Abstract:
Transaction analysis is an important part in studies aiming to understand consumer behaviour. The first step is defining a proper measure of similarity, or more specifically a distance metric, between transactions. Existing distance metrics on transactional data are built on retailer specific information, such as extensive product hierarchies or a large product catalogue. In this paper we propose a new distance metric that is retailer independent by design, allowing cross-retailer and cross-country analysis. The metric comes with a novel method of finding the importance of categories of products, alternating between unsupervised learning techniques and importance calibration. We test our methodology on a real-world dataset and show how we can identify clusters of consumer behaviour.
