An explainable deep learning framework for balancing cultural authenticity and user sentiment in product design

To advance the organic integration of intangible cultural heritage with innovative products, this study proposes an interpretable deep learning framework, using household service robots as an application scenario to systematically address the balance between cultural authenticity and user affective needs. The research first quantifies the weights of Yao cultural elements and user requirements through the Hesitant Fuzzy Analytic Hierarchy Process (HF-AHP) and Hesitant Fuzzy Quality Function Deployment (HF-QFD). Subsequently, a hybrid CPO-TCN-Transformer model is constructed, which integrates the multi-scale feature extraction capability of Temporal Convolutional Networks (TCN) with the global dependency modeling of the Transformer architecture. Furthermore, the Crowned Porcupine Optimizer (CPO) is incorporated to automatically tune key hyperparameters, effectively establishing complex nonlinear mappings between cultural features and design parameters. Experimental results demonstrate that the proposed method not only accurately predicts user affective preferences but also provides transparent explanations for design decisions via SHAP analysis. This approach achieves an organic fusion of traditional culture and modern technology, providing methodological support for innovative and clever product design, as well as the creative transformation of cultural heritage.