TDGN: A text-guided dual-gated network for multimodal sentiment analysis
Fig 5
t-SNE visualization of multimodal representations before and after applying the TGA module.
Left: the features are misaligned and distributed irregularly across modalities. Right: after TGA, the features exhibit improved alignment and compact clustering across text, video, and audio modalities.