Deep consensus semantic aware network for partial multi-view incomplete multi-label classification

Abstract Recently, multi-view multi-label classification (MVMLC) has attracted considerable attention, particularly in the computer vision field. However, due to overlooking incomplete views and missing labels caused by data collection limitations and unreliable annotations, many existing methods ha...

Full description

Saved in:
Bibliographic Details
Main Authors: Bo Shao, Yang Xu
Format: Article
Language:English
Published: Springer 2025-05-01
Series:Journal of King Saud University: Computer and Information Sciences
Subjects:
Online Access:https://doi.org/10.1007/s44443-025-00056-9
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849389102042972160
author Bo Shao
Yang Xu
author_facet Bo Shao
Yang Xu
author_sort Bo Shao
collection DOAJ
description Abstract Recently, multi-view multi-label classification (MVMLC) has attracted considerable attention, particularly in the computer vision field. However, due to overlooking incomplete views and missing labels caused by data collection limitations and unreliable annotations, many existing methods have difficulty in extracting consensus semantic information from diverse data sources. To overcome this challenge, we propose a deep consensus semantic aware network (CSA), which can effectively tackle the challenge of consensus extraction. The proposed CSA incorporates an enhanced contrastive learning (ECL) module, which contains cross-view and cross-fusion contrastive learning. The former facilitates the encoder in capturing and aligning consensus semantic information across multiple views, while the latter maximizes the semantic information of view-specific and view-common representations by decreasing potential interference introduced during the fusion process. In addition, CSA integrates a multi-view cycle generative adversarial (MCGA) network, where the generator reconstructs the original data based on the view-common representation, and the discriminator conducts cycle adversarial training to distinguish between original and generated data, which enhances the extraction of consensus information by forcing the generator to generate actual data. Experiments across five datasets highlight the superiority of CSA. On the Pascal07 dataset, CSA outperforms RANK, the second-best method, with a 0.8% increase in Average Precision (AP) and a 0.8% improvement in Area Under the ROC Curve (AUC). Similarly, on the MIRFLICKR dataset, CSA boosts AP by 1.6% and AUC by 0.5% over RANK.
format Article
id doaj-art-0ff9a341067f4f5188c3120ead8da561
institution Kabale University
issn 1319-1578
2213-1248
language English
publishDate 2025-05-01
publisher Springer
record_format Article
series Journal of King Saud University: Computer and Information Sciences
spelling doaj-art-0ff9a341067f4f5188c3120ead8da5612025-08-20T03:42:03ZengSpringerJournal of King Saud University: Computer and Information Sciences1319-15782213-12482025-05-0137411910.1007/s44443-025-00056-9Deep consensus semantic aware network for partial multi-view incomplete multi-label classificationBo Shao0Yang Xu1College of Big Data and Information Engineering, Guizhou UniversityCollege of Big Data and Information Engineering, Guizhou UniversityAbstract Recently, multi-view multi-label classification (MVMLC) has attracted considerable attention, particularly in the computer vision field. However, due to overlooking incomplete views and missing labels caused by data collection limitations and unreliable annotations, many existing methods have difficulty in extracting consensus semantic information from diverse data sources. To overcome this challenge, we propose a deep consensus semantic aware network (CSA), which can effectively tackle the challenge of consensus extraction. The proposed CSA incorporates an enhanced contrastive learning (ECL) module, which contains cross-view and cross-fusion contrastive learning. The former facilitates the encoder in capturing and aligning consensus semantic information across multiple views, while the latter maximizes the semantic information of view-specific and view-common representations by decreasing potential interference introduced during the fusion process. In addition, CSA integrates a multi-view cycle generative adversarial (MCGA) network, where the generator reconstructs the original data based on the view-common representation, and the discriminator conducts cycle adversarial training to distinguish between original and generated data, which enhances the extraction of consensus information by forcing the generator to generate actual data. Experiments across five datasets highlight the superiority of CSA. On the Pascal07 dataset, CSA outperforms RANK, the second-best method, with a 0.8% increase in Average Precision (AP) and a 0.8% improvement in Area Under the ROC Curve (AUC). Similarly, on the MIRFLICKR dataset, CSA boosts AP by 1.6% and AUC by 0.5% over RANK.https://doi.org/10.1007/s44443-025-00056-9Multi-view multi-label classificationConsensus extractionContrastive learningGenerative adversarial networks
spellingShingle Bo Shao
Yang Xu
Deep consensus semantic aware network for partial multi-view incomplete multi-label classification
Journal of King Saud University: Computer and Information Sciences
Multi-view multi-label classification
Consensus extraction
Contrastive learning
Generative adversarial networks
title Deep consensus semantic aware network for partial multi-view incomplete multi-label classification
title_full Deep consensus semantic aware network for partial multi-view incomplete multi-label classification
title_fullStr Deep consensus semantic aware network for partial multi-view incomplete multi-label classification
title_full_unstemmed Deep consensus semantic aware network for partial multi-view incomplete multi-label classification
title_short Deep consensus semantic aware network for partial multi-view incomplete multi-label classification
title_sort deep consensus semantic aware network for partial multi view incomplete multi label classification
topic Multi-view multi-label classification
Consensus extraction
Contrastive learning
Generative adversarial networks
url https://doi.org/10.1007/s44443-025-00056-9
work_keys_str_mv AT boshao deepconsensussemanticawarenetworkforpartialmultiviewincompletemultilabelclassification
AT yangxu deepconsensussemanticawarenetworkforpartialmultiviewincompletemultilabelclassification