Rethinking Fusion: Disentangled Learning of Shared and Modality-Specific Information for Stance Detection
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866914291062407168 |
|---|---|
| author | Xie, Zhiyu Niu, Fuqiang Dai, Genan Wang, Qianlong Dong, Li Zhang, Bowen Huang, Hu |
| author_facet | Xie, Zhiyu Niu, Fuqiang Dai, Genan Wang, Qianlong Dong, Li Zhang, Bowen Huang, Hu |
| contents | Multi-modal stance detection (MSD) aims to determine an author's stance toward a given target using both textual and visual content. While recent methods leverage multi-modal fusion and prompt-based learning, most fail to distinguish between modality-specific signals and cross-modal evidence, leading to suboptimal performance. We propose DiME (Disentangled Multi-modal Experts), a novel architecture that explicitly separates stance information into textual-dominant, visual-dominant, and cross-modal shared components. DiME first uses a target-aware Chain-of-Thought prompt to generate reasoning-guided textual input. Then, dual encoders extract modality features, which are processed by three expert modules with specialized loss functions: contrastive learning for modality-specific experts and cosine alignment for shared representation learning. A gating network adaptively fuses expert outputs for final prediction. Experiments on four benchmark datasets show that DiME consistently outperforms strong unimodal and multi-modal baselines under both in-target and zero-shot settings. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_21675 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Rethinking Fusion: Disentangled Learning of Shared and Modality-Specific Information for Stance Detection Xie, Zhiyu Niu, Fuqiang Dai, Genan Wang, Qianlong Dong, Li Zhang, Bowen Huang, Hu Multimedia Multi-modal stance detection (MSD) aims to determine an author's stance toward a given target using both textual and visual content. While recent methods leverage multi-modal fusion and prompt-based learning, most fail to distinguish between modality-specific signals and cross-modal evidence, leading to suboptimal performance. We propose DiME (Disentangled Multi-modal Experts), a novel architecture that explicitly separates stance information into textual-dominant, visual-dominant, and cross-modal shared components. DiME first uses a target-aware Chain-of-Thought prompt to generate reasoning-guided textual input. Then, dual encoders extract modality features, which are processed by three expert modules with specialized loss functions: contrastive learning for modality-specific experts and cosine alignment for shared representation learning. A gating network adaptively fuses expert outputs for final prediction. Experiments on four benchmark datasets show that DiME consistently outperforms strong unimodal and multi-modal baselines under both in-target and zero-shot settings. |
| title | Rethinking Fusion: Disentangled Learning of Shared and Modality-Specific Information for Stance Detection |
| topic | Multimedia |
| url | https://arxiv.org/abs/2601.21675 |