Transactional Database Analysis by Discovering Pairwise Interactions Strengths

摘要

We present a study that examines the complex interactions that could exist in transactional databases, among the items that constitute the products or purchases present in a market basket of consumers. Using simulated and real databases, we show that it is possible to reconstruct certain characteristics of aggregated purchasing behavior, with only second order interactions using the inverse Ising problem. While Ising's model is well known, the applications in this context are novel and have the potential to reveal useful information to the retail manager.