Abstract

Retail industry has evolved. Nowadays, companies around the world need a better and deeper understanding of their customers. In order to enhance store layout, generate customers groups, offers and personalized recommendations, among others. To accomplish these objectives, it is very important to know which products are related to each other. Classical approaches for clustering products, such as K-means or SOFM, do not work when exist scattered and large amounts of data. Even association rules give results that are difficult to interpret. These facts motivate us to use a novel approach that generates communities of products. One of the main advantages of these communities is that are meaningful and easily interpretable by retail analysts. This approach allows the processing of billions of transaction records within a reasonable time, according to the needs of companies. (C) 2014 The Authors. Published by Elsevier B.V.