A weighted difference loss approach for enhancing multi-label classification

Conventional multi-label classification methods often fail to capture the dynamic relationships and relative intensity shifts between labels, treating them as independent entities. This limitation is particularly detrimental in tasks like sentiment analysis where emotions co-occur in nuanced proport...

Full description

Bibliographic Details
Main Authors: Qiong, Hu, Azmi Murad, Masrah Azrifah, Azman, Azreen, Nasharuddin, Nurul Amelina
Format: Article
Language:English
Published: Nature Research 2025
Online Access:http://psasir.upm.edu.my/id/eprint/120145/
http://psasir.upm.edu.my/id/eprint/120145/1/120145.pdf
_version_ 1848868122592804864
author Qiong, Hu
Azmi Murad, Masrah Azrifah
Azman, Azreen
Nasharuddin, Nurul Amelina
author_facet Qiong, Hu
Azmi Murad, Masrah Azrifah
Azman, Azreen
Nasharuddin, Nurul Amelina
author_sort Qiong, Hu
building UPM Institutional Repository
collection Online Access
description Conventional multi-label classification methods often fail to capture the dynamic relationships and relative intensity shifts between labels, treating them as independent entities. This limitation is particularly detrimental in tasks like sentiment analysis where emotions co-occur in nuanced proportions. To address this, we introduce a novel Weighted Difference Loss (WDL) framework. WDL operates on three core principles: (1) transforming labels into a normalized distribution to model their relative proportions; (2) computing learnable, weighted differences across this distribution to explicitly capture inter-label dynamics and trends; and (3) employing a label-shuffling augmentation to ensure the model learns intrinsic, order-invariant relationships. Our framework not only achieves state-of-the-art performance on four public benchmarks, but more importantly, it substantially improves the recognition of minority classes. This demonstrates the framework’s ability to learn from sparse data by effectively leveraging the underlying label structure, offering a robust, loss-driven alternative to complex architectural modifications.
first_indexed 2025-11-15T14:47:22Z
format Article
id upm-120145
institution Universiti Putra Malaysia
institution_category Local University
language English
last_indexed 2025-11-15T14:47:22Z
publishDate 2025
publisher Nature Research
recordtype eprints
repository_type Digital Repository
spelling upm-1201452025-09-24T01:55:26Z http://psasir.upm.edu.my/id/eprint/120145/ A weighted difference loss approach for enhancing multi-label classification Qiong, Hu Azmi Murad, Masrah Azrifah Azman, Azreen Nasharuddin, Nurul Amelina Conventional multi-label classification methods often fail to capture the dynamic relationships and relative intensity shifts between labels, treating them as independent entities. This limitation is particularly detrimental in tasks like sentiment analysis where emotions co-occur in nuanced proportions. To address this, we introduce a novel Weighted Difference Loss (WDL) framework. WDL operates on three core principles: (1) transforming labels into a normalized distribution to model their relative proportions; (2) computing learnable, weighted differences across this distribution to explicitly capture inter-label dynamics and trends; and (3) employing a label-shuffling augmentation to ensure the model learns intrinsic, order-invariant relationships. Our framework not only achieves state-of-the-art performance on four public benchmarks, but more importantly, it substantially improves the recognition of minority classes. This demonstrates the framework’s ability to learn from sparse data by effectively leveraging the underlying label structure, offering a robust, loss-driven alternative to complex architectural modifications. Nature Research 2025 Article PeerReviewed text en http://psasir.upm.edu.my/id/eprint/120145/1/120145.pdf Qiong, Hu and Azmi Murad, Masrah Azrifah and Azman, Azreen and Nasharuddin, Nurul Amelina (2025) A weighted difference loss approach for enhancing multi-label classification. Scientific Reports, 15 (1). art. no. 25052. pp. 1-13. ISSN 2045-2322 https://www.nature.com/articles/s41598-025-09883-2?error=cookies_not_supported&code=b1a5fc06-d472-4597-8bfb-f0f263aebf2d 10.1038/s41598-025-09883-2
spellingShingle Qiong, Hu
Azmi Murad, Masrah Azrifah
Azman, Azreen
Nasharuddin, Nurul Amelina
A weighted difference loss approach for enhancing multi-label classification
title A weighted difference loss approach for enhancing multi-label classification
title_full A weighted difference loss approach for enhancing multi-label classification
title_fullStr A weighted difference loss approach for enhancing multi-label classification
title_full_unstemmed A weighted difference loss approach for enhancing multi-label classification
title_short A weighted difference loss approach for enhancing multi-label classification
title_sort weighted difference loss approach for enhancing multi-label classification
url http://psasir.upm.edu.my/id/eprint/120145/
http://psasir.upm.edu.my/id/eprint/120145/
http://psasir.upm.edu.my/id/eprint/120145/
http://psasir.upm.edu.my/id/eprint/120145/1/120145.pdf