Disentangled Variational Autoencoder for Emotion Recognition in Conversations

In Emotion Recognition in Conversations (ERC), the emotions of target utterances are closely dependent on their context. Therefore, existing works train model to generate response utterance, which aims recognise leveraging contextual information. However, adjacent generation ignores long-range dependencies and provides limited affective information many cases. addition, most ERC models learn a unified distributed representation for each lacks interpretability robustness. To address these issues, we propose VAD-disentangled Variational AutoEncoder (VAD-VAE), first introduces utterance reconstruction task based Autoencoder, then disentangles three affect representations Valence-Arousal-Dominance (VAD) from latent space. We also enhance disentangled by introducing VAD supervision signals sentiment lexicon minimising mutual between distributions. Experiments show that VAD-VAE outperforms state-of-the-art two datasets. Further analysis proves effectiveness proposed module quality representations. The code is available at https://github.com/SteveKGYang/VAD-VAE.