A Two-Stage Emotional Dialogue Generation Model Based on DialoGPT

The creation of emotionally rich dialogues has become an increasingly important area in natural language processing (NLP), aiming to improve human-computer interactions. Existing models frequently rely on a single emotional variable, which may produce repetitive and less varied responses. To overcome these challenges, we introduce a two-stage dialogue generation model featuring a novel emotional variable fusion strategy. In the initial stage, a fine-tuned DialoGPT model is employed to produce semantically coherent responses. In the subsequent stage, we enhance emotional diversity through emotional rewriting by integrating primary and auxiliary emotional variables, while preserving semantic accuracy. Experiments conducted on the EmpatheticDialogues dataset show that our model surpasses baseline models regarding fluency, relevance, and emotional engagement.