Objective: This paper studies the design ideas and concepts of the emotional interaction of the ordering robot, so as to design the human-based intelligent emotional interaction method of the ordering robot. Methods: Firstly, BLSTM network was used to construct the English semantic parameter coding network, then put forward thrust type attention model, this model can extract data by means of weighting, after that, the constrained SeqGAN network architecture is designed to complete the decoding, so as to adjust the parameters of the generating device and narrow the gap between the generated ordering language and human English emotional interaction response. Results: Compared with Du-Model method and HRED-Model method, the BLSTM-SeqGAN method has smaller confusion index and higher accuracy, and becomes more stable as the number of iterations increases. Conclusion: This method can obtain more natural, real and friendly emotional interaction response.