Abstract: Aiming at the problem that humanoid robots have difficulty perceiving users' emotions in real time and accurately and generating empathetic response dialogue, a humanoid robot head empathetic response dialogue model based on large language model (LLM) is constructed. The model focuses on the emotion recognition of the head system of humanoid robots and the response dialogue that integrates emotional information. On the one hand, implicit and explicit emotion enhancement techniques are utilized to obtain users' deep emotional information, addressing the issue of the lack of hidden emotion mining methods for emotional communication between the humanoid robot's head system and users. On the other hand, the emotional probability of the humanoid robot's head empathetic response is obtained by using the emotional transfer matrix, and the emotional information is integrated into the LLM to transform it into an empathetic response dialogue, solving the problem of humanoid robots' mechanical and formulaic responses during human-computer interaction. The experimental results show that the model can effectively enhance the emotional resonance of human-computer interaction and help achieve more natural and humanized human-computer interaction.