Adaptive feature interaction enhancement network for text classification
Abstract Text classification aims to establish text distinctions, which face difficulty in capturing global text semantics and local details. To address this issue, we propose an Adaptive Feature Interactive Enhancement Network (AFIENet). Specifically, AFIENet uses two branches to model the text glo...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-04-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-95492-y |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849764878379646976 |
|---|---|
| author | Rui Su Shangbing Gao Kefan Zhao Junqiang Zhang |
| author_facet | Rui Su Shangbing Gao Kefan Zhao Junqiang Zhang |
| author_sort | Rui Su |
| collection | DOAJ |
| description | Abstract Text classification aims to establish text distinctions, which face difficulty in capturing global text semantics and local details. To address this issue, we propose an Adaptive Feature Interactive Enhancement Network (AFIENet). Specifically, AFIENet uses two branches to model the text globally and locally. The adaptive segmentation module in the local network can dynamically split the text and capture key phrases, while the global network grasps the overall central semantics. After obtaining the results from the two branches, an interaction gate is designed to evaluate the confidence of the global features and selectively fuse them with the local features effectively. Finally, the interactively enhanced features are re-input into the classifier to improve text classification performance. Experiment results show that our proposed method can effectively enhance the performance of backbone networks such as TextCNN, RNN, and Transformer with fewer parameters. AFIENet achieved an average accuracy of 3.82% and an F1-score of 3.88% improvement across the three datasets when using Transformer as the backbone network. The comparable results to MacBERT that obtained with static word vectors also reflect the applicability of the proposed method. |
| format | Article |
| id | doaj-art-8a80d3ecc14349c7a0008810fb6aa73f |
| institution | DOAJ |
| issn | 2045-2322 |
| language | English |
| publishDate | 2025-04-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Scientific Reports |
| spelling | doaj-art-8a80d3ecc14349c7a0008810fb6aa73f2025-08-20T03:05:01ZengNature PortfolioScientific Reports2045-23222025-04-0115111410.1038/s41598-025-95492-yAdaptive feature interaction enhancement network for text classificationRui Su0Shangbing Gao1Kefan Zhao2Junqiang Zhang3School of Computer and Software Engineering, Huaiyin Institute of TechnologySchool of Computer and Software Engineering, Huaiyin Institute of TechnologySchool of Computer and Software Engineering, Huaiyin Institute of TechnologySchool of Computer and Software Engineering, Huaiyin Institute of TechnologyAbstract Text classification aims to establish text distinctions, which face difficulty in capturing global text semantics and local details. To address this issue, we propose an Adaptive Feature Interactive Enhancement Network (AFIENet). Specifically, AFIENet uses two branches to model the text globally and locally. The adaptive segmentation module in the local network can dynamically split the text and capture key phrases, while the global network grasps the overall central semantics. After obtaining the results from the two branches, an interaction gate is designed to evaluate the confidence of the global features and selectively fuse them with the local features effectively. Finally, the interactively enhanced features are re-input into the classifier to improve text classification performance. Experiment results show that our proposed method can effectively enhance the performance of backbone networks such as TextCNN, RNN, and Transformer with fewer parameters. AFIENet achieved an average accuracy of 3.82% and an F1-score of 3.88% improvement across the three datasets when using Transformer as the backbone network. The comparable results to MacBERT that obtained with static word vectors also reflect the applicability of the proposed method.https://doi.org/10.1038/s41598-025-95492-yNatural language processingText classificationAdaptive feature enhancementInteraction gatePre-training |
| spellingShingle | Rui Su Shangbing Gao Kefan Zhao Junqiang Zhang Adaptive feature interaction enhancement network for text classification Scientific Reports Natural language processing Text classification Adaptive feature enhancement Interaction gate Pre-training |
| title | Adaptive feature interaction enhancement network for text classification |
| title_full | Adaptive feature interaction enhancement network for text classification |
| title_fullStr | Adaptive feature interaction enhancement network for text classification |
| title_full_unstemmed | Adaptive feature interaction enhancement network for text classification |
| title_short | Adaptive feature interaction enhancement network for text classification |
| title_sort | adaptive feature interaction enhancement network for text classification |
| topic | Natural language processing Text classification Adaptive feature enhancement Interaction gate Pre-training |
| url | https://doi.org/10.1038/s41598-025-95492-y |
| work_keys_str_mv | AT ruisu adaptivefeatureinteractionenhancementnetworkfortextclassification AT shangbinggao adaptivefeatureinteractionenhancementnetworkfortextclassification AT kefanzhao adaptivefeatureinteractionenhancementnetworkfortextclassification AT junqiangzhang adaptivefeatureinteractionenhancementnetworkfortextclassification |