Emotional Dialogue Generation with Emotion Embedding

Abstract

As an important research content of artificial intelligence, dialogue system has received extensive attention from industry and academia. Existing dialogue systems mainly focus on solving problems such as content richness and semantic consistency. The research on emotion control has not received much attention. It’s still quite challenging to generate emotional responses. In this paper, we propose a dialogue generation model based on Seq2Seq and add emotion embedding to the decoder. Experiments show that the model can generate appropriate responses both in emotion and content. We also train a Bert_BiLSTM emotion classifier to improve the emotion annotation quality of the CDCG Dataset.