In traditional virtual health assistants, there is a lack of emotional understanding and lack of personalization. By using Seq2Seq-GAN (Sequence to Sequence Generative Adversarial Networks) technology, this article attempts to solve these problems. These technologies were used to realize the functions of emotion perception and understanding and personalized emotional interaction. First, an emotion intelligent interaction model was created, including two core modules: emotion perception and understanding and personalized emotion interaction. The emotion perception module uses the Seq2Seq model to serialize the emotions input by the user. Using GAN generator network, it can understand and describe emotions more deeply. The personalized emotional interaction module analyzes user historical data and behavior patterns, trains the generator network to generate personalized emotional responses, and improves user participation. Then, the model was evaluated, where the main evaluation indicators include emotion recognition accuracy and personalized interaction success rate. Experimental results showed that the accuracy rate of this model was 90% and the success rate was 85%, which verified the effectiveness and feasibility of the model. The emotion intelligent interaction model proposed in this article can solve the emotional adjustment and interaction problems of traditional virtual health assistants, and provide new ideas and methods for creating more intelligent and emotional virtual health assistants.