Traditional emotion models face multiple bottlenecks, such as simple emotion types or emotion modeling that is not aligned with real human emotions, resulting in fixed model outputs. To this end, this study proposes an emotion interaction system based on the Plutchik’s wheel of emotions (PWE), setting different stimulation and attenuation modules for different emotion types. Built on the integration of three-modal features, this framework performs modality concatenation by low-rank multimodal fusion (LMF), achieving real-time recognition of seven types of emotions. By PWE mapping and continuous stimulation of moods at the input, the continuous update on emotions is realized. The emotional feedback is sent to Transformer layer of DeepSeek for dynamic adjustment of internal emotional parameters and external hyperparameters, enabling personalized emotional dialog. A series of experiments is designed and conducted to verify the feasibility of this method. The results show that the accuracy of emotion mapping and satisfaction with DeepSeek outputs reach 78% and 84% respectively.