DGC-Net: Cross-Modal Dynamic Routing with Global-Guided Compensation in Multimodal Sentiment Analysis
| Source: arXiv AI
Tags: DGC-Net, Multimodal Sentiment Analysis, Mixture of Experts, Dynamic Routing, Multimodal Fusion
DGC-Net improves multimodal sentiment analysis by routing text, audio and visual cues through global-guided expert selection, aiming to fix inconsistent signals across modalities in user videos.
Details
Sentiment in opinion vlogs and video reviews comes through words, vocal tone and facial expressions, but these cues are often subtle, noisy and inconsistent across modalities. Existing methods either model local cues within each modality and miss global context, or fuse global and local signals with fixed implicit weighting that cannot adapt when local cues are unreliable. DGC-Net uses auxiliary global context to guide cross-modal dynamic expert routing, and a global-guided compensation fusion that keeps global semantics as the anchor while adding expert-refined local cues as each sample allows. The authors report experiments on five multimodal benchmark datasets; the provided abstract is cut off before specific results.