Emotion-Regularized Conditional Variational Autoencoder for Emotional Response Generation

This article presents an emotion-regularized conditional variational autoencoder (Emo-CVAE) model for generating emotional conversation responses. In conventional CVAE-based response generation, emotion labels are simply used as additional conditions in prior, posterior and decoder networks. Considering that styles naturally entangled with semantic contents the language space, Emo-CVAE utilizes to regularize CVAE latent space by introducing extra prediction network. training stage, estimated variables required predict token sequences of input responses simultaneously. Experimental results show our can learn a more informative structured than output better content performance baseline sequence-to-sequence (Seq2Seq) models.