Scholarlink

<p>Recommendation systems are pivotal in personalizing user experiences by predicting preferences and suggesting relevant items. However, these systems often face challenges due to data sparsity, where limited user-item interactions hinder accurate recommendations. To address this issue, we investigate transfer learning techniques that leverage cross-domain information to enhance recommendation performance. We explore collaborative filtering (CF) techniques, including Singular Value Decomposition (SVD), Matrix Factorization with Biases (MFB), and Deep Matrix Factorization (DPMF), both with and without transfer learning. Our experimental results show that DPMF combined with transfer learning outperforms other methods in terms of prediction accuracy. In the single-domain setting, we achieve an RMSE of 0.8313 and MAE of 0.6458 on the Book domain and an RMSE of 0.8340 and MAE of 0.6965 on the Music domain. In the cross-domain setting, we transfer knowledge from Books to Music, achieving an RMSE of 0.9159 and an MAE of 0.9025. These findings support the potential of our transfer-enhanced DPMF approach in improving recommendation quality across 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

Enhanced Cross-Domain Recommendation System Using Collaborative Filtering and Transfer Learning Methods

  • 期刊论文

  • 发表日期：

    2025年6月24日

查看全文
  • Atul Patel,
  • Vibhor Kant

摘要

Recommendation systems are pivotal in personalizing user experiences by predicting preferences and suggesting relevant items. However, these systems often face challenges due to data sparsity, where limited user-item interactions hinder accurate recommendations. To address this issue, we investigate transfer learning techniques that leverage cross-domain information to enhance recommendation performance. We explore collaborative filtering (CF) techniques, including Singular Value Decomposition (SVD), Matrix Factorization with Biases (MFB), and Deep Matrix Factorization (DPMF), both with and without transfer learning. Our experimental results show that DPMF combined with transfer learning outperforms other methods in terms of prediction accuracy. In the single-domain setting, we achieve an RMSE of 0.8313 and MAE of 0.6458 on the Book domain and an RMSE of 0.8340 and MAE of 0.6965 on the Music domain. In the cross-domain setting, we transfer knowledge from Books to Music, achieving an RMSE of 0.9159 and an MAE of 0.9025. These findings support the potential of our transfer-enhanced DPMF approach in improving recommendation quality across domains.