Original scientific paper

https://doi.org/10.17559/TV-20250524002692

Dual-Layer Meta-Learning Framework for Adaptive Multi-Task Scheduling of Digital Agents in Hotel Operations

Fuchun Xing ; HNU-ASU International College, Hainan University, No. 58 Renmin Avenue, Meilan District, Haikou City, Hainan Province *
Yang Cao ; HNU-ASU International College, Hainan University, No. 58 Renmin Avenue, Meilan District, Haikou City, Hainan Province
Feng Li ; China Telecom IoT Technology Co., Ltd. No. 1835, Pudong South Road, Pudong New Area, Shanghai 200211

* Corresponding author.

Abstract

In the evolving landscape of luxury hospitality, efficient and adaptive task scheduling of digital agents is crucial, particularly during peak operational periods characterized by simultaneous high-priority guest requests. Traditional scheduling systems, typically relying on rigid rules or deep learning models, often struggle with dynamic environments, resulting in prolonged customer wait times and inefficient resource utilization. To overcome these limitations, this paper proposes an innovative Dual-Layer Meta-Learning (DDL) framework. By modeling tasks scheduling as a Markov Decision Process (MDP) and integrating an attention-enhanced sequence-to-sequence (Seq2Seq) neural network, our framework effectively captures task dependencies and dynamically adjusts priorities in real-time. The proposed dual-layer meta-learning structure combines generalized cross-scenario knowledge extraction in the outer loop with rapid fine-tuning capabilities in the inner loop, enabling quick adaptation to new operational contexts with limited data. Experimental validation demonstrates substantial improvements, including a 27% increase in service response speed and a 19% reduction in customer waiting times compared to conventional methods. Ablation studies further highlight the pivotal role of the attention mechanism, emphasizing its effectiveness in accurately prioritizing critical tasks. These results underscore the framework's significant potential for enhancing operational efficiency in dynamic hospitality environments and its broader applicability to similar scheduling challenges across diverse service industries.

Keywords

adaptive service management; attention mechanism; digital agents; dual-layer Meta-Learning (DDL); few-shot learning; hotel operations; Markov decision process (MDP); multi-task scheduling; sequence-to-sequence (Seq2Seq) networks

Hrčak ID:

335075

URI

https://hrcak.srce.hr/335075

Publication date:

30.8.2025.

Visits: 305 *