CN111415265A - Social relationship data generation method of generative confrontation network - Google Patents
Social relationship data generation method of generative confrontation network Download PDFInfo
- Publication number
- CN111415265A CN111415265A CN202010104903.XA CN202010104903A CN111415265A CN 111415265 A CN111415265 A CN 111415265A CN 202010104903 A CN202010104903 A CN 202010104903A CN 111415265 A CN111415265 A CN 111415265A
- Authority
- CN
- China
- Prior art keywords
- relationship data
- generator
- discriminator
- data
- pseudo
- 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
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000013528 artificial neural network Methods 0.000 claims abstract description 25
- 239000011159 matrix material Substances 0.000 claims description 25
- 230000008569 process Effects 0.000 claims description 8
- 230000014509 gene expression Effects 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims description 3
- 238000010801 machine learning Methods 0.000 abstract description 6
- 230000006870 function Effects 0.000 description 8
- 238000013499 data model Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000006855 networking Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Marketing (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Primary Health Care (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Machine Translation (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a social relationship data generation method of a generative confrontation network, which comprises the following steps: A. storing relationship data of a social group, including user node information and user relationship information; B. constructing a generator comprising a first graph neural network receiving random noise for generating pseudo-relational data; C. constructing a discriminator which comprises a second graph neural network for receiving the relationship data and the pseudo relationship data and is used for judging the difference degree of the characteristic structure between the relationship data and the pseudo relationship data; D. and alternately training the generator and the discriminator until the discriminator judges that the pseudo-relationship data and the relationship data have the same characteristic structure, and using the current generator to generate a social relationship data sample. The method and the device can generate the social relationship data with the same characteristics as the real relationship data, more conveniently simulate the social relationship data and research the characteristics of the social relationship data, and provide richer samples for machine learning models related to the social network relationship data.
Description
Technical Field
The invention relates to a generation method of social relationship data, in particular to a generation method of social relationship data of a generative confrontation network.
Background
There are many data models and methods for the analysis and generation of social relationship network data, and different data models and methods have their own respective points of emphasis and deficiencies. For example, the common Barabasi-Albert data model, when generating social networking data, simulates the preferred contact of the social networking data by creating a random graph with n nodes.
The generation step of the random graph comprises the following steps:
The method comprises the following steps: executing the second step according to the probability p, otherwise executing the third step;
Connecting a new node to an old node which is uniformly selected randomly;
And step three, connecting the new node to the n old nodes with a probability proportional to the n old nodes, wherein the probability of connecting to the old nodes is the degree of the old nodes/the total degree.
The goal of this graph is to model preferential linking (i.e., assigning an amount based on the amount already present in each individual or object, which generally further increases the advantage of the dominant individual.
However, the Barabasi-Albert data model has no learning ability and cannot simulate the characteristics of a specific social network data set, so that the Barabasi-Albert data model has limitation in practical application.
Disclosure of Invention
The invention provides a social relationship data generation method of a generative confrontation network, which enables the generated social relationship data to have the same characteristics as real relationship data by continuously learning a machine learning model.
The invention discloses a social relationship data generation method of a generative confrontation network, which comprises the following steps:
A. Storing relationship data of a social group on a storage device, wherein the relationship data comprises user node information representing nodes and user relationship information representing edges;
B. Constructing a generator, wherein the generator comprises a first graph neural network for receiving random noise input, and generates pseudo relation data similar to the characteristic structure of the relation data according to the relation data stored on a storage device;
C. Constructing a discriminator, wherein the discriminator comprises a second graph neural network for receiving the relationship data and the pseudo relationship data generated by the generator, and the discriminator is used for judging the difference degree between the characteristic structure of the relationship data and the characteristic structure of the pseudo relationship data;
D. And alternately training the generator and the discriminator until the discriminator judges that the pseudo-relationship data and the relationship data have the same characteristic structure, and using the currently optimized generator to generate a generator of a social relationship data sample.
The pseudo relationship data and the relationship data have the same characteristic structure, which means that the pseudo relationship data and the relationship data have extremely similar characteristic structures. The relationship data in step a is relationship data of a real social group, for example, social group relationship data disclosed by FaceBook: https:// snap. The generator and the discriminator are alternately trained, so that the generator and the discriminator are respectively converged in the training and learning processes, the performance of the generator and the discriminator is not improved any more, the training is stopped after the number of iteration rounds of the training, and the discriminator cannot distinguish the pseudo data for two data samples of the real relationship data and the pseudo relationship data generated by the generator, namely, the generated pseudo relationship data and the real relationship data are considered to have the same characteristic structure.
The generator and the discriminator can be realized by a person skilled in the art according to the conventional technical means in the art, and the detailed process is not described in detail.
Further, the user node information in step a includes a user node information matrix V ═ V 1,v2,...,vnAnd the user relationship information comprises a user relationship information adjacency matrix Wherein v is irepresents the ith user node, i ∈ [1, n ] ],eijIndicating whether the ith user node is directly connected with the jth user node.
Further, in step B, the first graph neural network receiving the random noise is input to the generator, and the generator generates pseudo-relationship data similar to the feature structure of the relationship data according to the relationship data stored in the storage device by the random noise, where the pseudo-relationship data includes a pseudo-user node information matrix and a pseudo-user relationship information adjacency matrix.
Specifically, in step D, the expressions of the training generator and the discriminator are:
The generator G and the discriminator D are optimized by alternately maximizing and minimizing an objective function V (G; D), where V represents a user node information matrix and V ═ V { (V } 1,v2,...,vn},And All represent mathematical expectations, v to p true(·|vi) Representing user nodes v from a user node information matrix i,v~G(·|vi;θG) Representing user nodes v from a generator i,D(v;vc;θD) Denotes a discriminator, D (v; v. of c;θD) V in (1) cRepresenting other user nodes than V in V, θ GRepresenting the weight, θ, of the first graph neural network in the generator DRepresents weights of a second graph neural network in the discriminator, and ptrue(v) Representing a probability density function of a user node information matrix obeyed by a user node v, and dv represents the differential of v; the generator G and the discriminator D are optimized by continuously updating the weights of the respective graph neural networks through iteration until the discriminator D judges that the pseudo-relation data generated by the generator G and the relation data have the same characteristic structure.
Wherein V and p trueBoth represent a user node information matrix, i.e. a real data set of user nodes, the difference between them being that when p is used trueThe time is represented by emphasizing the probability distribution, and the time is represented by emphasizing the data composition mode by adopting V.
Further, in the iterative process of the generator G and the discriminator D, the discriminator D receives a positive sample of the user node information matrix and a negative sample of the generator G for training, and the training process is carried out through a first gradient Second graph neural network weights θ for discriminator D DUpdating to optimize an objective function Generator G passes through a second gradient First graph neural network weights θ for generator G GUpdating to optimize an objective function
The social relationship data generation method of the generative confrontation network can enable the machine learning model to generate the social relationship data with the same characteristics as the real relationship data through learning, so that the social relationship data can be simulated more conveniently, the characteristics of the social relationship data can be researched more conveniently, and richer samples are provided for the machine learning model related to the social network relationship data.
The present invention will be described in further detail with reference to the following examples. This should not be understood as limiting the scope of the above-described subject matter of the present invention to the following examples. Various substitutions and alterations according to the general knowledge and conventional practice in the art are intended to be included within the scope of the present invention without departing from the technical spirit of the present invention as described above.
Drawings
FIG. 1 is a flow chart of a method for generating social relationship data for a generative confrontation network according to the present invention.
Fig. 2 is a schematic diagram of a characteristic structure of real data.
FIG. 3 is a schematic diagram of the random noise initialization parameters of the generator.
Fig. 4 is a schematic diagram of a feature structure generated by the generator during an iterative update process according to the initial parameters of fig. 3, which approximates fig. 2.
Fig. 5 is a schematic diagram of a characteristic structure of the generator generated according to the initial parameters of fig. 3, which is substantially the same as that of fig. 2.
Detailed Description
The method for generating social relationship data of a generative confrontation network of the present invention as shown in fig. 1 to 5 includes:
A. Storing relationship data of a social group, such as social group relationship data disclosed by FaceBook, on a storage device: https:// snap. As shown in fig. 2, the relationship data includes user node information representing nodes and user relationship information representing edges. The user node information comprises a user node information matrix V ═ V 1,v2,...,vnAnd the user relationship information comprises a user relationship information adjacency matrix Wherein v is irepresents the ith user node, i ∈ [1, n ] ],eijIndicating whether the ith user node is directly connected with the jth user node.
B. A generator is constructed that includes a first graph neural network that receives a random noise input, the initial parameters of which are shown in fig. 3. According to the relation data stored on the storage device, the generator generates pseudo relation data similar to the characteristic structure of the relation data. The pseudo-relation data comprises a pseudo-user node information matrix and a pseudo-user relation information adjacency matrix.
C. And constructing a discriminator, wherein the discriminator comprises a second graph neural network which receives the relation data and the pseudo relation data generated by the generator, and the discriminator is used for judging the difference degree between the characteristic structure of the relation data and the characteristic structure of the pseudo relation data.
The generator and the discriminator can be realized by a person skilled in the art according to the conventional technical means in the art, and the detailed process is not described in detail.
D. Alternating the training generators and discriminators, as shown in FIG. 4, progressively generates a feature structure that approximates the true relational data during the iterative update process. The expressions for the training generator and the arbiter are:
The generator G and the discriminator D are optimized by alternately maximizing and minimizing an objective function V (G; D), where V represents a user node information matrix and V ═ V { (V } 1,v2,...,vn},And All represent mathematical expectations, v to p true(·|vi) Representing user nodes v from a user node information matrix i,v~G(·|vi;θG) Representing user nodes v from a generator i,D(v;vc;θD) Denotes a discriminator, D (v; v. of c;θD) V in (1) cRepresenting other user nodes than V in V, θ GRepresenting the weight, θ, of the first graph neural network in the generator DRepresents weights of a second graph neural network in the discriminator, and ptrue(v) Representing the probability density function of the user node information matrix to which the user node v is subjected, dv representing the differential of v. Wherein V and p trueBoth represent a user node information matrix, i.e. a real data set of user nodes, the difference between them being that when p is used trueThe time is represented by emphasizing the probability distribution, and the time is represented by emphasizing the data composition mode by adopting V.
During the iterative process of the generator G and the discriminator D, the discriminator D receives the real data set of the user node PtrueAnd the negative sample of the generator G, by a first gradient during the training process Second graph neural network weights θ for discriminator D DUpdating to optimize an objective function Generator G passes through a second gradient First graph neural network weights θ for generator G GUpdating to optimize an objective function
As shown in fig. 5, the generator G and the arbiter D perform optimization by iteratively updating their respective weights of the neural network continuously until the arbiter D determines that the pseudo relationship data generated by the generator G is the same as the feature structure of the relationship data, and then the currently optimized generator is used as a generator for generating a social relationship data sample.
According to the method, the machine learning model generates social relationship data with the same characteristics as the real relationship data through learning, so that the social relationship data are more conveniently simulated, the characteristics of the social relationship data are more conveniently researched, and richer samples are provided for the machine learning model related to the social network relationship data.
Claims (5)
1. The social relationship data generation method of the generative confrontation network is characterized by comprising the following steps:
A. Storing relationship data of a social group on a storage device, wherein the relationship data comprises user node information representing nodes and user relationship information representing edges;
B. Constructing a generator, wherein the generator comprises a first graph neural network for receiving random noise input, and generates pseudo relation data similar to the characteristic structure of the relation data according to the relation data stored on a storage device;
C. Constructing a discriminator, wherein the discriminator comprises a second graph neural network for receiving the relationship data and the pseudo relationship data generated by the generator, and the discriminator is used for judging the difference degree between the characteristic structure of the relationship data and the characteristic structure of the pseudo relationship data;
D. And alternately training the generator and the discriminator until the discriminator judges that the pseudo-relationship data and the relationship data have the same characteristic structure, and using the currently optimized generator to generate a generator of a social relationship data sample.
2. The method of generating social relationship data for a generative confrontation network as recited in claim 1, wherein: the user node information in step a includes a user node information matrix V ═ V 1,v2,...,vnAnd the user relationship information comprises a user relationship information adjacency matrix Wherein v is irepresents the ith user node, i ∈ [1, n ] ],eijIndicating whether the ith user node is directly connected with the jth user node.
3. The method of generating social relationship data for a generative confrontation network as recited in claim 2, wherein: and step B, inputting the received random noise into a first graph neural network of a generator, wherein the generator generates pseudo-relation data similar to the characteristic structure of the relation data according to the relation data stored in the storage device through the random noise, and the pseudo-relation data comprises a pseudo user node information matrix and a pseudo user relation information adjacency matrix.
4. The method of generating social relationship data for a generative confrontation network as recited in claim 1, wherein: in step D, the expressions of the training generator and the discriminator are:
The generator G and the discriminator D are optimized by alternately maximizing and minimizing an objective function V (G; D), where V represents a user node information matrix and V ═ V { (V } 1,v2,...,vn},And All represent mathematical expectations, v to p true(·|vi) Representing user nodes v from a user node information matrix i,v~G(·|vi;θG) Representing user nodes v from a generator i,D(v;vc;θD) Denotes a discriminator, D (v; v. of c;θD) V in (1) cRepresenting other user nodes than V in V, θ GRepresenting the weight, θ, of the first graph neural network in the generator DRepresents weights of a second graph neural network in the discriminator, and ptrue(v) Representing a probability density function of a user node information matrix obeyed by a user node v, and dv represents the differential of v; the generator G and the discriminator D are optimized by continuously updating the weights of the respective graph neural networks through iteration until the discriminator D judges that the pseudo-relation data generated by the generator G and the relation data have the same characteristic structure.
5. The method of generating social relationship data for a generative confrontation network as recited in claim 4, wherein: in the iterative process of the generator G and the discriminator D, the discriminator D receives a positive sample of the user node information matrix and a negative sample of the generator G for training, and the training process is carried out through a first gradient Second graph neural network weights θ for discriminator D DUpdating to optimize an objective function Generator G passes through a second gradient First graph neural network weights θ for generator G GUpdating to optimize an objective function
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010104903.XA CN111415265A (en) | 2020-02-20 | 2020-02-20 | Social relationship data generation method of generative confrontation network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010104903.XA CN111415265A (en) | 2020-02-20 | 2020-02-20 | Social relationship data generation method of generative confrontation network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111415265A true CN111415265A (en) | 2020-07-14 |
Family
ID=71490825
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010104903.XA Pending CN111415265A (en) | 2020-02-20 | 2020-02-20 | Social relationship data generation method of generative confrontation network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111415265A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112465006A (en) * | 2020-11-24 | 2021-03-09 | 中国人民解放军海军航空大学 | Graph neural network target tracking method and device |
CN113343123A (en) * | 2021-06-21 | 2021-09-03 | 中国科学技术大学 | Training method and detection method for generating confrontation multiple relation graph network |
-
2020
- 2020-02-20 CN CN202010104903.XA patent/CN111415265A/en active Pending
Non-Patent Citations (1)
Title |
---|
王坤峰等: ""生成式对抗网络GAN的研究进展与展望"", 《自动化学报》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112465006A (en) * | 2020-11-24 | 2021-03-09 | 中国人民解放军海军航空大学 | Graph neural network target tracking method and device |
CN112465006B (en) * | 2020-11-24 | 2022-08-05 | 中国人民解放军海军航空大学 | Target tracking method and device for graph neural network |
CN113343123A (en) * | 2021-06-21 | 2021-09-03 | 中国科学技术大学 | Training method and detection method for generating confrontation multiple relation graph network |
CN113343123B (en) * | 2021-06-21 | 2022-09-09 | 中国科学技术大学 | Training method and detection method for generating confrontation multiple relation graph network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109948029B (en) | Neural network self-adaptive depth Hash image searching method | |
CN110262233B (en) | Optimization method for technological parameters of magnetic control film plating instrument | |
CN112685504B (en) | Production process-oriented distributed migration chart learning method | |
CN111709534A (en) | Federal learning method, device, equipment and medium based on evolution calculation | |
CN111415265A (en) | Social relationship data generation method of generative confrontation network | |
CN109298930A (en) | A kind of cloud workflow schedule method and device based on multiple-objection optimization | |
Zhou et al. | Hybrid social spider optimization algorithm with differential mutation operator for the job-shop scheduling problem. | |
CN110688484B (en) | Microblog sensitive event speech detection method based on unbalanced Bayesian classification | |
CN115775026A (en) | Federated learning method based on organization similarity | |
CN107194469A (en) | Network reconstruction method based on time series data and stochastic gradient descent method | |
CN111126560A (en) | Method for optimizing BP neural network based on cloud genetic algorithm | |
CN107590538B (en) | Danger source identification method based on online sequence learning machine | |
CN113379536A (en) | Default probability prediction method for optimizing recurrent neural network based on gravity search algorithm | |
Sidajaya et al. | Neural network approach to the simulation of entangled states with one bit of communication | |
Fortier et al. | Learning Bayesian classifiers using overlapping swarm intelligence | |
Zhuang et al. | A Robust Computerized Adaptive Testing Approach in Educational Question Retrieval | |
CN109697511B (en) | Data reasoning method and device and computer equipment | |
CN115984025A (en) | Influence propagation estimation method and system based on deep learning graph network model | |
CN115022231A (en) | Optimal path planning method and system based on deep reinforcement learning | |
Altundogan et al. | Genetic algorithm based fuzzy cognitive map concept relationship determination and sigmoid configuration | |
CN114742292A (en) | Knowledge tracking process-oriented two-state co-evolution method for predicting future performance of students | |
CN112907004B (en) | Learning planning method, device and computer storage medium | |
CN117494119B (en) | Model construction method and device based on group intelligent optimization algorithm | |
CN107292321A (en) | Method and apparatus for obtaining model | |
Bevilacqua et al. | Bayesian Gene Regulatory Network Inference Optimization by means of Genetic Algorithms. |
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: 20200714 |