Identity Bridging Cluster Website Visits using Model-based Clustering

Abstract

Model-based clustering is becoming increasingly popular with the rise in computational power. Cluster analysis is used in many disciplines, for example biology, geography, image analysis and marketing. In this thesis we developed a model-based unsupervised clustering method to cluster website visits into clusters that represent a unique Internet user. As no ground truth exists we developed two evaluation methods to measure the quality of the clusters, one based on cluster content and size, and the other based on user behavior. The model-based clustering method was compared with a simple deterministic clustering model, the results were very similar. With further development of the model-based clustering we believe that it can generate better clusters of website visits that likely represent a single user.