Disentanglement Translation Network for multimodal sentiment analysis

Ying Zeng, Wenjun Yan, Sijie Mai, Haifeng Hu

Abstract: Highlights•Proposing Disentanglement Translation Network for multimodal sentiment analysis.•Devising a two-step translation to achieve more unified cross-modal distribution.•Proposing Slack Reconstruction to impose a more relaxed but effective constraint.•Leveraging multiple samples and considering sample difficulties for reconstruction.•Outperforming baseline methods on two widely-used datasets.