Extending market basket analysis with graph mining techniques: A real case

•We model two retail chains: a wholesale supermarket and a retail supermarket.•We present a novel approach to process million of records from transactional data.•We introduce a novel way to generate and to filter product networks.•We generate overlapped communities of related products.•Results give to the analyst a manageable set of products that can be characterized.

A common problem for many companies, like retail stores, it is to find sets of products that are sold together. The only source of information available is the history of sales transactional data. Common techniques of market basket analysis fail when processing huge amounts of scattered data, finding meaningless relationships. We developed a novel approach for market basket analysis based on graph mining techniques, able to process millions of scattered transactions. We demonstrate the effectiveness of our approach in a wholesale supermarket chain and a retail supermarket chain, processing around 238,000,000 and 128,000,000 transactions respectively compared to classical approach.