Scholarlink

<p>Next Point of Interest (POI) recommendation is crucial for many location-based service applications, fundamentally aiming to model user travel preferences to predict their next visit. Unfortunately, most current studies tend to adopt fixed preference dimensions for handling factors like POI, category and geographic location, thereby overlooking the variability in preference importance. In contrast, we prior learn and balance multi-dimensional user preferences. Specifically, we design a preference prior strategy that not only learns user preferences in advance but also supports the flexible expansion of new dimensions. Furthermore, we introduce an adaptive balancing algorithm that effectively adjusts the weights of different preference factors in users’ travel decisions. To the best of our knowledge, this is the first study to explore preference prior learning in Next POI recommendations. Extensive experiments demonstrate that our model outperforms state-of-the-art models, with Accuracy@1 increasing by 2.21%, 6.33%, and 12.62% on the NYC, TKY, and SIN datasets, respectively. Furthermore, the results of the ablation study suggest that our method can scale effectively to datasets with richer features.</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

User multi-dimensional prior preferences adaptive balancing based next POI recommendation

  • 期刊论文

  • 发表日期：

    2025年8月21日

查看全文
  • Chenghua Duan,
  • Wei Zhou,
  • Yuhang He,
  • Junhao Wen

摘要

Next Point of Interest (POI) recommendation is crucial for many location-based service applications, fundamentally aiming to model user travel preferences to predict their next visit. Unfortunately, most current studies tend to adopt fixed preference dimensions for handling factors like POI, category and geographic location, thereby overlooking the variability in preference importance. In contrast, we prior learn and balance multi-dimensional user preferences. Specifically, we design a preference prior strategy that not only learns user preferences in advance but also supports the flexible expansion of new dimensions. Furthermore, we introduce an adaptive balancing algorithm that effectively adjusts the weights of different preference factors in users’ travel decisions. To the best of our knowledge, this is the first study to explore preference prior learning in Next POI recommendations. Extensive experiments demonstrate that our model outperforms state-of-the-art models, with Accuracy@1 increasing by 2.21%, 6.33%, and 12.62% on the NYC, TKY, and SIN datasets, respectively. Furthermore, the results of the ablation study suggest that our method can scale effectively to datasets with richer features.