Scholarlink

<p>Denied Check-in (DCI), which refers to situations where guests with confirmed reservations are unable to check into their booked accommodations, has become a crucial concern for the online travel platforms (OTPs). In this paper, we address this issue with a novel trend-aware DCI prediction network. Our model is designed to capture the manifold causative factors and multi-scale temporal trends of DCI occurrences using an innovative attention mechanism. To tackle the challenge of data sparsity in DCI prediction, we develop a multi-task learning framework that simultaneously trains the model on both DCI prediction and order refusal prediction tasks. Comprehensive experiments on real-world dataset validate the superiority of our method over state-of-the-art baselines. Moreover, our model has been successfully deployed on a popular online travel platform to serve real traffic, leading to notable reduction in the platform’s overall DCI Rate.</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

Multi-faceted, multi-scale, and multi-task trend learning for denied check-in prediction on online travel platforms

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Fanwei Zhu,
  • Zulong Chen,
  • Wanjie Tao,
  • Quan Lu,
  • Li Lv,
  • Hailong Tan,
  • Zui Tao

摘要

Denied Check-in (DCI), which refers to situations where guests with confirmed reservations are unable to check into their booked accommodations, has become a crucial concern for the online travel platforms (OTPs). In this paper, we address this issue with a novel trend-aware DCI prediction network. Our model is designed to capture the manifold causative factors and multi-scale temporal trends of DCI occurrences using an innovative attention mechanism. To tackle the challenge of data sparsity in DCI prediction, we develop a multi-task learning framework that simultaneously trains the model on both DCI prediction and order refusal prediction tasks. Comprehensive experiments on real-world dataset validate the superiority of our method over state-of-the-art baselines. Moreover, our model has been successfully deployed on a popular online travel platform to serve real traffic, leading to notable reduction in the platform’s overall DCI Rate.