Emotional Response Generation using Conditional Variational Autoencoder

Neural response generation is to generate human-like response given human utterance by using a deep learning. In the previous studies, expressing emotion in response generation improve user performance, user engagement, and user satisfaction. Also, the conversational agents can communicate with users at the human level. However, the previous emotional response generation model cannot interpret why the model generates such response with emotions. We propose an interpretable emotional response generation model which generates emotional responses by using a latent space. The extraction part is to extract the emotion of input utterance as a vector form by using the Bidirectional GRU based classification model. The generation part is to generate an emotional response to the input utterance by exploiting emotion vector and latent space. All of these parts are jointly optimized at the training process. We will evaluate our model on the emotion-labeled dialogue dataset: DailyDialog.