Graph-Centric Intelligent Decision Support System with Application in Inventory Management

The Hong Kong University of Science and Technology (Guangzhou)

Data Science and Analytics Thrust

PhD Thesis Examination

By Ms. Luxuan WANG

ABSTRACT

This thesis addresses the inadequacy of conventional decision support systems in managing the complexity of modern inventory management. These traditional systems, which rely on simplified models and structured data, often result in significant inefficiencies. To overcome these limitations, a novel, graph-centric intelligent decision support system is proposed.

The new framework has three core technical contributions. First, it utilizes Large Language Models (LLMs) to automatically construct a knowledge graph, which unifies heterogeneous data into a semantic whole. Second, it employs a self-supervised representation learning method, to capture complex temporal patterns from unlabeled time-series data. This avoids the ”false-negative” problem common in traditional methods. Finally, an End-to-End Inventory Replenishment Framework based on Temporal Graph Neural Networks is introduced. This integrates the structural and dynamic layers to predict future network states, infer unobserved production functions, and generate proactive, cost-optimized replenishment decisions.

Validated with large-scale, real-world datasets from semiconductor manufacturing and logistics, the integrated system demonstrates state-of-the-art performance, significantly outperforming existing baselines. This research presents a new paradigm for creating adaptive, data-driven decision support systems capable of managing the complexities of modern socio-technical systems.

TEC

Chairperson: Prof Hai-Ning LIANG
Prime Supervisor: Prof Fugee TSUNG
Co-Supervisor: Prof Wenjia WANG
Examiners:
Prof Kaishun WU
Prof Lei ZHU
Prof Ruiting ZUO
Prof Zijun ZHANG