Article: Pattern recognition of digital trade flows in manufacturing clusters using graph neural networks Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.90 pp.93 - 116 Abstract: The digital trade flow patterns within manufacturing clusters are crucial for analysing supply chain resilience. However, existing graph neural network methods often treat enterprises within the cluster as homogeneous nodes, ignoring structural differences such as product specialisation levels and technical density, resulting in significant deviations in trade intensity predictions. This paper proposes the domain-aware attention graph network, which explicitly captures the heterogeneous trade relationships among enterprises by embedding learnable node domain feature modulation functions. Experiments show that domain-aware attention graph network achieves a determination coefficient of 0.842 in the trade intensity regression task, an improvement of approximately 0.049 compared to graph attention network; in the trade change direction classification task, its area under the curve reaches 0.891, an improvement of approximately 0.044 compared to graph attention network, indicating that the model can more accurately identify trade flow and intensity changes. Inderscience Publishers - linking academia, business and industry through research

Title: Pattern recognition of digital trade flows in manufacturing clusters using graph neural networks

Authors: Junqing Sheng; Yanhong Wang; Wenfa Zhang

Addresses: School of Economics and Management, Shandong Huayu University of Technology, Dezhou, 253034, China ' School of Economics and Management, Shandong Huayu University of Technology, Dezhou, 253034, China ' School of Economics and Management, Shandong Huayu University of Technology, Dezhou, 253034, China

Abstract: The digital trade flow patterns within manufacturing clusters are crucial for analysing supply chain resilience. However, existing graph neural network methods often treat enterprises within the cluster as homogeneous nodes, ignoring structural differences such as product specialisation levels and technical density, resulting in significant deviations in trade intensity predictions. This paper proposes the domain-aware attention graph network, which explicitly captures the heterogeneous trade relationships among enterprises by embedding learnable node domain feature modulation functions. Experiments show that domain-aware attention graph network achieves a determination coefficient of 0.842 in the trade intensity regression task, an improvement of approximately 0.049 compared to graph attention network; in the trade change direction classification task, its area under the curve reaches 0.891, an improvement of approximately 0.044 compared to graph attention network, indicating that the model can more accurately identify trade flow and intensity changes.

Keywords: manufacturing cluster; digital trade flow; hierarchical attention mechanism; gravitational model.

DOI: 10.1504/IJICT.2026.155734

International Journal of Information and Communication Technology, 2026 Vol.27 No.90, pp.93 - 116

Received: 02 May 2026
Accepted: 12 Jun 2026
Published online: 11 Aug 2026 *

Title: Pattern recognition of digital trade flows in manufacturing clusters using graph neural networks

Keep up-to-date