Bayesian User Behavior Models

Facilitating a satisfying user experience requires a detailed understanding of user behavior and intentions. The key is to leverage observations of activities, usually the clicks performed on Web pages. A common approach is to transform user sessions into Markov chains and analyze them using mixture models. However, model selection and interpretability of the results are often limiting factors. As a remedy, we present a Bayesian nonparametric approach to group user sessions and devise behavioral patterns. Empirical results on an electronic text book show that our approach reliably identifies underlying behavioral patterns and proves more robust than baseline competitors.

weitere Autoren: Jan Reubold, Technische Universität Dresden; Ahcène Boubekki, Wirtschaftsinformatik insb. Machine Learning, Lepuhana; Thorsten Strufe, Technische Universität Dresden;