the Multi-Granularity Causal Dynamic Router

Dynamic Interaction-Aware and Causality-Disentangled Framework for Multimodal Sentiment Analysis

arXiv:2605.30994v1 Announce Type: new Abstract: Although Multimodal Sentiment Analysis (MSA) effectively leverages rich information from language, visual, and acoustic modalities, existing methods still face two core challenges: 1) static conflict suppression mechanisms fail to adapt to dynamic variations across samples, and 2) the inherent sentimental bias within the language modality, which can misguide learning from other modalities, remains entangled. To this end, we propose a Dynamic...