Latent Session Model for Web User Clustering A case study on modeling users of an online real estate website

We analyze the web access log of Zillow.com – one of the largest real estate website and present a hierarchical mixture model which learns clusters of users and sessions from the combination of web usage and content data. The model is able to exploit the hierarchical structure of the usage data, and learns stereotypical session types and user segments such as high end or low end house buyers. We show that our model produces better clusters both qualitatively and quantitatively comparing to a 2-phase baseline model.