The Objective Way to Detect the Path to Purchase by Clustering Shoppers’ Trajectories

Abstract

Abstract The trajectories of shopping carts and baskets in a supermarket are an information-rich feature that can help in understanding the retail environment, gives idea about the interactions between objects and the ongoing events. This paper is interested in understanding, improving and personalising shopping experience by clustering the trajectories acquired by shopping carts and baskets in a real retail environment (a real supermarket during the business hours). In order to reach this, sCREEN (Consumer REtail ExperieNce) dataset has been previously built and the tracking system is built around Ultra-WideBand (UWB) technology. The approach proposed in this paper enables retailers to analyse the impact of different store layouts (maps) and shelf layouts (planograms) on issues such as ease of selection, affairs and the overall shopping experience. The first assumption is that some common sub-trajectories can be missed when seeing clustering trajectories as a whole. After the trajectory partitions into a set of line segments, similar line segments are grouped together to finally form a cluster. Then, a novel macro-clustering is developed, able to improve the robustness of the method and take into account the fact that trajectories are ordered collections of points. Finally, the performance of the proposed method is tested on a real-world dataset.