A personalized recommendation system for e-commerce visual design based on integrated user behavior and image data

With the increasing richness of visual content on e-commerce platforms, visual design has become a key factor influencing user decision-making. However, traditional recommendation systems often overlook users' individualized preferences for visual styles, resulting in limitations in recommendation relevance and appeal. This paper proposes a personalized visual recommendation model that integrates user behavior data with product image features. By employing a Transformer to model user behavior sequences and using CNNs to extract visual features from product images, an attention mechanism is designed to facilitate multimodal feature interaction. The model is evaluated on the Amazon Fashion dataset, and experimental results show that it outperforms traditional baseline models in metrics such as Precision@10, Recall@10, and NDCG@10. Moreover, the recommended results align more closely with users' visual preferences. This study validates the effectiveness of combining behavioral data with visual features and offers an interpretable and engaging recommendation system design path for e-commerce platforms.