CN111177384B - Multi-mark Chinese emotion marking method based on global and local mark correlation - Google Patents

Multi-mark Chinese emotion marking method based on global and local mark correlation Download PDF

Info

Publication number
CN111177384B
CN111177384B CN201911361911.6A CN201911361911A CN111177384B CN 111177384 B CN111177384 B CN 111177384B CN 201911361911 A CN201911361911 A CN 201911361911A CN 111177384 B CN111177384 B CN 111177384B
Authority
CN
China
Prior art keywords
mark
correlation
local
global
label
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.)
Active
Application number
CN201911361911.6A
Other languages
Chinese (zh)
Other versions
CN111177384A (en
Inventor
贾修一
朱赛赛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Science and Technology
Original Assignee
Nanjing University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Science and Technology filed Critical Nanjing University of Science and Technology
Priority to CN201911361911.6A priority Critical patent/CN111177384B/en
Publication of CN111177384A publication Critical patent/CN111177384A/en
Application granted granted Critical
Publication of CN111177384B publication Critical patent/CN111177384B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Databases & Information Systems (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a multi-mark Chinese emotion labeling method based on global and local mark correlation, which comprises the following steps: measuring global marker correlation between markers on the model parameter matrix by using a Pearson correlation coefficient; clustering the training set into m clusters by using a k-means clustering method, and calculating a local mark correlation matrix among marks through mark co-occurrence times; optimizing the objective function by using a quasi-Newton descent method L-BFGS algorithm, and solving the optimal value of a model parameter matrix; predicting the test example by using the trained linear regression model; the labeled set of samples is selected according to a threshold. The invention applies global and local mark correlation to multi-mark learning simultaneously, provides a multi-mark Chinese emotion marking method based on global and local mark correlation, and the method performance is superior to the prior method.

Description

Multi-mark Chinese emotion marking method based on global and local mark correlation
Technical Field
The invention relates to an emotion labeling technology, in particular to a multi-mark Chinese emotion labeling method based on global and local mark correlation.
Background
In the field of machine learning, traditional supervised learning is typically used to handle a classification problem where one example relates to only one label. However, objects in the real world tend to have ambiguity, and one object may be associated with multiple tags at the same time. Considering that a plurality of labels have certain correlation, if the multi-label problem is simply regarded as a plurality of single-label problems, the correlation information can not be effectively utilized. Thus, a multi-label learning framework is proposed for solving the problem of one example relating to multiple labels simultaneously. The goal of multi-label learning is to learn a model from training data that assigns a suitable set of labels to unseen examples.
The most straightforward and simplest approach to solving the multi-label learning problem is to decompose it into multiple single-label classification problems, and then train a two-class model for each label directly using traditional classification algorithms. However, this approach completely ignores the correlation that exists between different labels, and this information can provide additional useful information for the multi-label learning process. For example, in automatic annotation of images, when we know that a picture is annotated as "car", then it is very likely to be annotated as "road"; alternatively, in text classification, when a story is classified as "military" then the story is unlikely to be classified as "entertainment". In addition, many studies in recent years have shown that the classification performance of the multi-label learning algorithm can be significantly improved by using the label correlation performance. Therefore, in multi-label learning, it has become one of the most attractive research directions to study how to effectively explore and exploit label correlations.
To date, many multi-label learning algorithms based on the use of second or higher order label correlations have been proposed by scholars. But most of these algorithms use global tag dependencies, which they consider the inter-tag dependencies to be shared by all examples. However, considering only such global tag dependencies makes it difficult to correctly capture some local dependencies shared by only some of the examples. For example, in the text classification, the label "Washington" and the label "President" are highly relevant in the articles of the people biographical class, whereas in the articles of the geographical class, the label "Washington" is usually relevant to the label "City". To address this problem, much research has been devoted to how to exploit local marker correlations in multi-marker learning. In these studies, most of them use some clustering algorithm to estimate different local marker correlations shared by examples in different clusters. However, the estimation of local marker relevance is highly dependent on the accuracy of the clustering. Poor clustering results may lead to inaccurate estimates of local marker correlations and may even destroy some apparent global marker correlations.
Disclosure of Invention
The invention aims to provide a multi-mark Chinese emotion marking method based on global and local mark correlation,
the technical solution for realizing the purpose of the invention is as follows: a multi-mark Chinese emotion labeling method based on global and local mark correlation comprises the following steps:
step 1, measuring global mark correlation between marks on a model parameter matrix by utilizing a Pearson correlation coefficient;
step 2, clustering the training set into m clusters by using a k-means clustering method, and calculating a local mark correlation matrix among marks according to mark co-occurrence times;
step 3, optimizing the target function by using a quasi-Newton descent method L-BFGS algorithm, and solving an optimal value of a model parameter matrix;
step 4, predicting the test sample by using the trained linear regression model;
and 5, selecting a mark set of the sample according to a threshold value.
Compared with the prior art, the invention has the following remarkable advantages: the global and local mark correlation is simultaneously applied to multi-mark learning, the multi-mark Chinese emotion marking method based on the global and local mark correlation is provided, and experiments on different data sets show that the method provided by the invention has better performance.
Drawings
FIG. 1 is a flow chart of a multi-marker Chinese emotion labeling method based on global and local marker correlation.
Detailed Description
As shown in FIG. 1, a multi-label Chinese emotion labeling method based on global and local label correlation includes the following steps:
step 1, measuring global mark correlation between marks on a model parameter matrix W by utilizing a Pearson correlation coefficient;
step 2, clustering the training set into m clusters by using a k-means clustering method, and calculating a local mark correlation matrix R among marks according to mark co-occurrence times;
step 3, optimizing the target function by using a quasi-Newton descent method L-BFGS algorithm, and solving the optimal value of a model parameter matrix W;
step 4, predicting the test sample by using the trained linear regression model;
and 5, selecting a mark set of the sample according to a threshold value.
Furthermore, the correlation between columns of the Pearson correlation coefficient measurement model parameter matrix W is used as the global mark correlation between corresponding marks to be added into the model, and the specific formula is as follows:
Figure BDA0002335181380000031
where ρ (W) j ，W l ) Represents the pearson correlation coefficient between the jth and ith markers:
Figure BDA0002335181380000032
Figure BDA0002335181380000033
is the mean of the elements in column j of W, W jk Is the kth element of the jth column in W;
Figure BDA0002335181380000034
is the mean of the elements in column I of W, W lk Is the kth element of column l in W; q is the number of markers;
when rho (W) j ，W l ) When > 0, it means that the label j and the label l are positively correlated, and when ρ (W) j ，W l ) < 0 indicates that the mark j and the mark l are negatively correlated, when ρ (W) j ，W l ) A value of =0 indicates that the mark j and the mark l are not correlated; l ρ (W) j ，W l ) A larger value of | indicates a larger degree of correlation; dis (W) j ，W l ) Representing the distance between two parameter vectors.
Further, the square Dis (W) of Euclidean distance is adopted j ，W l )＝||W jk -W lk || 2 To measure similarity.
Further, clustering the training set into m clusters by using a k-means clustering method, and calculating a local mark correlation matrix R among marks according to mark co-occurrence times, wherein the specific steps are as follows: dividing training data into m clusters by a clustering method, and calculating the correlation between marks in each cluster; by X c Denotes the c-th cluster g c Example of (1), Y c Is a mark matrix corresponding to the mark matrix; selecting a standard k-means clustering algorithm to divide training data; assuming similar samples have similar label sets, clustering is performed on the label space; local marker correlations are then calculated using the number of marker co-occurrencesSex; in addition, the number of samples in each cluster after clustering may be different, so that the confidence of the correlation of the local marker obtained in each cluster is in direct proportion to the number of samples in the cluster where the local marker is located; the local marker correlation may be expressed as:
Figure BDA0002335181380000035
wherein R is c The tag correlation matrices representing examples in the c-th cluster, and the similarity between the jth tag and the ith tag in each correlation matrix:
Figure BDA0002335181380000041
y j is Y c In the jth column of (1), when pi is true,
Figure BDA0002335181380000042
otherwise
Figure BDA0002335181380000043
Figure BDA0002335181380000044
Is shown in the mark matrix Y c The number of times that the elements in the j-th column and the l-th column are not 0 at the same time; | g c | is the total number of instances in the c-th cluster; when in use
Figure BDA0002335181380000045
When R is cjl ＝0。
Further, considering both global and local tag correlation, a new objective function is established as follows:
Figure BDA0002335181380000046
x i represents the ith sample; regularization where the second term is used to prevent overfittingItem, the third item can take into account global marker correlations during model training, the last item is to take into account local marker correlations in different examples when training the model; lambda 1 、λ 2 And λ 3 And optimizing the objective function by using a quasi-Newton descent method L-BFGS algorithm to obtain an optimal value of the model parameter matrix W.
Further, the linear regression model trained in the above steps is used for predicting the test sample, and the specific steps are as follows:
the output model learned by the steps is as follows:
Figure BDA0002335181380000047
for a test sample, the predicted value of the sample on each mark can be obtained after the test sample is input into the model.
Further, selecting a mark set of the sample according to a threshold value;
and comparing the prediction result of the sample on each mark obtained in the last step with a set threshold value, wherein the sample contains the mark if the prediction result is larger than the threshold value, and otherwise, the sample does not contain the mark. The size of the threshold is here set to 0.5. Finally, five commonly used multi-marker evaluation indexes are used to evaluate the performance of each comparison algorithm, namely Hamming loss, ranking loss, one error, coverage and Average Precision.
The present invention will be described in detail with reference to examples.
Examples
With reference to fig. 1, a multi-tag chinese emotion labeling method based on global and local tag correlations includes the following steps:
step 1, measuring the correlation among columns of a model parameter matrix W by utilizing a Pearson correlation coefficient, and adding the correlation into a model as the global mark correlation among corresponding marks, wherein the specific process is as follows:
step S100, ρ (W) j ，W l ) Represents the pearson correlation coefficient between the jth and ith markers:
Figure BDA0002335181380000051
when rho (W) j ，W l ) When > 0, it means that the label j and the label l are positively correlated, and when ρ (W) j ，W l ) < 0 indicates that the mark j and the mark l are negatively correlated, when ρ (W) j ，W l ) And 0 indicates that the mark j and the mark l are not related. L ρ (W) j ，W l ) A larger value of | indicates a larger degree of correlation.
Step S101, adding the global marker correlation into the model to obtain:
Figure BDA0002335181380000052
wherein Dis (W) j ，W l ) Which represents the distance between two parameter vectors, in the case of euclidean distances,
Figure BDA0002335181380000053
for convenience, we use the squared Dis (W) of Euclidean distance j ，W l )＝||W jk -W lk || 2 To measure similarity.
Step 2, clustering the training set into m clusters by using a k-means clustering method, and calculating a local mark correlation matrix R among marks according to mark co-occurrence times; the specific process is as follows:
and step S200, dividing the training data into m clusters by a k-means clustering method, and respectively calculating the correlation between the marks in each cluster. By X i Represents the ith cluster g i Example of (1), Y i Is the corresponding label matrix, n is the number of samples of all clusters, and it is assumed that similar samples have similar label sets, so we perform clustering on the label space. The number of marker co-occurrences is then used to calculate the local marker correlation. In addition, the number of samples in each cluster after clustering may be different, so that the confidence of the local marker correlation obtained in each cluster is proportional to the number of samples in the cluster where the local marker correlation is located. The local marker correlation may be expressed as:
Figure BDA0002335181380000054
step S201, R c A label correlation matrix representing examples in the c-th cluster, and the similarity between the jth label and the ith label in each correlation matrix:
Figure BDA0002335181380000061
y j is Y c Column j in (1), when pi is true,
Figure BDA0002335181380000062
otherwise
Figure BDA0002335181380000063
Figure BDA0002335181380000064
Is shown in the mark matrix Y c The number of times that the elements in the j-th column and the l-th column are not 0 at the same time; | g c L is the total number of instances in the c-th cluster; when in use
Figure BDA0002335181380000065
To, R cjl ＝0。
Step 3, optimizing the target function by using a quasi-Newton descent method L-BFGS algorithm, and solving the optimal value of a model parameter matrix W; the specific process is as follows:
in step S300, the objective function of the algorithm is as follows:
Figure BDA0002335181380000066
wherein the second term is used as a regularization term to prevent overfitting, the third term can account for global marker correlations during model training, and the last term is to account for differences in training the modelLocal marker relevance in the example. Lambda [ alpha ] 1 ，λ 2 And λ 3 Is a balance factor.
S301, optimizing a target function by using a quasi-Newton descent method L-BFGS algorithm, and solving an optimal value of a model parameter matrix W;
step 4, predicting the test sample by using the trained linear regression model; the specific process is as follows:
the output model can be learned in step 3:
Figure BDA0002335181380000067
for a test sample, the output value of the sample on each mark can be obtained after the test sample is input into the model.
And 5, selecting a mark set of the sample according to a threshold, specifically:
and (5) comparing the prediction result of the sample on each mark predicted in the step (4) with a set threshold value, wherein the sample contains the mark if the prediction result is larger than the threshold value, and otherwise, the sample does not contain the mark. Here the size of the threshold is set to 0.5, resulting in a marked set of samples.
Five commonly used multi-marker evaluation indices were used to evaluate the performance of each comparison algorithm, hamming loss, ranking loss, one error, coverage, and Average Precision, respectively.
The invention provides a novel and simpler multi-label classification algorithm which simultaneously utilizes global and local label correlation. When utilizing global marker correlation, the present invention uses Pearson correlation coefficients to obtain different types of correlations between any two markers, including positive, negative, and uncorrelated. For example, if a picture is labeled "penguin" and "glacier," it is likely to be labeled "south pole," so we can say that the label "south pole" is positively correlated with "penguin" and "glacier"; if an article is marked as "war" and "disaster," it is unlikely to be marked as "entertainment," and so the marking "entertainment" can be seen as negatively correlated to "war" and "disaster. Most existing work does not take into account the negative correlation of labels when using label correlation, and in fact, the negative correlation between labels can also provide useful information for multi-label learning. Experimental results show that the method has better performance than other methods and has stronger multi-label learning ability.

Claims (5)

1. A multi-mark Chinese emotion labeling method based on global and local mark correlation is characterized by comprising the following steps:
step 1, measuring global mark correlation between marks on a model parameter matrix by utilizing a Pearson correlation coefficient;
the correlation between columns of the Pearson correlation coefficient measurement model parameter matrix W is used as the global mark correlation between corresponding marks to be added into the model, and the specific formula is as follows:
Figure FDA0003923360670000011
where ρ (W) j ,W l ) Represents the pearson correlation coefficient between the jth and ith markers:
Figure FDA0003923360670000012
Figure FDA0003923360670000013
is the mean of the elements in column j of W, W jk Is the kth element of the jth column in W;
Figure FDA0003923360670000014
is the mean of the elements in column I of W, W lk Is the kth element in the l column in W, and q is the number of marks;
when rho (W) j ,W l ) When > 0, it means that the label j and the label l are positively correlated, and when ρ (W) j ,W l ) < 0 indicates a mark j andthe label l is negatively correlated when p (W) j ,W l ) =0 indicates that the mark j and the mark l are not correlated; dis (W) j ,W l ) Representing the distance between two parameter vectors;
step 2, clustering the training set into m clusters by using a k-means clustering method, and calculating a local mark correlation matrix R among marks according to mark co-occurrence times, wherein the method comprises the following specific steps:
dividing the training data into m clusters by a clustering method, and calculating the correlation between the marks in each cluster; by X i Represents the ith cluster g i Example of (1), Y i Is the corresponding mark matrix, n is the number of samples of all clusters; selecting a k-means clustering algorithm to divide training data; clustering on the label space; then calculating the local marker correlation by using the marker co-occurrence times; making the confidence of the local marker correlation obtained in each cluster in direct proportion to the number of samples in the cluster where the local marker correlation is located; the local marker correlation is expressed as:
Figure FDA0003923360670000015
wherein R is i A tag correlation matrix representing an example in the ith cluster, and the similarity between the jth tag and the ith tag in each correlation matrix:
Figure FDA0003923360670000021
y j is Y i Column j in (1), when pi is true,
Figure FDA0003923360670000022
otherwise
Figure FDA0003923360670000023
Is shown in the mark matrix Y i The number of times that the elements in the j-th column and the l-th column are not 0 at the same time; | g i I is the ithTotal number of instances in the cluster; when the temperature is higher than the set temperature
Figure FDA0003923360670000024
When R is ijl ＝0；
Step 3, considering the global and local mark correlation, establishing an objective function as follows:
Figure FDA0003923360670000025
x i denotes the ith sample, λ 1 、λ 2 And λ 3 Is a balance factor;
optimizing the target function by using a quasi-Newton descent method L-BFGS algorithm, and solving the optimal value of a model parameter matrix W;
step 4, predicting the test sample by using the trained linear regression model;
and 5, selecting a mark set of the sample according to the threshold value.
2. The multi-tag Chinese emotion markup method of claim 1, wherein Dis (W) is a set of words in which the words are associated with different tags j ,W l )＝||W jk -W lk || 2 I.e. the square of the euclidean distance is used to measure the similarity.
3. The method for labeling Chinese emotion of multiple labels based on global and local label correlation as claimed in claim 1, wherein the linear regression model trained in the above steps is used to predict the test sample, and the specific steps are as follows:
the output model learned in the above steps is:
Figure FDA0003923360670000026
for a test sample, the predicted value of the sample on each mark can be obtained after the test sample is input into the model.
4. The multi-label Chinese emotion labeling method based on global and local label correlation as claimed in claim 3, wherein the label set of the samples is selected according to a threshold, specifically: and comparing the prediction result of the sample on each mark obtained in the last step with a set threshold value, wherein the fact that the sample contains the mark is indicated by the fact that the sample is larger than the threshold value, and otherwise, the sample does not contain the mark.
5. The method for multi-marker Chinese emotion markup based on global and local marker correlations as claimed in claim 4, wherein the threshold size is set to 0.5.
CN201911361911.6A 2019-12-25 2019-12-25 Multi-mark Chinese emotion marking method based on global and local mark correlation Active CN111177384B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911361911.6A CN111177384B (en) 2019-12-25 2019-12-25 Multi-mark Chinese emotion marking method based on global and local mark correlation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911361911.6A CN111177384B (en) 2019-12-25 2019-12-25 Multi-mark Chinese emotion marking method based on global and local mark correlation

Publications (2)

Publication Number Publication Date
CN111177384A CN111177384A (en) 2020-05-19
CN111177384B true CN111177384B (en) 2023-01-20

Family

ID=70657477

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911361911.6A Active CN111177384B (en) 2019-12-25 2019-12-25 Multi-mark Chinese emotion marking method based on global and local mark correlation

Country Status (1)

Country Link
CN (1) CN111177384B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113269226B (en) * 2021-04-14 2022-09-23 南京大学 Picture selection labeling method based on local and global information

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103455578A (en) * 2013-08-23 2013-12-18 华南师范大学 Association rule and bi-clustering-based airline customer data mining method
CN107092591A (en) * 2017-03-30 2017-08-25 南京理工大学 Multiple labeling Chinese emotional reaction categorization method based on correlation rule
CN107391492A (en) * 2017-08-04 2017-11-24 南京理工大学 Indicia distribution Chinese emotion Forecasting Methodology based on fractional sample correlation

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103455578A (en) * 2013-08-23 2013-12-18 华南师范大学 Association rule and bi-clustering-based airline customer data mining method
CN107092591A (en) * 2017-03-30 2017-08-25 南京理工大学 Multiple labeling Chinese emotional reaction categorization method based on correlation rule
CN107391492A (en) * 2017-08-04 2017-11-24 南京理工大学 Indicia distribution Chinese emotion Forecasting Methodology based on fractional sample correlation

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
标记分布学习中目标函数的选择;赵权等;《http://www.cnki.net/kcms/detail/11.5602.TP.20160513.1434.006.html》;20160513;第708-719页 *

Also Published As

Publication number Publication date
CN111177384A (en) 2020-05-19

Similar Documents

Publication Publication Date Title
CN107085585B (en) Accurate tag relevance prediction for image search
CN105117429B (en) Scene image mask method based on Active Learning and multi-tag multi-instance learning
US20170236032A1 (en) Accurate tag relevance prediction for image search
CN106095829A (en) Cross-media retrieval method based on degree of depth study with the study of concordance expression of space
CN109376796A (en) Image classification method based on active semi-supervised learning
CN112541355A (en) Few-sample named entity identification method and system with entity boundary class decoupling
CN106886601A (en) A kind of Cross-modality searching algorithm based on the study of subspace vehicle mixing
CN110569982A (en) Active sampling method based on meta-learning
CN112132186A (en) Multi-label classification method with partial deletion and unknown class labels
CN110009017A (en) A kind of multi-angle of view multiple labeling classification method based on the study of visual angle generic character
CN107330448A (en) A kind of combination learning method based on mark covariance and multiple labeling classification
CN113705570A (en) Few-sample target detection method based on deep learning
Zhu et al. A bug or a suggestion? an automatic way to label issues
Vezhnevets et al. Associative embeddings for large-scale knowledge transfer with self-assessment
CN111177384B (en) Multi-mark Chinese emotion marking method based on global and local mark correlation
US11829442B2 (en) Methods and systems for efficient batch active learning of a deep neural network
CN109993188B (en) Data tag identification method, behavior identification method and device
CN115527083B (en) Image annotation method and device and electronic equipment
CN116630694A (en) Target classification method and system for partial multi-label images and electronic equipment
CN107578069B (en) Image multi-scale automatic labeling method
CN112308097A (en) Sample identification method and device
Blount et al. Comparison of three individual identification algorithms for sperm whales (physeter macrocephalus) after automated detection
CN111723301B (en) Attention relation identification and labeling method based on hierarchical theme preference semantic matrix
CN107391492A (en) Indicia distribution Chinese emotion Forecasting Methodology based on fractional sample correlation
CN113420821A (en) Multi-label learning method based on local correlation of labels and features

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