Xian-Xun Zhu

Reliability-Aware Multi-Modal Sentiment Analysis Under Missing and Corrupted Modalities

A trainable reliability-aware evidential fusion framework that estimates not only sentiment predictions but also modality-specific evidence, predictive uncertainty, observable input quality, cross-modal disagreement, and normalized sample-dependent fusion weights is presented.