摘要
多视图偏多标记学习主要处理同时具有多个视图和多个相关标记但标记信息不完全准确的数据.现有多视图偏多标记学习方法大多采用两阶段的方式独立进行标记消歧与多标记分类,其分类性能有待提高.本文提出了一种多视图偏多标记分类与标记消歧联合学习(joint learning of multi-view partial multi-label classification and label disambiguation,JL-MVPML-LD)框架.首先,对多视图特征进行多核融合并考虑不同视图的重要性;其次,自动挖掘实例相关性和标记相关性,并利用它们来促进多视图偏多标记分类和标记消歧的联合学习;最后,采用交替迭代方法进行求解.在3个数据集上27种情况下的实验结果验证了本文方法的有效性.
Multi-view partial multi-label learning mainly aims to deal with data with multi-view features and multiple related labels,but the label information is not completely accurate.Most of the existing methods use a two-stage approach to perform label disambiguation and multi-label classification independently,but these methods are sub-optimal.In this paper,a new learning framework of Joint Learning of Multi-View Partial Multi-Label classification and Label Disambiguation(JL-MVPML-LD)is proposed.Firstly,the multi-view features are fused by using the multi-kernel method and the importance of each view is considered.Secondly,the instance correlation and label correlation are mined and learned,and then used to help the joint learning of multi-view partial multi-label classifier and label disambiguation.Finally,JL-MVPML_LD can be solved by alternating solution method.The experimental results in 27 cases on three datasets verify the effectiveness of the proposed method.
作者
徐远洋
何志芬
刘彬
Xu Yuanyang;He Zhifen;Liu Bin(School of Mathematics and Information Science,Nanchang Hangkong University,Nanchang 330063,China)
出处
《南京师大学报(自然科学版)》
北大核心
2025年第2期74-82,90,共10页
Journal of Nanjing Normal University(Natural Science Edition)
基金
国家自然科学基金项目(62362051)
江西省自然科学基金项目(2023BAB202047).
关键词
多视图偏多标记学习
核空间
标记消歧
标记相关性
multi-view partial multi-label learning
kernel space
label disambiguation
label correlation
