M.S. Applied Data Science - Capstone Chronicles 2025

2

Abstract This study investigated consumer cross-shopping behaviors across San Diego County businesses by leveraging network graph modeling to identify latent brand partnerships and structural patterns in the local market. Aggregated consumer spend data from SafeGraph was aggregated to places of interest (POI) and represented as nodes in downstream graph models. Clustering and community detection methodologies were applied to the graphs to enhance interpretability. Through our analysis, we found that businesses could be ranked based on the strength of cross shopping relationships derived from edge weights. This ranking supported more effective prioritization of cross-selling efforts and revealed opportunities for strategic partnerships between business locations. We found that geographically filtered graphs better captured the spatial and consumer cross-shopping patterns between physical locations when compared to the full graph containing all nodes; community-detection outcomes were also improved. Map-based visualizations were produced by our model; by translating graph structures into practical geographic representations, the analysis became more accessible to non-technical stakeholders and capable of supporting businesses partnership strategies and retail placement. Keywords: latent-class analysis, digraph, graph neural network, consumer behavior, network analysis

210