Scholarlink

In the dynamic realm of e-commerce, delivering efficient and accurate product recommendations is essential for enhancing user experience. This paper introduces a multimodal fashion product recommendation system that leverages both text and image inputs for improved relevance and personalization. Visual features are extracted using the VGG19 model, while textual features are processed using LSTM networks. To manage high-dimensional data, dimensionality reduction techniques PCA is employed, followed by similarity computation using KNN and Annoy. The proposed system achieves a notable accuracy of 92.2% with PCA and Annoy, significantly outperforming the benchmark accuracy of 89.3%. Furthermore, the system is almost eleven times faster compared to KNN, enabling real-time performance in large-scale e-Commerce environments compared to other recommendation techniques. By combining multimodal inputs and advanced algorithms, this work establishes a scalable, dynamic approach to intelligent product recommendations.

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

Deep Learning Multimodal Fashion Product Recommendation via Transfer Learning of VGG19

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Suraj Nishad,
  • Kamlesh Dutta,
  • Aman Kumar,
  • Ramakrishna Miryala

摘要

In the dynamic realm of e-commerce, delivering efficient and accurate product recommendations is essential for enhancing user experience. This paper introduces a multimodal fashion product recommendation system that leverages both text and image inputs for improved relevance and personalization. Visual features are extracted using the VGG19 model, while textual features are processed using LSTM networks. To manage high-dimensional data, dimensionality reduction techniques PCA is employed, followed by similarity computation using KNN and Annoy. The proposed system achieves a notable accuracy of 92.2% with PCA and Annoy, significantly outperforming the benchmark accuracy of 89.3%. Furthermore, the system is almost eleven times faster compared to KNN, enabling real-time performance in large-scale e-Commerce environments compared to other recommendation techniques. By combining multimodal inputs and advanced algorithms, this work establishes a scalable, dynamic approach to intelligent product recommendations.