Fig. 1

Methodology pipeline: A summary of the pipeline to extract and analyze the city consumption profiles. The transformation and analysis pipeline moves from left to right; from the raw purchase data, we build a bipartite graph (customer-to-merchant category). From the bipartite graph, we derive a customer consumption profile, which summarizes the single spending behavior of clients. The customer consumption profiles are nodes of a similarity graph where edges connect very similar consumption profiles. As the last step, a community detection algorithm, run on the similarity graph, identifies the representative consumption profiles used to obtain the city consumption profile