Scholarlink

<p>Next point of interest (POI) recommendation tasks are pivotal in location-based social networks (LBSN) service applications. Although recent studies have demonstrated the effectiveness of graph-based methods in next POI recommendation, they fall short in global spatial-temporal information integration and capturing high-level dependencies within user trajectories to model user preferences while also struggling with sparsity. Thus, global-local graph representation with attention network (GLGRAN) is proposed to surmount these challenges. GLGRAN acquires the general and specific user flavor by integrating a global spatial-temporal graph network module to fuse more abundant spatial and temporal information from a global perspective while employing a local graph attention module to capture high-level dependencies from user visits. Furthermore, the sparsity caused by inactive users is attenuated using a multi-task learning module to predict the next POI and G@<i>P</i> areas. Experimental results demonstrate that GLGRAN outperforms the state-of-the-art algorithms in various datasets and metrics. </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

GLGRAN: next POI recommendation on global-local graph representation with attention network

  • 期刊论文

  • 发表日期：

    2025年5月6日

查看全文
  • Weilong Li,
  • Qiang Zhang,
  • Lei Zhang,
  • Xiao-Yuan Jing

摘要

Next point of interest (POI) recommendation tasks are pivotal in location-based social networks (LBSN) service applications. Although recent studies have demonstrated the effectiveness of graph-based methods in next POI recommendation, they fall short in global spatial-temporal information integration and capturing high-level dependencies within user trajectories to model user preferences while also struggling with sparsity. Thus, global-local graph representation with attention network (GLGRAN) is proposed to surmount these challenges. GLGRAN acquires the general and specific user flavor by integrating a global spatial-temporal graph network module to fuse more abundant spatial and temporal information from a global perspective while employing a local graph attention module to capture high-level dependencies from user visits. Furthermore, the sparsity caused by inactive users is attenuated using a multi-task learning module to predict the next POI and G@P areas. Experimental results demonstrate that GLGRAN outperforms the state-of-the-art algorithms in various datasets and metrics.