Personalized Fashion Recommendation Fusing Multi-Behavior and Multi-Modal Features

Aiming at the problems of data sparsity, uneven behavior weight allocation, and insufficient timeliness modeling existing in traditional recommendation systems in the scenario of personalized fashion recommendation, this paper proposes a personalized recommendation method that integrates multi-behavior weights and multi-modal features. A dynamic weighted collaborative filtering algorithm is designed, which comprehensively considers the multi-dimensional behaviors of users, and introduces a time attenuation factor to construct a time-sensitive user-item scoring matrix, so as to more accurately depict the dynamic changes of user interests. A multi-modal deep fusion framework is built: ResNet-50 is used to extract commodity image features, and the pre-trained BERT model is combined to extract text features; meanwhile, the multi-head self-attention mechanism is adopted to realize semantic-level interaction and adaptive fusion of cross-modal features, thereby enhancing the expressive ability of commodity representation. Then, user preference score prediction is carried out based on the deep predictive network to generate a personalized recommendation list. Experimental results on real e-commerce datasets show that the method in this paper achieves 0.703 and 0.491 on HR@5 and NDCG@5, respectively, which is significantly superior to other baseline models. Ablation experiments further verify the effectiveness of each module including time attenuation, multi-behavior weights and multi-modal features. This study provides a more accurate, dynamic and transparently interpretable personalized recommendation solution for e-commerce platforms, and has certain theoretical value and practical significance.