| Item type |
文献 / Documents(1) |
| 公開日 |
2025-10-27 |
| アクセス権 |
|
|
アクセス権 |
embargoed access |
|
アクセス権URI |
http://purl.org/coar/access_right/c_f1cf |
| 資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_6501 |
|
資源タイプ |
journal article |
| 出版社版DOI |
|
|
|
関連識別子 |
https://doi.org/10.1016/j.asoc.2025.113665 |
|
|
関連名称 |
10.1016/j.asoc.2025.113665 |
| 出版タイプ |
|
|
出版タイプ |
AM |
|
出版タイプResource |
http://purl.org/coar/version/c_ab4af688f83e57aa |
| タイトル |
|
|
タイトル |
SAGE-Net : Single-layer augmented gated encoder network for efficient multimodal sentiment analysis |
| 著者 |
Zhou, Jiazheng
康, 鑫
Ding, Weiping
Wang, Linhuang
Ding, Fei
松本, 和幸
Zhang, Chenmeng
Chi, Huiwen
|
| 抄録 |
|
|
内容記述 |
Recently, the field of multimodal sentiment analysis has achieved remarkable progress. However, the increasing computational complexity of existing models poses significant challenges for resource-constrained scenarios. To address these challenges, this study introduces a single-layer augmented gated encoder network (SAGE-Net), a novel lightweight multimodal sentiment-analysis framework. In contrast to conventional multilayer, deeply stacked structures, SAGE-Net only uses a single-layer encoder to retain multimodal feature understanding, significantly reducing computational complexity. To further enhance information interaction, we introduce a single-layer cross-attention mechanism. We extensively experimented with diverse feature-fusion strategies and data-augmentation schemes. SAGE-Net maintained remarkable representational capacity and strong performance with low-complexity. Experimental results on the CMU-MOSI and CMU-MOSEI datasets indicate that SAGE-Net achieves state-of-the-art performance and significantly lowers model size and tuning costs. SAGE-Net is a viable, lightweight solution for multimodal sentiment analysis under resource-constrained, real-world scenarios. |
| キーワード |
|
|
主題 |
Multimodal sentiment analysis |
| キーワード |
|
|
主題 |
Lightweight model |
| キーワード |
|
|
主題 |
Cross-attention mechanism |
| キーワード |
|
|
主題 |
Data augmentation strategies |
| 書誌情報 |
en : Applied Soft Computing
巻 184,
号 B,
p. 113665,
発行日 2025-08-27
|
| 収録物ID |
|
|
収録物識別子タイプ |
PISSN |
|
収録物識別子 |
15684946 |
| 収録物ID |
|
|
収録物識別子タイプ |
EISSN |
|
収録物識別子 |
18729681 |
| 収録物ID |
|
|
収録物識別子タイプ |
NCID |
|
収録物識別子 |
AA11644645 |
| 収録物ID |
|
|
収録物識別子タイプ |
NCID |
|
収録物識別子 |
AA11926126 |
| 出版者 |
|
|
出版者 |
Elsevier |
| 権利情報 |
|
|
権利情報 |
© 2025. This manuscript version is made available under the CC-BY-NC-ND 4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/ |
| EID |
|
|
識別子 |
446049 |
| 言語 |
|
|
言語 |
eng |
asoc_184_B_113665.pdf (444 KB)
