Clickstream Data Analysis and Web User Profiling via Mixture Hidden Markov Models

Abstract

Using Mixture Hidden Markov Models (MHMMs), the study analyses clickstream data to identify users’ profiles with similar browsing behaviour. MHMMs enable us to analyse categorical sequences, assuming they evolve according to a mixture of latent Markov processes, each related to a different subpopulation. An empirical analysis of clickstream data from a hospitality industry website has been performed. Evidence shows the usefulness of MHMMs in exploring user behaviour and defining ad-hoc marketing strategies. Finally, as MHMMs entail identifying two latent classes, viz., the number of sub-populations and hidden states, the study proposes a model selection criterion based on an integrated completed likelihood approach that accounts for both latent classes.