Scholarlink

<p>With the rapid expansion of e-commerce platforms, personalized clothing recommendation has emerged as a critical technology for improving user satisfaction and platform revenue. This study addresses the challenges of effectively integrating multimodal item information and capturing temporally evolving user preferences in clothing recommendation. We propose a Multimodal Fusion based Dynamic Preference Learning Framework (MF-DPL) that employs iterative cross-modal attention mechanisms and gated preference decomposition to construct a unified recommendation model. This approach successfully optimizes ranking accuracy in sparse implicit feedback scenarios, achieving substantial performance gains. Experimental results on the Amazon Clothing and Polyvore Outfits datasets show that MF-DPL improves Recall@10 by 2.7%–8.0% and NDCG@10 by 2.0%–7.0% compared to recent state-of-the-art fashion recommendation baselines, while outperforming classical collaborative filtering methods by even larger margins. These findings provide practical guidance for multimodal personalized recommendation in fashion e-commerce and indicate new directions for dynamic preference modeling in trend sensitive domains.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fusion of multimodal data for clothing personalized recommendation with dynamic preference learning framework

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月24日

查看全文
  • Ying Yuan

摘要

With the rapid expansion of e-commerce platforms, personalized clothing recommendation has emerged as a critical technology for improving user satisfaction and platform revenue. This study addresses the challenges of effectively integrating multimodal item information and capturing temporally evolving user preferences in clothing recommendation. We propose a Multimodal Fusion based Dynamic Preference Learning Framework (MF-DPL) that employs iterative cross-modal attention mechanisms and gated preference decomposition to construct a unified recommendation model. This approach successfully optimizes ranking accuracy in sparse implicit feedback scenarios, achieving substantial performance gains. Experimental results on the Amazon Clothing and Polyvore Outfits datasets show that MF-DPL improves Recall@10 by 2.7%–8.0% and NDCG@10 by 2.0%–7.0% compared to recent state-of-the-art fashion recommendation baselines, while outperforming classical collaborative filtering methods by even larger margins. These findings provide practical guidance for multimodal personalized recommendation in fashion e-commerce and indicate new directions for dynamic preference modeling in trend sensitive domains.