Personalized Marketing Content Generation System Combined with Diffusion Model

Abstract

Current automated marketing content generation systems typically release content based on predefined rules or lightweight models, resulting in limited variation, creativity, and personalization for advanced engineering products, especially those involving electromagnetic shielding materials, wearable antennas, propagation-related devices, or other electromagnetic application scenarios. These systems often fail to account for product-specific visual characteristics, brand distinctions, electromagnetic application contexts, and diverse user preferences. To address this, the proposed system extracts interest and semantic features through multimodal user profiling. Next, it introduces an enhanced Stable Diffusion model to incorporate user semantic vectors during the denoising and generation process for personalized content synthesis. It implements a prompt control module to automatically vary marketing themes, uses the CLIP feature space to control style diversity, and filters high-quality content through a proposed multidimensional quality evaluation network. As a result, the system generates marketing content with improved creativity and personalization for engineering products with advanced electromagnetic application backgrounds. The results indicate that the proposed model outperforms all benchmarks, with an average semantic consistency score of 0.842, a creative diversity score consistently above 0.85, and an emotional fit score of 0.91 for middle- and high-income groups. The overall brand tone match score reaches 0.905, indicating reduced homogeneity and enhanced creative customization.