Method for constructing joint embedding model based on regular flow automatic encoder

A technology of autoencoder and construction method, which is applied in the field of joint embedding model construction, can solve problems such as calculation and similarity, and achieve good performance

Inactive Publication Date: 2020-05-15
SUN YAT SEN UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Because the above method only learns the low-dimensional expression of nodes, but not the expression of attributes, the similarity between them cannot be calculated by simple inner product

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Method for constructing joint embedding model based on regular flow automatic encoder
  • Method for constructing joint embedding model based on regular flow automatic encoder
  • Method for constructing joint embedding model based on regular flow automatic encoder

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0054] A method for constructing a joint embedding model based on a regular flow autoencoder, comprising the following steps:

[0055]Step 1. Define symbols and targets: Assuming that the number of nodes and attributes in the network are N and F respectively, then an attribute network can be represented by two matrices: adjacency matrix and the attribute matrix Each row of A represents the out-degree of the corresponding node, and each row of B represents the attributes of the corresponding node; so the goal is to pass the established model To learn the expression of this network in low dimension: node embedding matrix and the attribute embedding matrix Among them, D represents the dimension of the low-dimensional space; each row of N represents the representation of the corresponding node in the low-dimensional space, and each row of V represents the representation of the corresponding attribute in the low-dimensional space; note that the low-dimensional space mentio...

Embodiment 2

[0108] The present embodiment provides to build a neural network for the model of the present invention; including the following steps:

[0109] first:

[0110]

[0111]

[0112]

[0113]

[0114] where ReLU is the activation function, is a symmetric normalized adjacency matrix, and D is a degree matrix.

[0115] H=tanh(B T W 0 +b 0 )

[0116]

[0117]

[0118] h v =HW 3 +b 3

[0119] Where tanh is the activation function, W, b are the weight matrix and bias of the neural network, respectively. Note that the above two H and W are different, and are omitted here only for brevity.

[0120] Then sample:

[0121]

[0122]

[0123] For each function of the regular stream, we implement it like this:

[0124] h 1 =elu(Z (t-1) (M⊙W 0 )+b 0 )+h Z

[0125] h 2 =elu(H 1 (M⊙W 1 )+b 1 )

[0126]

[0127]

[0128]

[0129]

[0130] Among them, ELU and sigmoid are activation functions. Multiply the corresponding elements of M and the...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention belongs to the technical field of machine learning and artificial intelligence, and particularly relates to a method for constructing a joint embedding model based on a regular flow automatic encoder. The invention aims to create a new attribute network low-dimensional embedded model, and the model not only solves the defects of the prior art, but also has better performance than theprevious work. Therefore, the invention provides the F-CAN and a joint embedding model based on the regular flow. Different from the conventional algorithm, although the model also takes the topological structure and the attribute information of the attribute network as the input, the model not only outputs the low-dimensional expression of the nodes, but also outputs the low-dimensional expression of the attributes, and the low-dimensional expression of the attributes and the low-dimensional expression are located in the same space. Besides, aiming at the problem that posterior probability hypothesis in a joint embedding model is too simple, the invention provides a double-regular-flow method, parameters of the model are increased, the posterior probability hypothesis can fit the real posterior probability more easily, and therefore better performance is obtained.

Description

technical field [0001] The invention belongs to the technical field of machine learning and artificial intelligence, and more specifically, relates to a method for building a joint embedding model based on a regular flow autoencoder. Background technique [0002] As the most important network, the attribute network can be seen everywhere in many fields, such as the rapidly developing online social network in recent years, the user network of the e-commerce platform, and the network in the academic field, etc., can be abstracted as an attribute network. After data mining on these networks, a series of commercially valuable applications such as product recommendation, user portrait, and community discovery can be carried out. How to represent the network is crucial to the data mining task. [0003] In order to solve the above problems, some attribute network embedding algorithms are proposed, including mark-aware embedding, accelerated attribute network embedding, semi-superv...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06F30/20G06F17/18
CPCG06F17/18
Inventor 梁上松欧阳卓
Owner SUN YAT SEN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap