In very close future, human interests and behavior will play a central role in design of mobile networks. Thus, simulation of user behaviors is imperative for the design and evaluation of future mobile networks. However, multi-dimensionality of human interests makes it difficult for us to provide realistic simulations. In this paper, we address challenges in this regard and propose a multivariate technique based on Gaussian mixture models as the first step for simulation of mobile users' interests considering website visitations. We also introduce an evaluation technique for measuring the quality of high-dimensional simulation output. Using our collected dataset including billions of WLAN netflow records for 100 web domains at 79 buildings, we show that our technique not only keeps the characteristics of hidden behavioral groups, but also is flexible enough to approximate the real dataset with given required accuracy.