Adaptive Modality Reliability Diagnosis and Restoration for Robust Multimodal Intent Recognition

Multimodal intent recognition combines linguistic, acoustic, and visual evidence, but individual modalities may be noisy, missing, semantically conflicting, or disproportionately dominant. Existing methods typically infer modality importance implicitly and either reweight or suppress unreliable inpu...