Personalized design aesthetic preference modeling: a variational autoencoder and meta-learning approach for...

This research presents a comprehensive framework for personalized design aesthetic preference modeling that integrates variational autoencoders (VAE) with meta-learning approaches to address the limitations of traditional design evaluation methodologies. The proposed system combines multi-modal feature extraction from visual, textual, and behavioral data through attention-based fusion mechanisms, enabling robust capture of individual aesthetic preferences while maintaining generalization across diverse design domains. The VAE-based probabilistic modeling framework captures uncertainty in aesthetic judgments through learned latent representations, while the meta-learning component enables rapid adaptation to individual user preferences with minimal training data. Experimental evaluation across six design datasets demonstrates superior performance, achieving 84.7% accuracy in aesthetic preference prediction and 70% reduction in adaptation requirements compared to conventional transfer learning approaches. The framework successfully addresses the dual challenges of personalization and cross-domain generalization, providing practical foundations for intelligent design tools and automated content curation systems.