Unlock instant, AI-driven research and patent intelligence for your innovation.

A Multi-label Classification Learning Method Based on Matching Learning

A learning method and multi-label technology, applied in the field of extremely large-scale multi-label classification learning, can solve problems such as not being able to handle long-tail labels well, and achieve the effect of supporting online incremental learning and ensuring parallel learning

Active Publication Date: 2020-12-01
北京东方科诺科技发展有限公司
View PDF12 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

This type of algorithm uses the assumption of low-rank space and cannot handle long-tail labels well.

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
  • A Multi-label Classification Learning Method Based on Matching Learning
  • A Multi-label Classification Learning Method Based on Matching Learning
  • A Multi-label Classification Learning Method Based on Matching Learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0054] This embodiment provides a very large-scale multi-label classification learning method based on matching learning, combining figure 1 As shown, the specific steps are as follows:

[0055] Step 1. Collect user data on the Internet, including user tags.

[0056] Step 2. Extract features from data such as user text and images, and perform feature value calculation. In order to obtain the data set D={(x 1 ,w 1 ,y 1 )...(x n ,w n ,y n )}. Where x is the feature set, w is the corresponding feature value set, and y is the label set.

[0057] Step 3. Randomly sample a mini-batch from the data set for gradient descent mini-batch, and prepare to optimize the parameters of the multi-label model. The specific steps are as follows:

[0058] Step 301. Perform random shuffling on the data set D.

[0059] Step 302, traverse the shuffled data set with a step size M, and generate a mini-batch D at each step m .

[0060] Step 303, for D m Randomly sample N sets of negative la...

Embodiment 2

[0078] This embodiment provides a multi-label classification learning method based on matching learning, including the following steps:

[0079] S1: Collect client data in the Internet, perform feature value calculation on the client data, and obtain a training set D;

[0080] S2: traverse the training set D, and set the negative label set and positive label set in the training set D;

[0081] S3: Calculate the embedded representation E of the feature set in the training set D;

[0082] S4: Calculate the embedded representation Z of the positive label set + and the embedded representation Z of the set of negative labels - ;

[0083] S5: Comparing the embedded representation E and the embedded representation Z + Perform loss calculation to obtain a positive label loss value, for the embedded representation E and the embedded representation Z - Calculate the loss to get the negative label loss value;

[0084] S6: According to the positive label loss value and the negative ...

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 multi-label classification learning method based on matching learning provided by the present invention firstly calculates the eigenvalues ​​of the client data to obtain the training set; sets the positive and negative labels in the training set; then calculates the embedded representation E of the feature set in the training set; calculates Embedded representation Z for positive labels + and negatively labeled embedded representations Z ‑ ; followed by E and Z + For loss calculation, for E and Z ‑ Calculate the loss; according to the positive and negative loss value, use the gradient descent algorithm to train the model on the training set; when the loss value of the training model no longer decreases, the training ends, otherwise train again; finally use the test set to test the training model. The invention considers the adverse effect of long-tail labels on the learning of traditional extremely large-scale multi-label classification models, and proposes a matching learning method using feature sets and label sets. In addition, in order to ensure that the model can be extended to large-scale data sets, the model learning method of the gradient descent algorithm is adopted, which can not only ensure the parallel learning of the model, but also support the online incremental learning of the model.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to an extremely large-scale multi-label classification learning method based on matching learning and neural network models. Background technique [0002] With the rapid development of Web 2.0 technology, a large amount of user-related or even user-generated content has accumulated on the Internet. In a wide variety of massive data, there is a wealth of information that can reflect user characteristics, which is an important data support for user portraits. User profiling has always been an important issue in the research of social computing. User portraits, that is, labeling user information, provide companies with basic information about users, and can help companies quickly find accurate user groups and more extensive feedback information such as user needs. The core task of user portrait is to "label" users. How to use data mining or machine learning to automatically...

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 Patents(China)
IPC IPC(8): G06K9/62
CPCG06F18/214G06F18/2431
Inventor 翟书杰李晨
Owner 北京东方科诺科技发展有限公司