In the market basket setting, we are given a series of transactions each composed of one or more items and the goal is to find relationships between items, usually sets of items that tend to occur in the same transaction. Association rules, a popular approach for mining such data, are limited in the ability to express complex interactions between items. Our work defines some of these limitations and addresses them by modeling the set of transactions as a network. We develop both a general methodology for analyzing networks of products, and a privacy-preserving protocol such that product network information can be securely shared among stores. In general, our network based view of transactional data is able to infer relationships that are more expressive and expansive than those produced by a typical association rules analysis.
