Dynamic Network Analysis of International Oil Trade: A Graph Autoencoder Approach to Density Peaks Clustering

ABSTRACT

Traditional methods for analyzing international crude oil trade (ICOT), such as economic and international trade theory, gravity model, and complex network theory, often overlook the network's topological structure information and lack the ability to adaptively learn node relations. To address the limitations of traditional analytical approaches, we propose a graph‐autoencoder‐based clustering algorithm (GAE‐CA) for analyzing ICOT trend. The GAE‐CA utilizes a graph neural network (GNN) as an encoder to obtain embeddings for nodes, while an inner‐product decoder is employed to reconstruct the original graph. Then, the density peaks clustering algorithm is employed to conduct cluster analysis on the embedding representation of nodes. Finally, by analyzing the ICOT data from 2003 to 2022 using the GAE‐CA model, we observe that during the period 2003–2007, the network forms one cluster. After the economic crisis in 2008, crude oil trading countries form two clusters, and during the period 2018–2022, all countries reintegrate into one cluster.