many
User is not intended to the information of oneself to be used in business activity, such as the politics exchange of politician, the investigation of plainclothes policeman are living
Move etc..These are not intended to the tag attributes of oneself to be found and be used by third party per capita.In this case, it needs to borrow
Network privacy protection algorism is helped, it is by carrying out the larger change that destination node classification results are realized in lesser change to network structure
It is dynamic, to influence the result of each Network algorithm in downstream.
The network huge for one, influence of the different parts to destination node attribute are different.Therefore it needs to borrow
Suitable model is helped to find these even Bian Zuhes maximum to destination node properties affect.
Summary of the invention
The object of the present invention is to provide a kind of privacy of user guard method based on Dynamic iterations Fast Field, the users
Method for secret protection can fast and effeciently hide the information of user, realize the protection to privacy of user.
The technical solution of the present invention is as follows:
A kind of privacy of user guard method based on Dynamic iterations Fast Field, comprising the following steps:
(1) for primitive network, using primitive network and corresponding true category training figure convolutional neural networks (Graph
Convolutional Network, GCN), and determine picture scroll product neural network parameter;
(2) according to picture scroll product neural network parameter, cross entropy using in primitive network before and after destination node reconnection side as
Objective function is hidden, and calculates and hides objective function to the gradient information of adjacency matrix, then comprehensive last gradient information meter
Momentum information is calculated, the corresponding adjacency matrix of primitive network is updated according to the momentum information of acquisition.
In privacy of user guard method provided in the present invention, the optimal solution on reconnection side is searched for by GCN, such one
Come, the attribute change of metric objective node is just converted into the variation issue of metric objective node diagnostic vector.In view of being based on
The case where simple iteration of gradient is easily trapped into local optimum also considers upper primary gradient information when updating adjacency matrix,
Gradient information twice is merged into momentum information and determines the direction of search, to help algorithm to jump out local best points and increase calculation
The transportable property of method.
Preferably, in training figure convolutional neural networks, to the expression formula of transmitting before figure convolutional neural networks are as follows:
Wherein, X is the eigenmatrix of primitive network G interior joint, ForDegree matrix,The adjacency matrix from connection is had for primitive network G, I is unit matrix, and A is the adjacency matrix of primitive network G,
W0∈RC×HAnd W1∈RH×|F|It is output layer respectively to hidden layer and hidden layer to the weight of output layer, F is category set, f ()
It is softmax activation primitive and Relu activation primitive respectively with σ ();
And using objective function L as target, picture scroll product neural network parameter is updated,
Wherein, VLFor the node set containing category, F is category set, and Y is true category matrix, if first of node vl
Class be designated as h, then Ylh=1, otherwise Ylh=0, Y 'lhIt (A) is the output result of figure convolutional neural networks.
In training figure convolutional neural networks, i-th layer is updated in figure convolutional neural networks using Fast Field descent method
Weight Wi:
Wherein, η is learning rate, WiWeight W comprising i-th of output layer to hidden layeri,0With i-th of output layer to hide
The weight W of layeri,1。
After training, figure convolutional neural networks have been determined, the picture scroll product mind determined when being subsequently generated confrontation network
It is constant through network parameter.It is to regard as to the adjacency matrix of primitive network when destination node is hidden in primitive network
One variable, update is iterated to it.Specifically, to the hiding objective function L of destination node t buildingtAre as follows:
Wherein, F is category set, and Y is true category matrix, if first of node vlClass be designated as h, then Yth=1, otherwise
Yth=0, Y 'thIt (A) is the output result of figure convolutional neural networks;
Hide objective function LtIt indicates for the prediction category of destination node t and the difference size of true category.For hidden
For private protection, target is the value for increasing community as far as possible by modification adjacency matrix A and hiding objective function.Therefore, will
Hide objective function LtDerivation is carried out to adjacency matrix A, obtains gradient matrix
Wherein, i and j is node index.
Due to gradient matrixBe it is asymmetrical, before calculating momentum information, it is also necessary to construct gradient network g:
In gradient network g, all there is a numerical value, if the numerical value is positive, then it represents that in the node between any node pair
Increase even side between pair so as to hide objective function LtIncrease, otherwise, indicates to delete the company side between the node pair so as to hide target
Function LtIncrease.
Wherein, the comprehensive last gradient information calculating momentum information includes:
The momentum that kth time iteration generatesIt is as follows:
Wherein, μ is decay factor,||gk-1||1For previous generation gradient network gk-1A norm.
The present invention directly updates adjacency matrix A without using gradient network g, but uses the gradient network of -1 iteration of kth
gk-1And momentumCalculate new momentumLast adjacency matrix A is updated by new momentumk-1.It is such good
Place is the not stringent current gradient direction of the new direction of search, the new side that the gradient information before integrating obtains
To facilitating algorithm and jump out locally optimal solution.
The momentum information according to acquisition, which is updated the corresponding adjacency matrix of primitive network, includes:
From momentumIn select the node pair of element maximum absolute value, if the element of maximum absolute value is positive,
Otherwise increasing even side between the node pair, indicates to delete the company side between the node pair, to update adjacency matrixIt obtainsNote
Meaning
Beneficial effects of the present invention are mainly manifested in:
Globally optimal solution can effectively be searched by being updated based on gradient momentum to adjacency matrix, avoid falling into part
Optimal solution reinforces the transportable property of algorithm.Using depth model GCN search reconnection side strategy, by the attribute change of destination node
Be converted into the variation of its distance between feature vector before and after reconnection side, it is more intuitive in this way and convenient for algorithm faster, it is more smart
Really finding influences maximum even side to destination node.Finally in real data set the experimental results showed that, which has
Good applicability and scalability can fast and effeciently hide the information of user, realize the protection to privacy of user.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to do simply to introduce, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art, can be with root under the premise of not making the creative labor
Other accompanying drawings are obtained according to these attached drawings.
Fig. 1 is the flow chart of the privacy of user guard method the present invention is based on Dynamic iterations Fast Field;
Fig. 2 is the structural block diagram of the privacy of user guard method the present invention is based on Dynamic iterations Fast Field.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention more comprehensible, with reference to the accompanying drawings and embodiments to this
Invention is described in further detail.It should be appreciated that the specific embodiments described herein are only used to explain the present invention,
And the scope of protection of the present invention is not limited.
As depicted in figs. 1 and 2, the privacy of user guard method provided by the invention based on Dynamic iterations Fast Field, packet
Include following steps:
The present embodiment obtains the weight of each layer matrix of GCN using former network and true tag training GCN with this first, this
A little weights no longer change when generating confrontation network later.For a Undirected networks G, its adjacency matrix is A, thenThe adjacency matrix from connection is had for network G, I is unit matrix.The forward process of GCN can be indicated simply
Are as follows:
Wherein X is node diagnostic matrix, ForDegree matrix, W0∈RC×HAnd W1∈RH×|F|Point
It is not output layer to hidden layer and hidden layer to the weight of output layer, F is category set, and f () and σ () are respectively
Softmax activation primitive and Relu activation primitive;
For node-classification, the present invention uses cross entropy as objective function, and definition is as shown in formula (2).
Wherein VLFor the node set containing category, F is category set, and Y is true category matrix, wherein if node vl
Class be designated as h, then Ylh=1, otherwise Ylh=0, Y 'lh(A) output for being GCN is as a result, A is the adjacency matrix of primitive network.
In the m times iteration, each layer weight W in neural network is updated using Fast Field descent methodi:
Wherein η is learning rate.
After the complete GCN of training, it will abut against matrix A and regard a variable as and be iterated update to it.Firstly for target
Node t defines community and hides objective function LtAre as follows:
This hides objective function LtIt indicates for the prediction category of destination node t and the difference size of true category.For
For secret protection, the target of this paper is the value for increasing community as far as possible by modification adjacency matrix A and hiding objective function.
For this purpose, by LtDerivation is carried out to adjacency matrix A, to obtain gradient matrix
Because of gradient matrixIt is asymmetrical, so shown in building gradient network g such as formula (6).
In gradient network g, all there is a numerical value between any node pair.If the value is positive, then it represents that in the node pair
Between increase even side will make objective function LtIncrease, otherwise indicates that the company side deleted between the node pair will make LtIncrease.And the value
Absolute value is bigger, and current additions and deletions connect side to LtInfluence it is bigger.It should be noted that if the value is positive (negative), this node pair
Between have (not having) even side, then ignore such node pair.
The present invention directly updates adjacency matrix A without using gradient network g, but uses the gradient network of -1 iteration of kth
gk-1And momentumCalculate new momentumLast adjacency matrix A is updated by new momentumk-1.It counts in this way
The benefit of calculation is the not stringent current gradient direction of the new direction of search, obtain one of the gradient information before integrating
New direction is therefore this facilitates algorithm and jumps out locally optimal solution.Define the momentum that kth time iteration generatesIt is as follows.
Finally from momentumIn select the node pair of element maximum absolute value, according to the positive and negative of maximum absolute value element
Update confrontation network adjacent matrixIt obtainsPay attention to
The step of generating confrontation network is as follows:
A-1: primitive network G training GCN model is used;
A-2: the adjacency matrix of initialization confrontation networkInitial momentum
A-3: iteration secondary for kth, according toConstruct gradient network gk-1;
A-4: momentum is calculated according to formula 7
A-5: fromIn select the node pair of maximum absolute value, new confrontation net is obtained according to update rule described above
Network adjacency matrix
A-6: if k < K, repeatedly a-3~a-5.
Experiment simulation
The effect of MIFGS algorithm is tested in experiment using political blog data collection.The data set reflects blog in network
Political orientation, wherein node indicates blog, and even side is crawled automatically from blog homepage.1490 nodes are shared in network,
19090 sides, 2 classes.
It selects secret protection success rate (ASR) and averagely protects successfully required reconnection number of edges (AML) as evaluation index
To measure the ability that each algorithm hides certain specific node.
ASR: the ratio that average protective success ratio, i.e. certain figure embedded mobile GIS classify destination node mistake.Here it disturbs
Size K takes 1 to 20, and totally 20.
AML: reconnection side number needed for average successfully hiding destination node, limiting the reconnection side upper limit herein is 20, if certain
Node cannot be hidden successfully by modifying 20 Lian Bianlai, then take the 20 reconnection number of edges hiding as success.
In order to verify the privacy of user guard method (abbreviation MIFGS) provided by the invention based on Dynamic iterations Fast Field
Middle momentum contributes positively to algorithm and jumps out local best points, updates adjacency matrix by MIFGS and directly using gradient network's
FGS algorithm and DICE are compared.It is as follows to sketch two kinds of control algorithms:
FGS: the adjacency matrix of confrontation network is directly updated using gradient network gSpecifically, ladder is directly selected
The node pair for spending maximum absolute value in network g, by update Policy Updates described above
DICE: a kind of simple heuristic network privacy protection algorism, if K is disturbance, size (is in MIFGS and FGS
Maximum number of iterations), then the company side of random erasure b (b < K) destination node, and at random by destination node and K-b other classes
Other node is connected.
Take μ=0.5 herein in an experiment, each algorithm final result is as shown in table 1.
Each algorithm final result of the political blog data collection of table 1
From table 1 it follows that MIFGS all indicators are better than FGS, this, which has been absolutely proved, uses momentum information update
Confrontation network will be helpful to algorithm and quickly jump out local best points, so that result is more acurrate.In addition either MIFGS is still
FGS algorithm is all substantially better than DICE algorithm, this shows that it is feasible for converting mathematical problem for traditional network problem analysis by GCN
, and accurate result can be obtained.
Technical solution of the present invention and beneficial effect is described in detail in above-described specific embodiment, Ying Li
Solution is not intended to restrict the invention the foregoing is merely presently most preferred embodiment of the invention, all in principle model of the invention
Interior done any modification, supplementary, and equivalent replacement etc. are enclosed, should all be included in the protection scope of the present invention.
Claims (7)
1. a kind of privacy of user guard method based on Dynamic iterations Fast Field, comprising the following steps:
(1) for primitive network, using primitive network and corresponding true category training figure convolutional neural networks, and picture scroll is determined
Product neural network parameter;
(2) according to picture scroll product neural network parameter, cross entropy using in primitive network before and after destination node reconnection side is as hiding
Objective function, and it is dynamic to the gradient information of adjacency matrix, then comprehensive last gradient information calculating to calculate hiding objective function
Information is measured, the corresponding adjacency matrix of primitive network is updated according to the momentum information of acquisition.
2. as described in claim 1 based on the privacy of user guard method of Dynamic iterations Fast Field, which is characterized in that instructing
When practicing figure convolutional neural networks, to the expression formula of transmitting before figure convolutional neural networks are as follows:
Wherein, X is the eigenmatrix of primitive network G interior joint, ForDegree matrix,The adjacency matrix from connection is had for primitive network G, I is unit matrix, and A is the adjacency matrix of primitive network G,
W0∈RC×HAnd W1∈RH×|F|It is output layer respectively to hidden layer and hidden layer to the weight of output layer, F is category set, f ()
It is softmax activation primitive and Relu activation primitive respectively with σ ();
And using objective function L as target, picture scroll product neural network parameter is updated,
Wherein, VLFor the node set containing category, F is category set, and Y is true category matrix, if first of node vlClass
It is designated as h, then Ylh=1, otherwise Ylh=0, Y 'lhIt (A) is the output result of figure convolutional neural networks.
3. as claimed in claim 2 based on the privacy of user guard method of Dynamic iterations Fast Field, which is characterized in that instructing
When practicing figure convolutional neural networks, i-th layer of weight W in figure convolutional neural networks is updated using Fast Field descent methodi:
Wherein, η is learning rate, WiWeight W comprising i-th of output layer to hidden layerI, 0With the power of i-th of output layer to hidden layer
Weight WI, 1。
4. as described in claim 1 based on the privacy of user guard method of Dynamic iterations Fast Field, which is characterized in that mesh
Mark the hiding objective function L of node t buildingtAre as follows:
Wherein, F is category set, and Y is true category matrix, if first of node vlClass be designated as h, then Yth=1, otherwise Yth=
0, Y 'thIt (A) is the output result of figure convolutional neural networks;
By hiding objective function LtDerivation is carried out to adjacency matrix A, obtains gradient matrix
Wherein, i and j is node index.
5. as described in claim 1 based on the privacy of user guard method of Dynamic iterations Fast Field, which is characterized in that counting
Before calculation momentum information, it is also necessary to construct gradient network g:
In gradient network g, all there is a numerical value, if the numerical value is positive, then it represents that between the node pair between any node pair
Increase even side so as to hide objective function LtIncrease, otherwise, indicates to delete the company side between the node pair so as to hide objective function Lt
Increase.
6. as claimed in claim 5 based on the privacy of user guard method of Dynamic iterations Fast Field, which is characterized in that described
Comprehensive last gradient information calculates momentum information
The momentum that kth time iteration generatesIt is as follows:
Wherein, μ is decay factor,For previous generation gradient network gk-1A norm.
7. as claimed in claim 6 based on the privacy of user guard method of Dynamic iterations Fast Field, which is characterized in that described
The corresponding adjacency matrix of primitive network is updated according to the momentum information of acquisition and includes:
From momentumIn select the node pair of element maximum absolute value, if the element of maximum absolute value is positive, in the section
Otherwise increasing even side between point pair, indicates to delete the company side between the node pair, to update adjacency matrixIt obtainsPay attention to
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910125911.XA CN109960755B (en) | 2019-02-20 | 2019-02-20 | User privacy protection method based on dynamic iteration fast gradient |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910125911.XA CN109960755B (en) | 2019-02-20 | 2019-02-20 | User privacy protection method based on dynamic iteration fast gradient |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109960755A true CN109960755A (en) | 2019-07-02 |
CN109960755B CN109960755B (en) | 2021-03-05 |
Family
ID=67023594
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910125911.XA Active CN109960755B (en) | 2019-02-20 | 2019-02-20 | User privacy protection method based on dynamic iteration fast gradient |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109960755B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110674869A (en) * | 2019-09-23 | 2020-01-10 | 腾讯科技(深圳)有限公司 | Classification processing and graph convolution neural network model training method and device |
CN111275189A (en) * | 2020-02-25 | 2020-06-12 | 支付宝(杭州)信息技术有限公司 | Network representation learning method, system and device |
CN111859454A (en) * | 2020-07-28 | 2020-10-30 | 桂林慧谷人工智能产业技术研究院 | Privacy protection method for defending link prediction based on graph neural network |
CN111882054A (en) * | 2020-05-27 | 2020-11-03 | 杭州中奥科技有限公司 | Method and related equipment for cross training of network data of encryption relationship between two parties |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2010144645A2 (en) * | 2009-06-12 | 2010-12-16 | Brian Burk | Healthcare provider resources online |
CN103488637A (en) * | 2012-06-11 | 2014-01-01 | 北京大学 | Method for carrying out expert search based on dynamic community mining |
CN104834747A (en) * | 2015-05-25 | 2015-08-12 | 中国科学院自动化研究所 | Short text classification method based on convolution neutral network |
CN106097355A (en) * | 2016-06-14 | 2016-11-09 | 山东大学 | The micro-Hyperspectral imagery processing method of gastroenteric tumor based on convolutional neural networks |
CN106847302A (en) * | 2017-02-17 | 2017-06-13 | 大连理工大学 | Single channel mixing voice time-domain seperation method based on convolutional neural networks |
US20180046762A1 (en) * | 2014-08-17 | 2018-02-15 | Bridget Osetinsky | Comprehension Normalization Method For Networks |
CN108196251A (en) * | 2017-12-25 | 2018-06-22 | 电子科技大学 | Accelerated iteration regularization super-resolution imaging method based on vector extrapolation |
CN108664687A (en) * | 2018-03-22 | 2018-10-16 | 浙江工业大学 | A kind of industrial control system space-time data prediction technique based on deep learning |
-
2019
- 2019-02-20 CN CN201910125911.XA patent/CN109960755B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2010144645A2 (en) * | 2009-06-12 | 2010-12-16 | Brian Burk | Healthcare provider resources online |
CN103488637A (en) * | 2012-06-11 | 2014-01-01 | 北京大学 | Method for carrying out expert search based on dynamic community mining |
US20180046762A1 (en) * | 2014-08-17 | 2018-02-15 | Bridget Osetinsky | Comprehension Normalization Method For Networks |
CN104834747A (en) * | 2015-05-25 | 2015-08-12 | 中国科学院自动化研究所 | Short text classification method based on convolution neutral network |
CN106097355A (en) * | 2016-06-14 | 2016-11-09 | 山东大学 | The micro-Hyperspectral imagery processing method of gastroenteric tumor based on convolutional neural networks |
CN106847302A (en) * | 2017-02-17 | 2017-06-13 | 大连理工大学 | Single channel mixing voice time-domain seperation method based on convolutional neural networks |
CN108196251A (en) * | 2017-12-25 | 2018-06-22 | 电子科技大学 | Accelerated iteration regularization super-resolution imaging method based on vector extrapolation |
CN108664687A (en) * | 2018-03-22 | 2018-10-16 | 浙江工业大学 | A kind of industrial control system space-time data prediction technique based on deep learning |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110674869A (en) * | 2019-09-23 | 2020-01-10 | 腾讯科技(深圳)有限公司 | Classification processing and graph convolution neural network model training method and device |
CN110674869B (en) * | 2019-09-23 | 2023-04-18 | 腾讯科技(深圳)有限公司 | Classification processing and graph convolution neural network model training method and device |
CN111275189A (en) * | 2020-02-25 | 2020-06-12 | 支付宝(杭州)信息技术有限公司 | Network representation learning method, system and device |
CN111882054A (en) * | 2020-05-27 | 2020-11-03 | 杭州中奥科技有限公司 | Method and related equipment for cross training of network data of encryption relationship between two parties |
CN111882054B (en) * | 2020-05-27 | 2024-04-12 | 杭州中奥科技有限公司 | Method for cross training of encryption relationship network data of two parties and related equipment |
CN111859454A (en) * | 2020-07-28 | 2020-10-30 | 桂林慧谷人工智能产业技术研究院 | Privacy protection method for defending link prediction based on graph neural network |
CN111859454B (en) * | 2020-07-28 | 2024-03-29 | 桂林慧谷人工智能产业技术研究院 | Privacy protection method for defending link prediction based on graph neural network |
Also Published As
Publication number | Publication date |
---|---|
CN109960755B (en) | 2021-03-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109960755A (en) | A kind of privacy of user guard method based on Dynamic iterations Fast Field | |
CN110457404B (en) | Social media account classification method based on complex heterogeneous network | |
Serrano et al. | A novel agent-based rumor spreading model in twitter | |
CN107330461A (en) | Collaborative filtering recommending method based on emotion with trust | |
Baggio | Studying complex tourism systems: a novel approach based on networks derived from a time series | |
CN109325525A (en) | Sample attribute assessment models training method, device and server | |
CN112084373B (en) | Graph embedding-based multi-source heterogeneous network user alignment method | |
CN112036445B (en) | User identity recognition method across social networks based on neural tensor network | |
CN112765653B (en) | Multi-source data fusion privacy protection method based on multi-privacy policy combination optimization | |
CN109344887A (en) | Short video classification methods, system and medium based on multi-modal dictionary learning | |
CN113255895A (en) | Graph neural network representation learning-based structure graph alignment method and multi-graph joint data mining method | |
CN109522416A (en) | A kind of construction method of Financial Risk Control knowledge mapping | |
CN105825430A (en) | Heterogeneous social network-based detection method | |
CN105760649A (en) | Big-data-oriented creditability measuring method | |
CN109492027B (en) | Cross-community potential character relation analysis method based on weak credible data | |
CN109492076A (en) | A kind of network-based community's question and answer website answer credible evaluation method | |
Sun et al. | GPOGC: Gaussian pigeon-oriented graph clustering algorithm for social networks cluster | |
CN109948242A (en) | Network representation learning method based on feature Hash | |
CN111598252B (en) | University computer basic knowledge problem solving method based on deep learning | |
Han et al. | A topic representation model for online social networks based on hybrid human–artificial intelligence | |
CN109783805A (en) | A kind of network community user recognition methods and device | |
CN116521936A (en) | Course recommendation method and device based on user behavior analysis and storage medium | |
CN109885797A (en) | A kind of relational network construction method based on the mapping of more identity spaces | |
He et al. | A topic community-based method for friend recommendation in online social networks via joint nonnegative matrix factorization | |
Madhan et al. | Analyzing complexity nature inspired optimization algorithms using halstead metrics |
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 |