Scholarlink

<p>Recent research in recommendation systems has increasingly focused on understanding user preferences using collaborative ranking from implicit feedback like clicks and purchases. While previous studies using pairwise ranking techniques assuming relative scores have been successful, the independence assumption in these methods may not always reflect reality. Therefore, current methods have evolved to model user preferences for different items from a setwise viewpoint, thereby relaxing the independence assumption of pairwise preference techniques and yielding certain outcomes. However, these methods typically only consider individual users’ preferences for different items, neglecting the collaborative effects between users. Motivated by these insights and further considering user-side modeling, we propose an improved <b>S</b>etwise collaborative <b>R</b>anking method (SGSR) that aggregates information from users’ <b>S</b>ocial neighbors and a <b>G</b>roup of potential relevant neighbors. Specifically, we integrate information from users’ social networks, thereby enriching our understanding of their social neighbors. Concurrently, we leverage users’ historical interaction data to identify potential relevant neighbors who lack explicit social connections with the user. To assess the efficacy and efficiency of our proposed model, we conduct comprehensive experiments using multiple benchmark datasets. The experimental outcomes demonstrate that our method outperforms existing state-of-the-art techniques.</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

An improved setwise collaborative ranking method leveraging neighborhood information

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Jianyi Wu,
  • Bilian Chen,
  • Langcai Cao

摘要

Recent research in recommendation systems has increasingly focused on understanding user preferences using collaborative ranking from implicit feedback like clicks and purchases. While previous studies using pairwise ranking techniques assuming relative scores have been successful, the independence assumption in these methods may not always reflect reality. Therefore, current methods have evolved to model user preferences for different items from a setwise viewpoint, thereby relaxing the independence assumption of pairwise preference techniques and yielding certain outcomes. However, these methods typically only consider individual users’ preferences for different items, neglecting the collaborative effects between users. Motivated by these insights and further considering user-side modeling, we propose an improved Setwise collaborative Ranking method (SGSR) that aggregates information from users’ Social neighbors and a Group of potential relevant neighbors. Specifically, we integrate information from users’ social networks, thereby enriching our understanding of their social neighbors. Concurrently, we leverage users’ historical interaction data to identify potential relevant neighbors who lack explicit social connections with the user. To assess the efficacy and efficiency of our proposed model, we conduct comprehensive experiments using multiple benchmark datasets. The experimental outcomes demonstrate that our method outperforms existing state-of-the-art techniques.