A variational autoencoder that reconstructs each modality from a fused latent representation improves multimodal sentiment classification by 1.5 to 3.1 average F1 points on MOSI, MOSEI, and IEMOCAP.
Kingma and Max Welling
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Variational Fusion for Multimodal Sentiment Analysis
A variational autoencoder that reconstructs each modality from a fused latent representation improves multimodal sentiment classification by 1.5 to 3.1 average F1 points on MOSI, MOSEI, and IEMOCAP.