Unsupervised Customer Segmentation with Knowledge Graph Embeddings

Abstract

We propose an unsupervised customer segmentation method from behavioral data. We model sequences of beer consumption from a publicly available dataset of 2.9M reviews of more than 110,000 brands over 12 years as a knowledge graph, learn their representations with knowledge graph embedding models, and apply off-the-shelf cluster analysis. Experiments and clusters interpretation show that we learn meaningful clusters of beer customers, without relying on expensive consumer surveys or time-consuming data annotation campaigns.