Incorporating manifold regularization into adaptive label thresholding algorithm for online semi-supervised multi-label classification

Online multi-label classification aims to assign multiple appropriate categories to arriving instances. Currently, most approaches focus on supervised scenarios where instances are all labeled. The adaptive label thresholding approach, for example, can dynamically adjust the label threshold for labe...

Full description

Saved in:
Bibliographic Details
Published in:Evolving systems Vol. 16; no. 4; p. 124
Main Authors: Fang, Nannan, Zhai, Tingting
Format: Journal Article
Language:English
Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.12.2025
Springer Nature B.V
Subjects:
Adaptive algorithms
Algorithms
Artificial Intelligence
Classification
Complex Systems
Complexity
Datasets
Distance learning
Engineering
Feature selection
Labels
Manifolds
Optimization
Original Paper
Regularization
Regularization methods
Online multi-label classification
Adaptive label thresholding
Online semi-supervised learning
Manifold regularization
ISSN:1868-6478, 1868-6486
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Online multi-label classification aims to assign multiple appropriate categories to arriving instances. Currently, most approaches focus on supervised scenarios where instances are all labeled. The adaptive label thresholding approach, for example, can dynamically adjust the label threshold for labeled instances to distinguish relevant and irrelevant labels. However, in reality, labeling every instance is time-consuming and resource-intensive, and typically only a small fraction of instances are labeled. To address this challenge, we propose an innovative online semi-supervised multi-label classification algorithm. We design a novel online manifold regularization term that can leverage the similarity between labeled and unlabeled instances for prediction, and integrate it with the optimization objective of the adaptive label thresholding algorithm. The Online Gradient Descent method is employed to solve the improved optimization problem. We further extend the model to handle nonlinear classification problems using Mercer kernels, and derive detailed coefficient update formulas for support vectors. Experiments are conducted on eight open datasets spanning text, image, music, and biology domains. Performance is evaluated using seven common multi-label classification metrics. Paired t-tests on F1-measure results show that our proposed algorithm achieves statistically significant improvements over the second-best method on six datasets, confirming its effectiveness in semi-supervised multi-label classification tasks.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1868-6478
1868-6486
DOI:10.1007/s12530-025-09752-3