An Empathetic Dialogue Generation Model Based on External Knowledge Fusion

Aiming at the problem that traditional dialogue systems are difficult to capture the hidden emotional factors in dialogue history, this paper proposes an external knowledge-enhanced empathetic dialogue generation (KEDG) model based on external knowledge fusion. The model explicitly understands and expresses the corresponding emotion in the dialogue history, thus realizing the empathy ability of the model. First, a context map of emotional context is constructed by enriching the dialogue history by fusing with external knowledge. Then, the model is broken down into three main parts and described separately, including emotion context encoder, emotion prediction and emotion-dependent decoder. Based on the emotional features extracted from the dialogue model, user emotions can be modeled to generate empathic responses. Finally, extensive experimental results on benchmark datasets verify the effectiveness of the proposed model.