Abstract Despite exponential growth in digital commerce, conventional customer analytics fail to capture relational dynamics and temporal patterns critical for optimizing targeting precision, cross-selling, and churn prevention. This paper advances consumer behaviour research by framing e-commerce activity as a relational system shaped by economic incentives and temporal decision-making, building on social embeddedness theory and (Recency–Frequency–Monetary) RFM modelling principles. We model transactions as a weighted bipartite network where customers and products form interconnected nodes, addressing gaps in traditional collaborative filtering and demographic segmentation that overlook relational structures. Using 38,995 consumers and 44,802 transactional links across four product categories, we introduce an original edge-weighting methodology fusing linear economic attributes (profit margins, discounts, shipping costs) with nonlinear behavioural components (purchase recency, order prioritisation). Applying graph neural networks (GCN, GAT, GraphSAGE), the framework achieves 94–95% classification accuracy enabling practical applications in personalized marketing automation, dynamic pricing, and recommendation systems with 12–15% gains in targeting precision. Significant variations in network centrality reveal differentiated strategic roles among product segments. By integrating network analytics with social science perspectives, this research illuminates how relational data exposes underlying power relations and behavioural logics in contemporary e-commerce ecosystems, bridging computational innovation with sociological understanding of digital marketplace dynamics.
Batrancea et al. (Wed,) studied this question.