This study proposes an NLP-based prediction model to enhance the Kansei quality evaluation of appearance in competitive mouse design. The objective is to align the mouse’s appearance with the emotional needs of the target users while resolving issues related to subjective design and real-time performance. The model incorporates multidimensional scale analysis and K-Means to acquire representative samples and employs LDA and PCA for extracting evaluation indexes from online reviews. By utilizing an SSA-optimized RBF neural network, the model develops an evaluation prediction model where sample modeling feature coding acts as the input layer and user Kansei evaluation scores serve as the output layer during training. The prediction results are later compared against those obtained from PSO and GA-optimized RBF models. Through this research, it is demonstrated that the proposed prediction model effectively forecasts the Kansei quality evaluation of competitive mouse appearance, thus contributing to a more efficient and scientifically grounded approach to emotional design.