REDH-Net: Relation-Enhanced Dynamic Routing and Distribution-Consistent Hierarchical Dual-Head Prediction for Multimodal Sentiment Analysis.
Saved in:
| Title: | REDH-Net: Relation-Enhanced Dynamic Routing and Distribution-Consistent Hierarchical Dual-Head Prediction for Multimodal Sentiment Analysis. |
|---|---|
| Authors: | An, Honghui1 anna011002@163.com, Wang, Liang1 wang_liang@syuct.edu.cn |
| Source: | Engineering Letters. Oct2026, Vol. 34 Issue 10, p4542-4554. 13p. |
| Subjects: | Affective computing, Adaptive routing (Computer network management) |
| Abstract: | Multimodal sentiment analysis (MSA) integrates textual, acoustic, and visual cues to infer sentiment polarity and intensity. Recent dynamic routing and mixture-of-experts methods improve sample-level fusion by assigning adaptive modality weights, but their routing decisions are still often generated mainly from joint representations, with limited explicit modeling of modality-specific informativeness and cross-modal relations. In addition, conventional single-head predictors and loosely coupled dual-head predictors may not sufficiently coordinate coarse-grained polarity discrimination with fine-grained sentiment intensity modeling. To address these issues, this paper proposes REDH-Net, a relation-enhanced dynamic routing and distribution-consistent hierarchical dual-head network for MSA. The routing module estimates variance-based modality informativeness, computes pairwise cross-modal consistency, and adaptively fuses the resulting relation prior with a basic routing branch. The prediction module introduces a seven-class primary head, a binary auxiliary polarity head, and a distribution consistency constraint that aligns the primary-head induced polarity distribution with the auxiliary polarity distribution. Experimental results show that, under the Unaligned setting, REDH-Net achieves relative improvements over the strong EMOE dynamic expert model of 0.82% in ACC2 and 1.46% in ACC7 on CMU-MOSI, and 1.05% in ACC2 and 0.56% in ACC7 on CMU-MOSEI. Further ablation studies, robustness analysis, routing behavior visualization, probability consistency visualization, and qualitative results verify the effectiveness and interpretability of the proposed design. [ABSTRACT FROM AUTHOR] |
| Copyright of Engineering Letters is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
Be the first to leave a comment!
