Scholarlink

Collaborative Filtering (CF) is a widely utilized method in recommendation systems. However, a major challenge faced by CF is data sparsity, which can significantly impact the performance and effectiveness of CF-based recommendation systems. To address this problem, we introduce a method named Structured Neighbor and Uniform Noise-Enhanced Learning(SNUNEL). This method enhances CF by incorporating the structured neighbors into the contrastive learning framework. Specifically, SNUNEL utilizes an objective based on structural contrast that considers users or items alongside their structural neighbors as positive pairs, thereby enriching the learning process. Furthermore, SNUNEL introduces uniform noise in the embedding process rather than relying on graph augmentations, which helps to adjust the uniformity of learned representations and alleviates the influence of popularity bias. Diverse benchmark datasets have been utilized to demonstrate the effectiveness of SNUNEL in addressing data sparsity and improving recommendation performance. Compared to competitive CF baselines, SNUNEL achieves significant performance gains, notably with 9.6% and 18.9% performance gains over the MovieLens-1M and Amazon-book datasets, respectively.

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

SNUNEL: Neighborhood-Enriched Contrastive Learning with Uniform Noise to Improved Collaborative Filtering

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xijia Lin,
  • Fei Ye,
  • Jianguo Li,
  • Guohua Chen,
  • Zesong Tan,
  • Jiemin Chen

摘要

Collaborative Filtering (CF) is a widely utilized method in recommendation systems. However, a major challenge faced by CF is data sparsity, which can significantly impact the performance and effectiveness of CF-based recommendation systems. To address this problem, we introduce a method named Structured Neighbor and Uniform Noise-Enhanced Learning(SNUNEL). This method enhances CF by incorporating the structured neighbors into the contrastive learning framework. Specifically, SNUNEL utilizes an objective based on structural contrast that considers users or items alongside their structural neighbors as positive pairs, thereby enriching the learning process. Furthermore, SNUNEL introduces uniform noise in the embedding process rather than relying on graph augmentations, which helps to adjust the uniformity of learned representations and alleviates the influence of popularity bias. Diverse benchmark datasets have been utilized to demonstrate the effectiveness of SNUNEL in addressing data sparsity and improving recommendation performance. Compared to competitive CF baselines, SNUNEL achieves significant performance gains, notably with 9.6% and 18.9% performance gains over the MovieLens-1M and Amazon-book datasets, respectively.