The problem of predicting human behavior has been a great challenge for several disciplines including computer science. In particular, web user browsing behavior has been studied from the machine learning point of view, a field that has been coined web usage mining (WUM). However, current WUM techniques can be negatively impacted by changes in web site structure and content (e.g. Web 2.0). The key reason behind this issue may be that machine learning algorithms learn the observed behavior according to a particular training set, but do not model the user behavior under different conditions. We propose a simulation model that mimics human interaction with the web by recovering observed navigational steps. This web usage model is inspired by a neurophysiology's stochastic description of decision making and by the information utility of web page content. The proposed model corresponds to a high-dimensional stochastic process based on the leaky competing accumulator (LCA) neural model. We solve high-dimensional issues by considering a mesh-less symbolic interpolation. As a proof-of-concept we test the web user simulation system on an academic web site by recovering most of the observed behavior (73%). Therefore, our approach operationally describes web users that seem to react as observed users confronted by changes in the web site interface.